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/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.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
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static inline int ext2_feature_set_ok(struct super_block *sb);
  79 static inline int ext3_feature_set_ok(struct super_block *sb);
  80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  81 static void ext4_destroy_lazyinit_thread(void);
  82 static void ext4_unregister_li_request(struct super_block *sb);
  83 static void ext4_clear_request_list(void);
  84 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 MODULE_ALIAS_FS("ext2");
  95 MODULE_ALIAS("ext2");
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 MODULE_ALIAS_FS("ext3");
 111 MODULE_ALIAS("ext3");
 112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 113 #else
 114 #define IS_EXT3_SB(sb) (0)
 115 #endif
 116
 117 static int ext4_verify_csum_type(struct super_block *sb,
 118                                  struct ext4_super_block *es)
 119 {
 120         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 121                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 122                 return 1;
 123
 124         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 125 }
 126
 127 static __le32 ext4_superblock_csum(struct super_block *sb,
 128                                    struct ext4_super_block *es)
 129 {
 130         struct ext4_sb_info *sbi = EXT4_SB(sb);
 131         int offset = offsetof(struct ext4_super_block, s_checksum);
 132         __u32 csum;
 133
 134         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 135
 136         return cpu_to_le32(csum);
 137 }
 138
 139 int ext4_superblock_csum_verify(struct super_block *sb,
 140                                 struct ext4_super_block *es)
 141 {
 142         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 143                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 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_RO_COMPAT_FEATURE(sb,
 154                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 155                 return;
 156
 157         es->s_checksum = ext4_superblock_csum(sb, es);
 158 }
 159
 160 void *ext4_kvmalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kmalloc(size, flags);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 void *ext4_kvzalloc(size_t size, gfp_t flags)
 171 {
 172         void *ret;
 173
 174         ret = kzalloc(size, flags);
 175         if (!ret)
 176                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 177         return ret;
 178 }
 179
 180 void ext4_kvfree(void *ptr)
 181 {
 182         if (is_vmalloc_addr(ptr))
 183                 vfree(ptr);
 184         else
 185                 kfree(ptr);
 186
 187 }
 188
 189 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 190                                struct ext4_group_desc *bg)
 191 {
 192         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 193                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 194                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 195 }
 196
 197 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 198                                struct ext4_group_desc *bg)
 199 {
 200         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 201                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 202                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 203 }
 204
 205 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 206                               struct ext4_group_desc *bg)
 207 {
 208         return le32_to_cpu(bg->bg_inode_table_lo) |
 209                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 210                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 211 }
 212
 213 __u32 ext4_free_group_clusters(struct super_block *sb,
 214                                struct ext4_group_desc *bg)
 215 {
 216         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 217                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 218                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 219 }
 220
 221 __u32 ext4_free_inodes_count(struct super_block *sb,
 222                               struct ext4_group_desc *bg)
 223 {
 224         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 225                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 226                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 227 }
 228
 229 __u32 ext4_used_dirs_count(struct super_block *sb,
 230                               struct ext4_group_desc *bg)
 231 {
 232         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 233                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 234                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 235 }
 236
 237 __u32 ext4_itable_unused_count(struct super_block *sb,
 238                               struct ext4_group_desc *bg)
 239 {
 240         return le16_to_cpu(bg->bg_itable_unused_lo) |
 241                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 242                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 243 }
 244
 245 void ext4_block_bitmap_set(struct super_block *sb,
 246                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 247 {
 248         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 249         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 250                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 251 }
 252
 253 void ext4_inode_bitmap_set(struct super_block *sb,
 254                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 255 {
 256         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 257         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 258                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 259 }
 260
 261 void ext4_inode_table_set(struct super_block *sb,
 262                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 263 {
 264         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 265         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 266                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 267 }
 268
 269 void ext4_free_group_clusters_set(struct super_block *sb,
 270                                   struct ext4_group_desc *bg, __u32 count)
 271 {
 272         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 273         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 274                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 275 }
 276
 277 void ext4_free_inodes_set(struct super_block *sb,
 278                           struct ext4_group_desc *bg, __u32 count)
 279 {
 280         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 281         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 282                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 283 }
 284
 285 void ext4_used_dirs_set(struct super_block *sb,
 286                           struct ext4_group_desc *bg, __u32 count)
 287 {
 288         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 289         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 290                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 291 }
 292
 293 void ext4_itable_unused_set(struct super_block *sb,
 294                           struct ext4_group_desc *bg, __u32 count)
 295 {
 296         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 297         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 298                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 299 }
 300
 301
 302 static void __save_error_info(struct super_block *sb, const char *func,
 303                             unsigned int line)
 304 {
 305         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 306
 307         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 308         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 309         es->s_last_error_time = cpu_to_le32(get_seconds());
 310         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 311         es->s_last_error_line = cpu_to_le32(line);
 312         if (!es->s_first_error_time) {
 313                 es->s_first_error_time = es->s_last_error_time;
 314                 strncpy(es->s_first_error_func, func,
 315                         sizeof(es->s_first_error_func));
 316                 es->s_first_error_line = cpu_to_le32(line);
 317                 es->s_first_error_ino = es->s_last_error_ino;
 318                 es->s_first_error_block = es->s_last_error_block;
 319         }
 320         /*
 321          * Start the daily error reporting function if it hasn't been
 322          * started already
 323          */
 324         if (!es->s_error_count)
 325                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 326         le32_add_cpu(&es->s_error_count, 1);
 327 }
 328
 329 static void save_error_info(struct super_block *sb, const char *func,
 330                             unsigned int line)
 331 {
 332         __save_error_info(sb, func, line);
 333         ext4_commit_super(sb, 1);
 334 }
 335
 336 /*
 337  * The del_gendisk() function uninitializes the disk-specific data
 338  * structures, including the bdi structure, without telling anyone
 339  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 340  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 341  * This is a kludge to prevent these oops until we can put in a proper
 342  * hook in del_gendisk() to inform the VFS and file system layers.
 343  */
 344 static int block_device_ejected(struct super_block *sb)
 345 {
 346         struct inode *bd_inode = sb->s_bdev->bd_inode;
 347         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 348
 349         return bdi->dev == NULL;
 350 }
 351
 352 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 353 {
 354         struct super_block              *sb = journal->j_private;
 355         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 356         int                             error = is_journal_aborted(journal);
 357         struct ext4_journal_cb_entry    *jce;
 358
 359         BUG_ON(txn->t_state == T_FINISHED);
 360         spin_lock(&sbi->s_md_lock);
 361         while (!list_empty(&txn->t_private_list)) {
 362                 jce = list_entry(txn->t_private_list.next,
 363                                  struct ext4_journal_cb_entry, jce_list);
 364                 list_del_init(&jce->jce_list);
 365                 spin_unlock(&sbi->s_md_lock);
 366                 jce->jce_func(sb, jce, error);
 367                 spin_lock(&sbi->s_md_lock);
 368         }
 369         spin_unlock(&sbi->s_md_lock);
 370 }
 371
 372 /* Deal with the reporting of failure conditions on a filesystem such as
 373  * inconsistencies detected or read IO failures.
 374  *
 375  * On ext2, we can store the error state of the filesystem in the
 376  * superblock.  That is not possible on ext4, because we may have other
 377  * write ordering constraints on the superblock which prevent us from
 378  * writing it out straight away; and given that the journal is about to
 379  * be aborted, we can't rely on the current, or future, transactions to
 380  * write out the superblock safely.
 381  *
 382  * We'll just use the jbd2_journal_abort() error code to record an error in
 383  * the journal instead.  On recovery, the journal will complain about
 384  * that error until we've noted it down and cleared it.
 385  */
 386
 387 static void ext4_handle_error(struct super_block *sb)
 388 {
 389         if (sb->s_flags & MS_RDONLY)
 390                 return;
 391
 392         if (!test_opt(sb, ERRORS_CONT)) {
 393                 journal_t *journal = EXT4_SB(sb)->s_journal;
 394
 395                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 396                 if (journal)
 397                         jbd2_journal_abort(journal, -EIO);
 398         }
 399         if (test_opt(sb, ERRORS_RO)) {
 400                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 401                 sb->s_flags |= MS_RDONLY;
 402         }
 403         if (test_opt(sb, ERRORS_PANIC))
 404                 panic("EXT4-fs (device %s): panic forced after error\n",
 405                         sb->s_id);
 406 }
 407
 408 void __ext4_error(struct super_block *sb, const char *function,
 409                   unsigned int line, const char *fmt, ...)
 410 {
 411         struct va_format vaf;
 412         va_list args;
 413
 414         va_start(args, fmt);
 415         vaf.fmt = fmt;
 416         vaf.va = &args;
 417         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 418                sb->s_id, function, line, current->comm, &vaf);
 419         va_end(args);
 420         save_error_info(sb, function, line);
 421
 422         ext4_handle_error(sb);
 423 }
 424
 425 void ext4_error_inode(struct inode *inode, const char *function,
 426                       unsigned int line, ext4_fsblk_t block,
 427                       const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct va_format vaf;
 431         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 432
 433         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 434         es->s_last_error_block = cpu_to_le64(block);
 435         save_error_info(inode->i_sb, function, line);
 436         va_start(args, fmt);
 437         vaf.fmt = fmt;
 438         vaf.va = &args;
 439         if (block)
 440                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 441                        "inode #%lu: block %llu: comm %s: %pV\n",
 442                        inode->i_sb->s_id, function, line, inode->i_ino,
 443                        block, current->comm, &vaf);
 444         else
 445                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 446                        "inode #%lu: comm %s: %pV\n",
 447                        inode->i_sb->s_id, function, line, inode->i_ino,
 448                        current->comm, &vaf);
 449         va_end(args);
 450
 451         ext4_handle_error(inode->i_sb);
 452 }
 453
 454 void ext4_error_file(struct file *file, const char *function,
 455                      unsigned int line, ext4_fsblk_t block,
 456                      const char *fmt, ...)
 457 {
 458         va_list args;
 459         struct va_format vaf;
 460         struct ext4_super_block *es;
 461         struct inode *inode = file_inode(file);
 462         char pathname[80], *path;
 463
 464         es = EXT4_SB(inode->i_sb)->s_es;
 465         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 466         save_error_info(inode->i_sb, function, line);
 467         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 468         if (IS_ERR(path))
 469                 path = "(unknown)";
 470         va_start(args, fmt);
 471         vaf.fmt = fmt;
 472         vaf.va = &args;
 473         if (block)
 474                 printk(KERN_CRIT
 475                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 476                        "block %llu: comm %s: path %s: %pV\n",
 477                        inode->i_sb->s_id, function, line, inode->i_ino,
 478                        block, current->comm, path, &vaf);
 479         else
 480                 printk(KERN_CRIT
 481                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 482                        "comm %s: path %s: %pV\n",
 483                        inode->i_sb->s_id, function, line, inode->i_ino,
 484                        current->comm, path, &vaf);
 485         va_end(args);
 486
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EIO:
 497                 errstr = "IO failure";
 498                 break;
 499         case -ENOMEM:
 500                 errstr = "Out of memory";
 501                 break;
 502         case -EROFS:
 503                 if (!sb || (EXT4_SB(sb)->s_journal &&
 504                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 505                         errstr = "Journal has aborted";
 506                 else
 507                         errstr = "Readonly filesystem";
 508                 break;
 509         default:
 510                 /* If the caller passed in an extra buffer for unknown
 511                  * errors, textualise them now.  Else we just return
 512                  * NULL. */
 513                 if (nbuf) {
 514                         /* Check for truncated error codes... */
 515                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 516                                 errstr = nbuf;
 517                 }
 518                 break;
 519         }
 520
 521         return errstr;
 522 }
 523
 524 /* __ext4_std_error decodes expected errors from journaling functions
 525  * automatically and invokes the appropriate error response.  */
 526
 527 void __ext4_std_error(struct super_block *sb, const char *function,
 528                       unsigned int line, int errno)
 529 {
 530         char nbuf[16];
 531         const char *errstr;
 532
 533         /* Special case: if the error is EROFS, and we're not already
 534          * inside a transaction, then there's really no point in logging
 535          * an error. */
 536         if (errno == -EROFS && journal_current_handle() == NULL &&
 537             (sb->s_flags & MS_RDONLY))
 538                 return;
 539
 540         errstr = ext4_decode_error(sb, errno, nbuf);
 541         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 542                sb->s_id, function, line, errstr);
 543         save_error_info(sb, function, line);
 544
 545         ext4_handle_error(sb);
 546 }
 547
 548 /*
 549  * ext4_abort is a much stronger failure handler than ext4_error.  The
 550  * abort function may be used to deal with unrecoverable failures such
 551  * as journal IO errors or ENOMEM at a critical moment in log management.
 552  *
 553  * We unconditionally force the filesystem into an ABORT|READONLY state,
 554  * unless the error response on the fs has been set to panic in which
 555  * case we take the easy way out and panic immediately.
 556  */
 557
 558 void __ext4_abort(struct super_block *sb, const char *function,
 559                 unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         save_error_info(sb, function, line);
 564         va_start(args, fmt);
 565         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 566                function, line);
 567         vprintk(fmt, args);
 568         printk("\n");
 569         va_end(args);
 570
 571         if ((sb->s_flags & MS_RDONLY) == 0) {
 572                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 573                 sb->s_flags |= MS_RDONLY;
 574                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 575                 if (EXT4_SB(sb)->s_journal)
 576                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 577                 save_error_info(sb, function, line);
 578         }
 579         if (test_opt(sb, ERRORS_PANIC))
 580                 panic("EXT4-fs panic from previous error\n");
 581 }
 582
 583 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 584 {
 585         struct va_format vaf;
 586         va_list args;
 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 void __ext4_warning(struct super_block *sb, const char *function,
 596                     unsigned int line, const char *fmt, ...)
 597 {
 598         struct va_format vaf;
 599         va_list args;
 600
 601         va_start(args, fmt);
 602         vaf.fmt = fmt;
 603         vaf.va = &args;
 604         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 605                sb->s_id, function, line, &vaf);
 606         va_end(args);
 607 }
 608
 609 void __ext4_grp_locked_error(const char *function, unsigned int line,
 610                              struct super_block *sb, ext4_group_t grp,
 611                              unsigned long ino, ext4_fsblk_t block,
 612                              const char *fmt, ...)
 613 __releases(bitlock)
 614 __acquires(bitlock)
 615 {
 616         struct va_format vaf;
 617         va_list args;
 618         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 619
 620         es->s_last_error_ino = cpu_to_le32(ino);
 621         es->s_last_error_block = cpu_to_le64(block);
 622         __save_error_info(sb, function, line);
 623
 624         va_start(args, fmt);
 625
 626         vaf.fmt = fmt;
 627         vaf.va = &args;
 628         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 629                sb->s_id, function, line, grp);
 630         if (ino)
 631                 printk(KERN_CONT "inode %lu: ", ino);
 632         if (block)
 633                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 634         printk(KERN_CONT "%pV\n", &vaf);
 635         va_end(args);
 636
 637         if (test_opt(sb, ERRORS_CONT)) {
 638                 ext4_commit_super(sb, 0);
 639                 return;
 640         }
 641
 642         ext4_unlock_group(sb, grp);
 643         ext4_handle_error(sb);
 644         /*
 645          * We only get here in the ERRORS_RO case; relocking the group
 646          * may be dangerous, but nothing bad will happen since the
 647          * filesystem will have already been marked read/only and the
 648          * journal has been aborted.  We return 1 as a hint to callers
 649          * who might what to use the return value from
 650          * ext4_grp_locked_error() to distinguish between the
 651          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 652          * aggressively from the ext4 function in question, with a
 653          * more appropriate error code.
 654          */
 655         ext4_lock_group(sb, grp);
 656         return;
 657 }
 658
 659 void ext4_update_dynamic_rev(struct super_block *sb)
 660 {
 661         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 662
 663         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 664                 return;
 665
 666         ext4_warning(sb,
 667                      "updating to rev %d because of new feature flag, "
 668                      "running e2fsck is recommended",
 669                      EXT4_DYNAMIC_REV);
 670
 671         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 672         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 673         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 674         /* leave es->s_feature_*compat flags alone */
 675         /* es->s_uuid will be set by e2fsck if empty */
 676
 677         /*
 678          * The rest of the superblock fields should be zero, and if not it
 679          * means they are likely already in use, so leave them alone.  We
 680          * can leave it up to e2fsck to clean up any inconsistencies there.
 681          */
 682 }
 683
 684 /*
 685  * Open the external journal device
 686  */
 687 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 688 {
 689         struct block_device *bdev;
 690         char b[BDEVNAME_SIZE];
 691
 692         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 693         if (IS_ERR(bdev))
 694                 goto fail;
 695         return bdev;
 696
 697 fail:
 698         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 699                         __bdevname(dev, b), PTR_ERR(bdev));
 700         return NULL;
 701 }
 702
 703 /*
 704  * Release the journal device
 705  */
 706 static void ext4_blkdev_put(struct block_device *bdev)
 707 {
 708         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 709 }
 710
 711 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 712 {
 713         struct block_device *bdev;
 714         bdev = sbi->journal_bdev;
 715         if (bdev) {
 716                 ext4_blkdev_put(bdev);
 717                 sbi->journal_bdev = NULL;
 718         }
 719 }
 720
 721 static inline struct inode *orphan_list_entry(struct list_head *l)
 722 {
 723         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 724 }
 725
 726 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 727 {
 728         struct list_head *l;
 729
 730         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 731                  le32_to_cpu(sbi->s_es->s_last_orphan));
 732
 733         printk(KERN_ERR "sb_info orphan list:\n");
 734         list_for_each(l, &sbi->s_orphan) {
 735                 struct inode *inode = orphan_list_entry(l);
 736                 printk(KERN_ERR "  "
 737                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 738                        inode->i_sb->s_id, inode->i_ino, inode,
 739                        inode->i_mode, inode->i_nlink,
 740                        NEXT_ORPHAN(inode));
 741         }
 742 }
 743
 744 static void ext4_put_super(struct super_block *sb)
 745 {
 746         struct ext4_sb_info *sbi = EXT4_SB(sb);
 747         struct ext4_super_block *es = sbi->s_es;
 748         int i, err;
 749
 750         ext4_unregister_li_request(sb);
 751         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 752
 753         flush_workqueue(sbi->dio_unwritten_wq);
 754         destroy_workqueue(sbi->dio_unwritten_wq);
 755
 756         if (sbi->s_journal) {
 757                 err = jbd2_journal_destroy(sbi->s_journal);
 758                 sbi->s_journal = NULL;
 759                 if (err < 0)
 760                         ext4_abort(sb, "Couldn't clean up the journal");
 761         }
 762
 763         ext4_es_unregister_shrinker(sb);
 764         del_timer(&sbi->s_err_report);
 765         ext4_release_system_zone(sb);
 766         ext4_mb_release(sb);
 767         ext4_ext_release(sb);
 768         ext4_xattr_put_super(sb);
 769
 770         if (!(sb->s_flags & MS_RDONLY)) {
 771                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 772                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 773         }
 774         if (!(sb->s_flags & MS_RDONLY))
 775                 ext4_commit_super(sb, 1);
 776
 777         if (sbi->s_proc) {
 778                 remove_proc_entry("options", sbi->s_proc);
 779                 remove_proc_entry(sb->s_id, ext4_proc_root);
 780         }
 781         kobject_del(&sbi->s_kobj);
 782
 783         for (i = 0; i < sbi->s_gdb_count; i++)
 784                 brelse(sbi->s_group_desc[i]);
 785         ext4_kvfree(sbi->s_group_desc);
 786         ext4_kvfree(sbi->s_flex_groups);
 787         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 788         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 789         percpu_counter_destroy(&sbi->s_dirs_counter);
 790         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 791         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 792         brelse(sbi->s_sbh);
 793 #ifdef CONFIG_QUOTA
 794         for (i = 0; i < MAXQUOTAS; i++)
 795                 kfree(sbi->s_qf_names[i]);
 796 #endif
 797
 798         /* Debugging code just in case the in-memory inode orphan list
 799          * isn't empty.  The on-disk one can be non-empty if we've
 800          * detected an error and taken the fs readonly, but the
 801          * in-memory list had better be clean by this point. */
 802         if (!list_empty(&sbi->s_orphan))
 803                 dump_orphan_list(sb, sbi);
 804         J_ASSERT(list_empty(&sbi->s_orphan));
 805
 806         sync_blockdev(sb->s_bdev);
 807         invalidate_bdev(sb->s_bdev);
 808         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 809                 /*
 810                  * Invalidate the journal device's buffers.  We don't want them
 811                  * floating about in memory - the physical journal device may
 812                  * hotswapped, and it breaks the `ro-after' testing code.
 813                  */
 814                 sync_blockdev(sbi->journal_bdev);
 815                 invalidate_bdev(sbi->journal_bdev);
 816                 ext4_blkdev_remove(sbi);
 817         }
 818         if (sbi->s_mmp_tsk)
 819                 kthread_stop(sbi->s_mmp_tsk);
 820         sb->s_fs_info = NULL;
 821         /*
 822          * Now that we are completely done shutting down the
 823          * superblock, we need to actually destroy the kobject.
 824          */
 825         kobject_put(&sbi->s_kobj);
 826         wait_for_completion(&sbi->s_kobj_unregister);
 827         if (sbi->s_chksum_driver)
 828                 crypto_free_shash(sbi->s_chksum_driver);
 829         kfree(sbi->s_blockgroup_lock);
 830         kfree(sbi);
 831 }
 832
 833 static struct kmem_cache *ext4_inode_cachep;
 834
 835 /*
 836  * Called inside transaction, so use GFP_NOFS
 837  */
 838 static struct inode *ext4_alloc_inode(struct super_block *sb)
 839 {
 840         struct ext4_inode_info *ei;
 841
 842         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 843         if (!ei)
 844                 return NULL;
 845
 846         ei->vfs_inode.i_version = 1;
 847         INIT_LIST_HEAD(&ei->i_prealloc_list);
 848         spin_lock_init(&ei->i_prealloc_lock);
 849         ext4_es_init_tree(&ei->i_es_tree);
 850         rwlock_init(&ei->i_es_lock);
 851         INIT_LIST_HEAD(&ei->i_es_lru);
 852         ei->i_es_lru_nr = 0;
 853         ei->i_reserved_data_blocks = 0;
 854         ei->i_reserved_meta_blocks = 0;
 855         ei->i_allocated_meta_blocks = 0;
 856         ei->i_da_metadata_calc_len = 0;
 857         ei->i_da_metadata_calc_last_lblock = 0;
 858         spin_lock_init(&(ei->i_block_reservation_lock));
 859 #ifdef CONFIG_QUOTA
 860         ei->i_reserved_quota = 0;
 861 #endif
 862         ei->jinode = NULL;
 863         INIT_LIST_HEAD(&ei->i_completed_io_list);
 864         spin_lock_init(&ei->i_completed_io_lock);
 865         ei->i_sync_tid = 0;
 866         ei->i_datasync_tid = 0;
 867         atomic_set(&ei->i_ioend_count, 0);
 868         atomic_set(&ei->i_unwritten, 0);
 869         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
 870
 871         return &ei->vfs_inode;
 872 }
 873
 874 static int ext4_drop_inode(struct inode *inode)
 875 {
 876         int drop = generic_drop_inode(inode);
 877
 878         trace_ext4_drop_inode(inode, drop);
 879         return drop;
 880 }
 881
 882 static void ext4_i_callback(struct rcu_head *head)
 883 {
 884         struct inode *inode = container_of(head, struct inode, i_rcu);
 885         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 886 }
 887
 888 static void ext4_destroy_inode(struct inode *inode)
 889 {
 890         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 891                 ext4_msg(inode->i_sb, KERN_ERR,
 892                          "Inode %lu (%p): orphan list check failed!",
 893                          inode->i_ino, EXT4_I(inode));
 894                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 895                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 896                                 true);
 897                 dump_stack();
 898         }
 899         call_rcu(&inode->i_rcu, ext4_i_callback);
 900 }
 901
 902 static void init_once(void *foo)
 903 {
 904         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 905
 906         INIT_LIST_HEAD(&ei->i_orphan);
 907         init_rwsem(&ei->xattr_sem);
 908         init_rwsem(&ei->i_data_sem);
 909         inode_init_once(&ei->vfs_inode);
 910 }
 911
 912 static int init_inodecache(void)
 913 {
 914         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 915                                              sizeof(struct ext4_inode_info),
 916                                              0, (SLAB_RECLAIM_ACCOUNT|
 917                                                 SLAB_MEM_SPREAD),
 918                                              init_once);
 919         if (ext4_inode_cachep == NULL)
 920                 return -ENOMEM;
 921         return 0;
 922 }
 923
 924 static void destroy_inodecache(void)
 925 {
 926         /*
 927          * Make sure all delayed rcu free inodes are flushed before we
 928          * destroy cache.
 929          */
 930         rcu_barrier();
 931         kmem_cache_destroy(ext4_inode_cachep);
 932 }
 933
 934 void ext4_clear_inode(struct inode *inode)
 935 {
 936         invalidate_inode_buffers(inode);
 937         clear_inode(inode);
 938         dquot_drop(inode);
 939         ext4_discard_preallocations(inode);
 940         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 941         ext4_es_lru_del(inode);
 942         if (EXT4_I(inode)->jinode) {
 943                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 944                                                EXT4_I(inode)->jinode);
 945                 jbd2_free_inode(EXT4_I(inode)->jinode);
 946                 EXT4_I(inode)->jinode = NULL;
 947         }
 948 }
 949
 950 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 951                                         u64 ino, u32 generation)
 952 {
 953         struct inode *inode;
 954
 955         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 956                 return ERR_PTR(-ESTALE);
 957         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 958                 return ERR_PTR(-ESTALE);
 959
 960         /* iget isn't really right if the inode is currently unallocated!!
 961          *
 962          * ext4_read_inode will return a bad_inode if the inode had been
 963          * deleted, so we should be safe.
 964          *
 965          * Currently we don't know the generation for parent directory, so
 966          * a generation of 0 means "accept any"
 967          */
 968         inode = ext4_iget_normal(sb, ino);
 969         if (IS_ERR(inode))
 970                 return ERR_CAST(inode);
 971         if (generation && inode->i_generation != generation) {
 972                 iput(inode);
 973                 return ERR_PTR(-ESTALE);
 974         }
 975
 976         return inode;
 977 }
 978
 979 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 980                                         int fh_len, int fh_type)
 981 {
 982         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 983                                     ext4_nfs_get_inode);
 984 }
 985
 986 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 987                                         int fh_len, int fh_type)
 988 {
 989         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 990                                     ext4_nfs_get_inode);
 991 }
 992
 993 /*
 994  * Try to release metadata pages (indirect blocks, directories) which are
 995  * mapped via the block device.  Since these pages could have journal heads
 996  * which would prevent try_to_free_buffers() from freeing them, we must use
 997  * jbd2 layer's try_to_free_buffers() function to release them.
 998  */
 999 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1000                                  gfp_t wait)
1001 {
1002         journal_t *journal = EXT4_SB(sb)->s_journal;
1003
1004         WARN_ON(PageChecked(page));
1005         if (!page_has_buffers(page))
1006                 return 0;
1007         if (journal)
1008                 return jbd2_journal_try_to_free_buffers(journal, page,
1009                                                         wait & ~__GFP_WAIT);
1010         return try_to_free_buffers(page);
1011 }
1012
1013 #ifdef CONFIG_QUOTA
1014 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1015 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1016
1017 static int ext4_write_dquot(struct dquot *dquot);
1018 static int ext4_acquire_dquot(struct dquot *dquot);
1019 static int ext4_release_dquot(struct dquot *dquot);
1020 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1021 static int ext4_write_info(struct super_block *sb, int type);
1022 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1023                          struct path *path);
1024 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1025                                  int format_id);
1026 static int ext4_quota_off(struct super_block *sb, int type);
1027 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1028 static int ext4_quota_on_mount(struct super_block *sb, int type);
1029 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1030                                size_t len, loff_t off);
1031 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1032                                 const char *data, size_t len, loff_t off);
1033 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1034                              unsigned int flags);
1035 static int ext4_enable_quotas(struct super_block *sb);
1036
1037 static const struct dquot_operations ext4_quota_operations = {
1038         .get_reserved_space = ext4_get_reserved_space,
1039         .write_dquot    = ext4_write_dquot,
1040         .acquire_dquot  = ext4_acquire_dquot,
1041         .release_dquot  = ext4_release_dquot,
1042         .mark_dirty     = ext4_mark_dquot_dirty,
1043         .write_info     = ext4_write_info,
1044         .alloc_dquot    = dquot_alloc,
1045         .destroy_dquot  = dquot_destroy,
1046 };
1047
1048 static const struct quotactl_ops ext4_qctl_operations = {
1049         .quota_on       = ext4_quota_on,
1050         .quota_off      = ext4_quota_off,
1051         .quota_sync     = dquot_quota_sync,
1052         .get_info       = dquot_get_dqinfo,
1053         .set_info       = dquot_set_dqinfo,
1054         .get_dqblk      = dquot_get_dqblk,
1055         .set_dqblk      = dquot_set_dqblk
1056 };
1057
1058 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1059         .quota_on_meta  = ext4_quota_on_sysfile,
1060         .quota_off      = ext4_quota_off_sysfile,
1061         .quota_sync     = dquot_quota_sync,
1062         .get_info       = dquot_get_dqinfo,
1063         .set_info       = dquot_set_dqinfo,
1064         .get_dqblk      = dquot_get_dqblk,
1065         .set_dqblk      = dquot_set_dqblk
1066 };
1067 #endif
1068
1069 static const struct super_operations ext4_sops = {
1070         .alloc_inode    = ext4_alloc_inode,
1071         .destroy_inode  = ext4_destroy_inode,
1072         .write_inode    = ext4_write_inode,
1073         .dirty_inode    = ext4_dirty_inode,
1074         .drop_inode     = ext4_drop_inode,
1075         .evict_inode    = ext4_evict_inode,
1076         .put_super      = ext4_put_super,
1077         .sync_fs        = ext4_sync_fs,
1078         .freeze_fs      = ext4_freeze,
1079         .unfreeze_fs    = ext4_unfreeze,
1080         .statfs         = ext4_statfs,
1081         .remount_fs     = ext4_remount,
1082         .show_options   = ext4_show_options,
1083 #ifdef CONFIG_QUOTA
1084         .quota_read     = ext4_quota_read,
1085         .quota_write    = ext4_quota_write,
1086 #endif
1087         .bdev_try_to_free_page = bdev_try_to_free_page,
1088 };
1089
1090 static const struct super_operations ext4_nojournal_sops = {
1091         .alloc_inode    = ext4_alloc_inode,
1092         .destroy_inode  = ext4_destroy_inode,
1093         .write_inode    = ext4_write_inode,
1094         .dirty_inode    = ext4_dirty_inode,
1095         .drop_inode     = ext4_drop_inode,
1096         .evict_inode    = ext4_evict_inode,
1097         .put_super      = ext4_put_super,
1098         .statfs         = ext4_statfs,
1099         .remount_fs     = ext4_remount,
1100         .show_options   = ext4_show_options,
1101 #ifdef CONFIG_QUOTA
1102         .quota_read     = ext4_quota_read,
1103         .quota_write    = ext4_quota_write,
1104 #endif
1105         .bdev_try_to_free_page = bdev_try_to_free_page,
1106 };
1107
1108 static const struct export_operations ext4_export_ops = {
1109         .fh_to_dentry = ext4_fh_to_dentry,
1110         .fh_to_parent = ext4_fh_to_parent,
1111         .get_parent = ext4_get_parent,
1112 };
1113
1114 enum {
1115         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1116         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1117         Opt_nouid32, Opt_debug, Opt_removed,
1118         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1119         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1120         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1121         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1122         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1123         Opt_data_err_abort, Opt_data_err_ignore,
1124         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1125         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1126         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1127         Opt_usrquota, Opt_grpquota, Opt_i_version,
1128         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1129         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1130         Opt_inode_readahead_blks, Opt_journal_ioprio,
1131         Opt_dioread_nolock, Opt_dioread_lock,
1132         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1133         Opt_max_dir_size_kb,
1134 };
1135
1136 static const match_table_t tokens = {
1137         {Opt_bsd_df, "bsddf"},
1138         {Opt_minix_df, "minixdf"},
1139         {Opt_grpid, "grpid"},
1140         {Opt_grpid, "bsdgroups"},
1141         {Opt_nogrpid, "nogrpid"},
1142         {Opt_nogrpid, "sysvgroups"},
1143         {Opt_resgid, "resgid=%u"},
1144         {Opt_resuid, "resuid=%u"},
1145         {Opt_sb, "sb=%u"},
1146         {Opt_err_cont, "errors=continue"},
1147         {Opt_err_panic, "errors=panic"},
1148         {Opt_err_ro, "errors=remount-ro"},
1149         {Opt_nouid32, "nouid32"},
1150         {Opt_debug, "debug"},
1151         {Opt_removed, "oldalloc"},
1152         {Opt_removed, "orlov"},
1153         {Opt_user_xattr, "user_xattr"},
1154         {Opt_nouser_xattr, "nouser_xattr"},
1155         {Opt_acl, "acl"},
1156         {Opt_noacl, "noacl"},
1157         {Opt_noload, "norecovery"},
1158         {Opt_noload, "noload"},
1159         {Opt_removed, "nobh"},
1160         {Opt_removed, "bh"},
1161         {Opt_commit, "commit=%u"},
1162         {Opt_min_batch_time, "min_batch_time=%u"},
1163         {Opt_max_batch_time, "max_batch_time=%u"},
1164         {Opt_journal_dev, "journal_dev=%u"},
1165         {Opt_journal_checksum, "journal_checksum"},
1166         {Opt_journal_async_commit, "journal_async_commit"},
1167         {Opt_abort, "abort"},
1168         {Opt_data_journal, "data=journal"},
1169         {Opt_data_ordered, "data=ordered"},
1170         {Opt_data_writeback, "data=writeback"},
1171         {Opt_data_err_abort, "data_err=abort"},
1172         {Opt_data_err_ignore, "data_err=ignore"},
1173         {Opt_offusrjquota, "usrjquota="},
1174         {Opt_usrjquota, "usrjquota=%s"},
1175         {Opt_offgrpjquota, "grpjquota="},
1176         {Opt_grpjquota, "grpjquota=%s"},
1177         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1178         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1179         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1180         {Opt_grpquota, "grpquota"},
1181         {Opt_noquota, "noquota"},
1182         {Opt_quota, "quota"},
1183         {Opt_usrquota, "usrquota"},
1184         {Opt_barrier, "barrier=%u"},
1185         {Opt_barrier, "barrier"},
1186         {Opt_nobarrier, "nobarrier"},
1187         {Opt_i_version, "i_version"},
1188         {Opt_stripe, "stripe=%u"},
1189         {Opt_delalloc, "delalloc"},
1190         {Opt_nodelalloc, "nodelalloc"},
1191         {Opt_removed, "mblk_io_submit"},
1192         {Opt_removed, "nomblk_io_submit"},
1193         {Opt_block_validity, "block_validity"},
1194         {Opt_noblock_validity, "noblock_validity"},
1195         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1196         {Opt_journal_ioprio, "journal_ioprio=%u"},
1197         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1198         {Opt_auto_da_alloc, "auto_da_alloc"},
1199         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1200         {Opt_dioread_nolock, "dioread_nolock"},
1201         {Opt_dioread_lock, "dioread_lock"},
1202         {Opt_discard, "discard"},
1203         {Opt_nodiscard, "nodiscard"},
1204         {Opt_init_itable, "init_itable=%u"},
1205         {Opt_init_itable, "init_itable"},
1206         {Opt_noinit_itable, "noinit_itable"},
1207         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1208         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1209         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1210         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1211         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1212         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1213         {Opt_err, NULL},
1214 };
1215
1216 static ext4_fsblk_t get_sb_block(void **data)
1217 {
1218         ext4_fsblk_t    sb_block;
1219         char            *options = (char *) *data;
1220
1221         if (!options || strncmp(options, "sb=", 3) != 0)
1222                 return 1;       /* Default location */
1223
1224         options += 3;
1225         /* TODO: use simple_strtoll with >32bit ext4 */
1226         sb_block = simple_strtoul(options, &options, 0);
1227         if (*options && *options != ',') {
1228                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1229                        (char *) *data);
1230                 return 1;
1231         }
1232         if (*options == ',')
1233                 options++;
1234         *data = (void *) options;
1235
1236         return sb_block;
1237 }
1238
1239 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1240 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1241         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1242
1243 #ifdef CONFIG_QUOTA
1244 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1245 {
1246         struct ext4_sb_info *sbi = EXT4_SB(sb);
1247         char *qname;
1248         int ret = -1;
1249
1250         if (sb_any_quota_loaded(sb) &&
1251                 !sbi->s_qf_names[qtype]) {
1252                 ext4_msg(sb, KERN_ERR,
1253                         "Cannot change journaled "
1254                         "quota options when quota turned on");
1255                 return -1;
1256         }
1257         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1258                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1259                          "when QUOTA feature is enabled");
1260                 return -1;
1261         }
1262         qname = match_strdup(args);
1263         if (!qname) {
1264                 ext4_msg(sb, KERN_ERR,
1265                         "Not enough memory for storing quotafile name");
1266                 return -1;
1267         }
1268         if (sbi->s_qf_names[qtype]) {
1269                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1270                         ret = 1;
1271                 else
1272                         ext4_msg(sb, KERN_ERR,
1273                                  "%s quota file already specified",
1274                                  QTYPE2NAME(qtype));
1275                 goto errout;
1276         }
1277         if (strchr(qname, '/')) {
1278                 ext4_msg(sb, KERN_ERR,
1279                         "quotafile must be on filesystem root");
1280                 goto errout;
1281         }
1282         sbi->s_qf_names[qtype] = qname;
1283         set_opt(sb, QUOTA);
1284         return 1;
1285 errout:
1286         kfree(qname);
1287         return ret;
1288 }
1289
1290 static int clear_qf_name(struct super_block *sb, int qtype)
1291 {
1292
1293         struct ext4_sb_info *sbi = EXT4_SB(sb);
1294
1295         if (sb_any_quota_loaded(sb) &&
1296                 sbi->s_qf_names[qtype]) {
1297                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1298                         " when quota turned on");
1299                 return -1;
1300         }
1301         kfree(sbi->s_qf_names[qtype]);
1302         sbi->s_qf_names[qtype] = NULL;
1303         return 1;
1304 }
1305 #endif
1306
1307 #define MOPT_SET        0x0001
1308 #define MOPT_CLEAR      0x0002
1309 #define MOPT_NOSUPPORT  0x0004
1310 #define MOPT_EXPLICIT   0x0008
1311 #define MOPT_CLEAR_ERR  0x0010
1312 #define MOPT_GTE0       0x0020
1313 #ifdef CONFIG_QUOTA
1314 #define MOPT_Q          0
1315 #define MOPT_QFMT       0x0040
1316 #else
1317 #define MOPT_Q          MOPT_NOSUPPORT
1318 #define MOPT_QFMT       MOPT_NOSUPPORT
1319 #endif
1320 #define MOPT_DATAJ      0x0080
1321 #define MOPT_NO_EXT2    0x0100
1322 #define MOPT_NO_EXT3    0x0200
1323 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1324
1325 static const struct mount_opts {
1326         int     token;
1327         int     mount_opt;
1328         int     flags;
1329 } ext4_mount_opts[] = {
1330         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1331         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1332         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1333         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1334         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1335         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1336         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1337          MOPT_EXT4_ONLY | MOPT_SET},
1338         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1339          MOPT_EXT4_ONLY | MOPT_CLEAR},
1340         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1341         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1342         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1343          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1344         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1345          MOPT_EXT4_ONLY | MOPT_CLEAR},
1346         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1347          MOPT_EXT4_ONLY | MOPT_SET},
1348         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1349                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1350          MOPT_EXT4_ONLY | MOPT_SET},
1351         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1352         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1353         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1354         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1355         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1356          MOPT_NO_EXT2 | MOPT_SET},
1357         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1358          MOPT_NO_EXT2 | MOPT_CLEAR},
1359         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1360         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1361         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1362         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1363         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1364         {Opt_commit, 0, MOPT_GTE0},
1365         {Opt_max_batch_time, 0, MOPT_GTE0},
1366         {Opt_min_batch_time, 0, MOPT_GTE0},
1367         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1368         {Opt_init_itable, 0, MOPT_GTE0},
1369         {Opt_stripe, 0, MOPT_GTE0},
1370         {Opt_resuid, 0, MOPT_GTE0},
1371         {Opt_resgid, 0, MOPT_GTE0},
1372         {Opt_journal_dev, 0, MOPT_GTE0},
1373         {Opt_journal_ioprio, 0, MOPT_GTE0},
1374         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1375         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1376         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1377          MOPT_NO_EXT2 | MOPT_DATAJ},
1378         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1379         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1380 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1381         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1382         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1383 #else
1384         {Opt_acl, 0, MOPT_NOSUPPORT},
1385         {Opt_noacl, 0, MOPT_NOSUPPORT},
1386 #endif
1387         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1388         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1389         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1390         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1391                                                         MOPT_SET | MOPT_Q},
1392         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1393                                                         MOPT_SET | MOPT_Q},
1394         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1395                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1396         {Opt_usrjquota, 0, MOPT_Q},
1397         {Opt_grpjquota, 0, MOPT_Q},
1398         {Opt_offusrjquota, 0, MOPT_Q},
1399         {Opt_offgrpjquota, 0, MOPT_Q},
1400         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1401         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1402         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1403         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1404         {Opt_err, 0, 0}
1405 };
1406
1407 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1408                             substring_t *args, unsigned long *journal_devnum,
1409                             unsigned int *journal_ioprio, int is_remount)
1410 {
1411         struct ext4_sb_info *sbi = EXT4_SB(sb);
1412         const struct mount_opts *m;
1413         kuid_t uid;
1414         kgid_t gid;
1415         int arg = 0;
1416
1417 #ifdef CONFIG_QUOTA
1418         if (token == Opt_usrjquota)
1419                 return set_qf_name(sb, USRQUOTA, &args[0]);
1420         else if (token == Opt_grpjquota)
1421                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1422         else if (token == Opt_offusrjquota)
1423                 return clear_qf_name(sb, USRQUOTA);
1424         else if (token == Opt_offgrpjquota)
1425                 return clear_qf_name(sb, GRPQUOTA);
1426 #endif
1427         switch (token) {
1428         case Opt_noacl:
1429         case Opt_nouser_xattr:
1430                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1431                 break;
1432         case Opt_sb:
1433                 return 1;       /* handled by get_sb_block() */
1434         case Opt_removed:
1435                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1436                 return 1;
1437         case Opt_abort:
1438                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1439                 return 1;
1440         case Opt_i_version:
1441                 sb->s_flags |= MS_I_VERSION;
1442                 return 1;
1443         }
1444
1445         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1446                 if (token == m->token)
1447                         break;
1448
1449         if (m->token == Opt_err) {
1450                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1451                          "or missing value", opt);
1452                 return -1;
1453         }
1454
1455         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1456                 ext4_msg(sb, KERN_ERR,
1457                          "Mount option \"%s\" incompatible with ext2", opt);
1458                 return -1;
1459         }
1460         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1461                 ext4_msg(sb, KERN_ERR,
1462                          "Mount option \"%s\" incompatible with ext3", opt);
1463                 return -1;
1464         }
1465
1466         if (args->from && match_int(args, &arg))
1467                 return -1;
1468         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1469                 return -1;
1470         if (m->flags & MOPT_EXPLICIT)
1471                 set_opt2(sb, EXPLICIT_DELALLOC);
1472         if (m->flags & MOPT_CLEAR_ERR)
1473                 clear_opt(sb, ERRORS_MASK);
1474         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1475                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1476                          "options when quota turned on");
1477                 return -1;
1478         }
1479
1480         if (m->flags & MOPT_NOSUPPORT) {
1481                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1482         } else if (token == Opt_commit) {
1483                 if (arg == 0)
1484                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1485                 sbi->s_commit_interval = HZ * arg;
1486         } else if (token == Opt_max_batch_time) {
1487                 sbi->s_max_batch_time = arg;
1488         } else if (token == Opt_min_batch_time) {
1489                 sbi->s_min_batch_time = arg;
1490         } else if (token == Opt_inode_readahead_blks) {
1491                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1492                         ext4_msg(sb, KERN_ERR,
1493                                  "EXT4-fs: inode_readahead_blks must be "
1494                                  "0 or a power of 2 smaller than 2^31");
1495                         return -1;
1496                 }
1497                 sbi->s_inode_readahead_blks = arg;
1498         } else if (token == Opt_init_itable) {
1499                 set_opt(sb, INIT_INODE_TABLE);
1500                 if (!args->from)
1501                         arg = EXT4_DEF_LI_WAIT_MULT;
1502                 sbi->s_li_wait_mult = arg;
1503         } else if (token == Opt_max_dir_size_kb) {
1504                 sbi->s_max_dir_size_kb = arg;
1505         } else if (token == Opt_stripe) {
1506                 sbi->s_stripe = arg;
1507         } else if (token == Opt_resuid) {
1508                 uid = make_kuid(current_user_ns(), arg);
1509                 if (!uid_valid(uid)) {
1510                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1511                         return -1;
1512                 }
1513                 sbi->s_resuid = uid;
1514         } else if (token == Opt_resgid) {
1515                 gid = make_kgid(current_user_ns(), arg);
1516                 if (!gid_valid(gid)) {
1517                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1518                         return -1;
1519                 }
1520                 sbi->s_resgid = gid;
1521         } else if (token == Opt_journal_dev) {
1522                 if (is_remount) {
1523                         ext4_msg(sb, KERN_ERR,
1524                                  "Cannot specify journal on remount");
1525                         return -1;
1526                 }
1527                 *journal_devnum = arg;
1528         } else if (token == Opt_journal_ioprio) {
1529                 if (arg > 7) {
1530                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1531                                  " (must be 0-7)");
1532                         return -1;
1533                 }
1534                 *journal_ioprio =
1535                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1536         } else if (m->flags & MOPT_DATAJ) {
1537                 if (is_remount) {
1538                         if (!sbi->s_journal)
1539                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1540                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1541                                 ext4_msg(sb, KERN_ERR,
1542                                          "Cannot change data mode on remount");
1543                                 return -1;
1544                         }
1545                 } else {
1546                         clear_opt(sb, DATA_FLAGS);
1547                         sbi->s_mount_opt |= m->mount_opt;
1548                 }
1549 #ifdef CONFIG_QUOTA
1550         } else if (m->flags & MOPT_QFMT) {
1551                 if (sb_any_quota_loaded(sb) &&
1552                     sbi->s_jquota_fmt != m->mount_opt) {
1553                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1554                                  "quota options when quota turned on");
1555                         return -1;
1556                 }
1557                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1558                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1559                         ext4_msg(sb, KERN_ERR,
1560                                  "Cannot set journaled quota options "
1561                                  "when QUOTA feature is enabled");
1562                         return -1;
1563                 }
1564                 sbi->s_jquota_fmt = m->mount_opt;
1565 #endif
1566         } else {
1567                 if (!args->from)
1568                         arg = 1;
1569                 if (m->flags & MOPT_CLEAR)
1570                         arg = !arg;
1571                 else if (unlikely(!(m->flags & MOPT_SET))) {
1572                         ext4_msg(sb, KERN_WARNING,
1573                                  "buggy handling of option %s", opt);
1574                         WARN_ON(1);
1575                         return -1;
1576                 }
1577                 if (arg != 0)
1578                         sbi->s_mount_opt |= m->mount_opt;
1579                 else
1580                         sbi->s_mount_opt &= ~m->mount_opt;
1581         }
1582         return 1;
1583 }
1584
1585 static int parse_options(char *options, struct super_block *sb,
1586                          unsigned long *journal_devnum,
1587                          unsigned int *journal_ioprio,
1588                          int is_remount)
1589 {
1590         struct ext4_sb_info *sbi = EXT4_SB(sb);
1591         char *p;
1592         substring_t args[MAX_OPT_ARGS];
1593         int token;
1594
1595         if (!options)
1596                 return 1;
1597
1598         while ((p = strsep(&options, ",")) != NULL) {
1599                 if (!*p)
1600                         continue;
1601                 /*
1602                  * Initialize args struct so we know whether arg was
1603                  * found; some options take optional arguments.
1604                  */
1605                 args[0].to = args[0].from = NULL;
1606                 token = match_token(p, tokens, args);
1607                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1608                                      journal_ioprio, is_remount) < 0)
1609                         return 0;
1610         }
1611 #ifdef CONFIG_QUOTA
1612         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1613             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1614                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1615                          "feature is enabled");
1616                 return 0;
1617         }
1618         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1619                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1620                         clear_opt(sb, USRQUOTA);
1621
1622                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1623                         clear_opt(sb, GRPQUOTA);
1624
1625                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1626                         ext4_msg(sb, KERN_ERR, "old and new quota "
1627                                         "format mixing");
1628                         return 0;
1629                 }
1630
1631                 if (!sbi->s_jquota_fmt) {
1632                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1633                                         "not specified");
1634                         return 0;
1635                 }
1636         }
1637 #endif
1638         if (test_opt(sb, DIOREAD_NOLOCK)) {
1639                 int blocksize =
1640                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1641
1642                 if (blocksize < PAGE_CACHE_SIZE) {
1643                         ext4_msg(sb, KERN_ERR, "can't mount with "
1644                                  "dioread_nolock if block size != PAGE_SIZE");
1645                         return 0;
1646                 }
1647         }
1648         return 1;
1649 }
1650
1651 static inline void ext4_show_quota_options(struct seq_file *seq,
1652                                            struct super_block *sb)
1653 {
1654 #if defined(CONFIG_QUOTA)
1655         struct ext4_sb_info *sbi = EXT4_SB(sb);
1656
1657         if (sbi->s_jquota_fmt) {
1658                 char *fmtname = "";
1659
1660                 switch (sbi->s_jquota_fmt) {
1661                 case QFMT_VFS_OLD:
1662                         fmtname = "vfsold";
1663                         break;
1664                 case QFMT_VFS_V0:
1665                         fmtname = "vfsv0";
1666                         break;
1667                 case QFMT_VFS_V1:
1668                         fmtname = "vfsv1";
1669                         break;
1670                 }
1671                 seq_printf(seq, ",jqfmt=%s", fmtname);
1672         }
1673
1674         if (sbi->s_qf_names[USRQUOTA])
1675                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1676
1677         if (sbi->s_qf_names[GRPQUOTA])
1678                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1679 #endif
1680 }
1681
1682 static const char *token2str(int token)
1683 {
1684         const struct match_token *t;
1685
1686         for (t = tokens; t->token != Opt_err; t++)
1687                 if (t->token == token && !strchr(t->pattern, '='))
1688                         break;
1689         return t->pattern;
1690 }
1691
1692 /*
1693  * Show an option if
1694  *  - it's set to a non-default value OR
1695  *  - if the per-sb default is different from the global default
1696  */
1697 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1698                               int nodefs)
1699 {
1700         struct ext4_sb_info *sbi = EXT4_SB(sb);
1701         struct ext4_super_block *es = sbi->s_es;
1702         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1703         const struct mount_opts *m;
1704         char sep = nodefs ? '\n' : ',';
1705
1706 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1707 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1708
1709         if (sbi->s_sb_block != 1)
1710                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1711
1712         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1713                 int want_set = m->flags & MOPT_SET;
1714                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1715                     (m->flags & MOPT_CLEAR_ERR))
1716                         continue;
1717                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1718                         continue; /* skip if same as the default */
1719                 if ((want_set &&
1720                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1721                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1722                         continue; /* select Opt_noFoo vs Opt_Foo */
1723                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1724         }
1725
1726         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1727             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1728                 SEQ_OPTS_PRINT("resuid=%u",
1729                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1730         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1731             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1732                 SEQ_OPTS_PRINT("resgid=%u",
1733                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1734         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1735         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1736                 SEQ_OPTS_PUTS("errors=remount-ro");
1737         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1738                 SEQ_OPTS_PUTS("errors=continue");
1739         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1740                 SEQ_OPTS_PUTS("errors=panic");
1741         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1742                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1743         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1744                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1745         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1746                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1747         if (sb->s_flags & MS_I_VERSION)
1748                 SEQ_OPTS_PUTS("i_version");
1749         if (nodefs || sbi->s_stripe)
1750                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1751         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1752                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1753                         SEQ_OPTS_PUTS("data=journal");
1754                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1755                         SEQ_OPTS_PUTS("data=ordered");
1756                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1757                         SEQ_OPTS_PUTS("data=writeback");
1758         }
1759         if (nodefs ||
1760             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1761                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1762                                sbi->s_inode_readahead_blks);
1763
1764         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1765                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1766                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1767         if (nodefs || sbi->s_max_dir_size_kb)
1768                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1769
1770         ext4_show_quota_options(seq, sb);
1771         return 0;
1772 }
1773
1774 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1775 {
1776         return _ext4_show_options(seq, root->d_sb, 0);
1777 }
1778
1779 static int options_seq_show(struct seq_file *seq, void *offset)
1780 {
1781         struct super_block *sb = seq->private;
1782         int rc;
1783
1784         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1785         rc = _ext4_show_options(seq, sb, 1);
1786         seq_puts(seq, "\n");
1787         return rc;
1788 }
1789
1790 static int options_open_fs(struct inode *inode, struct file *file)
1791 {
1792         return single_open(file, options_seq_show, PDE_DATA(inode));
1793 }
1794
1795 static const struct file_operations ext4_seq_options_fops = {
1796         .owner = THIS_MODULE,
1797         .open = options_open_fs,
1798         .read = seq_read,
1799         .llseek = seq_lseek,
1800         .release = single_release,
1801 };
1802
1803 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1804                             int read_only)
1805 {
1806         struct ext4_sb_info *sbi = EXT4_SB(sb);
1807         int res = 0;
1808
1809         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1810                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1811                          "forcing read-only mode");
1812                 res = MS_RDONLY;
1813         }
1814         if (read_only)
1815                 goto done;
1816         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1817                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1818                          "running e2fsck is recommended");
1819         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1820                 ext4_msg(sb, KERN_WARNING,
1821                          "warning: mounting fs with errors, "
1822                          "running e2fsck is recommended");
1823         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1824                  le16_to_cpu(es->s_mnt_count) >=
1825                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1826                 ext4_msg(sb, KERN_WARNING,
1827                          "warning: maximal mount count reached, "
1828                          "running e2fsck is recommended");
1829         else if (le32_to_cpu(es->s_checkinterval) &&
1830                 (le32_to_cpu(es->s_lastcheck) +
1831                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1832                 ext4_msg(sb, KERN_WARNING,
1833                          "warning: checktime reached, "
1834                          "running e2fsck is recommended");
1835         if (!sbi->s_journal)
1836                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1837         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1838                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1839         le16_add_cpu(&es->s_mnt_count, 1);
1840         es->s_mtime = cpu_to_le32(get_seconds());
1841         ext4_update_dynamic_rev(sb);
1842         if (sbi->s_journal)
1843                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1844
1845         ext4_commit_super(sb, 1);
1846 done:
1847         if (test_opt(sb, DEBUG))
1848                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1849                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1850                         sb->s_blocksize,
1851                         sbi->s_groups_count,
1852                         EXT4_BLOCKS_PER_GROUP(sb),
1853                         EXT4_INODES_PER_GROUP(sb),
1854                         sbi->s_mount_opt, sbi->s_mount_opt2);
1855
1856         cleancache_init_fs(sb);
1857         return res;
1858 }
1859
1860 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1861 {
1862         struct ext4_sb_info *sbi = EXT4_SB(sb);
1863         struct flex_groups *new_groups;
1864         int size;
1865
1866         if (!sbi->s_log_groups_per_flex)
1867                 return 0;
1868
1869         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1870         if (size <= sbi->s_flex_groups_allocated)
1871                 return 0;
1872
1873         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1874         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1875         if (!new_groups) {
1876                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1877                          size / (int) sizeof(struct flex_groups));
1878                 return -ENOMEM;
1879         }
1880
1881         if (sbi->s_flex_groups) {
1882                 memcpy(new_groups, sbi->s_flex_groups,
1883                        (sbi->s_flex_groups_allocated *
1884                         sizeof(struct flex_groups)));
1885                 ext4_kvfree(sbi->s_flex_groups);
1886         }
1887         sbi->s_flex_groups = new_groups;
1888         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1889         return 0;
1890 }
1891
1892 static int ext4_fill_flex_info(struct super_block *sb)
1893 {
1894         struct ext4_sb_info *sbi = EXT4_SB(sb);
1895         struct ext4_group_desc *gdp = NULL;
1896         ext4_group_t flex_group;
1897         unsigned int groups_per_flex = 0;
1898         int i, err;
1899
1900         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1901         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1902                 sbi->s_log_groups_per_flex = 0;
1903                 return 1;
1904         }
1905         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1906
1907         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1908         if (err)
1909                 goto failed;
1910
1911         for (i = 0; i < sbi->s_groups_count; i++) {
1912                 gdp = ext4_get_group_desc(sb, i, NULL);
1913
1914                 flex_group = ext4_flex_group(sbi, i);
1915                 atomic_add(ext4_free_inodes_count(sb, gdp),
1916                            &sbi->s_flex_groups[flex_group].free_inodes);
1917                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1918                              &sbi->s_flex_groups[flex_group].free_clusters);
1919                 atomic_add(ext4_used_dirs_count(sb, gdp),
1920                            &sbi->s_flex_groups[flex_group].used_dirs);
1921         }
1922
1923         return 1;
1924 failed:
1925         return 0;
1926 }
1927
1928 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1929                                    struct ext4_group_desc *gdp)
1930 {
1931         int offset;
1932         __u16 crc = 0;
1933         __le32 le_group = cpu_to_le32(block_group);
1934
1935         if ((sbi->s_es->s_feature_ro_compat &
1936              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1937                 /* Use new metadata_csum algorithm */
1938                 __le16 save_csum;
1939                 __u32 csum32;
1940
1941                 save_csum = gdp->bg_checksum;
1942                 gdp->bg_checksum = 0;
1943                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1944                                      sizeof(le_group));
1945                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1946                                      sbi->s_desc_size);
1947                 gdp->bg_checksum = save_csum;
1948
1949                 crc = csum32 & 0xFFFF;
1950                 goto out;
1951         }
1952
1953         /* old crc16 code */
1954         if (!(sbi->s_es->s_feature_ro_compat &
1955               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
1956                 return 0;
1957
1958         offset = offsetof(struct ext4_group_desc, bg_checksum);
1959
1960         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1961         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1962         crc = crc16(crc, (__u8 *)gdp, offset);
1963         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1964         /* for checksum of struct ext4_group_desc do the rest...*/
1965         if ((sbi->s_es->s_feature_incompat &
1966              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1967             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1968                 crc = crc16(crc, (__u8 *)gdp + offset,
1969                             le16_to_cpu(sbi->s_es->s_desc_size) -
1970                                 offset);
1971
1972 out:
1973         return cpu_to_le16(crc);
1974 }
1975
1976 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1977                                 struct ext4_group_desc *gdp)
1978 {
1979         if (ext4_has_group_desc_csum(sb) &&
1980             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1981                                                       block_group, gdp)))
1982                 return 0;
1983
1984         return 1;
1985 }
1986
1987 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
1988                               struct ext4_group_desc *gdp)
1989 {
1990         if (!ext4_has_group_desc_csum(sb))
1991                 return;
1992         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
1993 }
1994
1995 /* Called at mount-time, super-block is locked */
1996 static int ext4_check_descriptors(struct super_block *sb,
1997                                   ext4_group_t *first_not_zeroed)
1998 {
1999         struct ext4_sb_info *sbi = EXT4_SB(sb);
2000         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2001         ext4_fsblk_t last_block;
2002         ext4_fsblk_t block_bitmap;
2003         ext4_fsblk_t inode_bitmap;
2004         ext4_fsblk_t inode_table;
2005         int flexbg_flag = 0;
2006         ext4_group_t i, grp = sbi->s_groups_count;
2007
2008         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2009                 flexbg_flag = 1;
2010
2011         ext4_debug("Checking group descriptors");
2012
2013         for (i = 0; i < sbi->s_groups_count; i++) {
2014                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2015
2016                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2017                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2018                 else
2019                         last_block = first_block +
2020                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2021
2022                 if ((grp == sbi->s_groups_count) &&
2023                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2024                         grp = i;
2025
2026                 block_bitmap = ext4_block_bitmap(sb, gdp);
2027                 if (block_bitmap < first_block || block_bitmap > last_block) {
2028                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2029                                "Block bitmap for group %u not in group "
2030                                "(block %llu)!", i, block_bitmap);
2031                         return 0;
2032                 }
2033                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2034                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2035                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2036                                "Inode bitmap for group %u not in group "
2037                                "(block %llu)!", i, inode_bitmap);
2038                         return 0;
2039                 }
2040                 inode_table = ext4_inode_table(sb, gdp);
2041                 if (inode_table < first_block ||
2042                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2043                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2044                                "Inode table for group %u not in group "
2045                                "(block %llu)!", i, inode_table);
2046                         return 0;
2047                 }
2048                 ext4_lock_group(sb, i);
2049                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2050                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2051                                  "Checksum for group %u failed (%u!=%u)",
2052                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2053                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2054                         if (!(sb->s_flags & MS_RDONLY)) {
2055                                 ext4_unlock_group(sb, i);
2056                                 return 0;
2057                         }
2058                 }
2059                 ext4_unlock_group(sb, i);
2060                 if (!flexbg_flag)
2061                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2062         }
2063         if (NULL != first_not_zeroed)
2064                 *first_not_zeroed = grp;
2065
2066         ext4_free_blocks_count_set(sbi->s_es,
2067                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2068         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2069         return 1;
2070 }
2071
2072 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2073  * the superblock) which were deleted from all directories, but held open by
2074  * a process at the time of a crash.  We walk the list and try to delete these
2075  * inodes at recovery time (only with a read-write filesystem).
2076  *
2077  * In order to keep the orphan inode chain consistent during traversal (in
2078  * case of crash during recovery), we link each inode into the superblock
2079  * orphan list_head and handle it the same way as an inode deletion during
2080  * normal operation (which journals the operations for us).
2081  *
2082  * We only do an iget() and an iput() on each inode, which is very safe if we
2083  * accidentally point at an in-use or already deleted inode.  The worst that
2084  * can happen in this case is that we get a "bit already cleared" message from
2085  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2086  * e2fsck was run on this filesystem, and it must have already done the orphan
2087  * inode cleanup for us, so we can safely abort without any further action.
2088  */
2089 static void ext4_orphan_cleanup(struct super_block *sb,
2090                                 struct ext4_super_block *es)
2091 {
2092         unsigned int s_flags = sb->s_flags;
2093         int nr_orphans = 0, nr_truncates = 0;
2094 #ifdef CONFIG_QUOTA
2095         int i;
2096 #endif
2097         if (!es->s_last_orphan) {
2098                 jbd_debug(4, "no orphan inodes to clean up\n");
2099                 return;
2100         }
2101
2102         if (bdev_read_only(sb->s_bdev)) {
2103                 ext4_msg(sb, KERN_ERR, "write access "
2104                         "unavailable, skipping orphan cleanup");
2105                 return;
2106         }
2107
2108         /* Check if feature set would not allow a r/w mount */
2109         if (!ext4_feature_set_ok(sb, 0)) {
2110                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2111                          "unknown ROCOMPAT features");
2112                 return;
2113         }
2114
2115         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2116                 /* don't clear list on RO mount w/ errors */
2117                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2118                         jbd_debug(1, "Errors on filesystem, "
2119                                   "clearing orphan list.\n");
2120                         es->s_last_orphan = 0;
2121                 }
2122                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2123                 return;
2124         }
2125
2126         if (s_flags & MS_RDONLY) {
2127                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2128                 sb->s_flags &= ~MS_RDONLY;
2129         }
2130 #ifdef CONFIG_QUOTA
2131         /* Needed for iput() to work correctly and not trash data */
2132         sb->s_flags |= MS_ACTIVE;
2133         /* Turn on quotas so that they are updated correctly */
2134         for (i = 0; i < MAXQUOTAS; i++) {
2135                 if (EXT4_SB(sb)->s_qf_names[i]) {
2136                         int ret = ext4_quota_on_mount(sb, i);
2137                         if (ret < 0)
2138                                 ext4_msg(sb, KERN_ERR,
2139                                         "Cannot turn on journaled "
2140                                         "quota: error %d", ret);
2141                 }
2142         }
2143 #endif
2144
2145         while (es->s_last_orphan) {
2146                 struct inode *inode;
2147
2148                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2149                 if (IS_ERR(inode)) {
2150                         es->s_last_orphan = 0;
2151                         break;
2152                 }
2153
2154                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2155                 dquot_initialize(inode);
2156                 if (inode->i_nlink) {
2157                         ext4_msg(sb, KERN_DEBUG,
2158                                 "%s: truncating inode %lu to %lld bytes",
2159                                 __func__, inode->i_ino, inode->i_size);
2160                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2161                                   inode->i_ino, inode->i_size);
2162                         mutex_lock(&inode->i_mutex);
2163                         ext4_truncate(inode);
2164                         mutex_unlock(&inode->i_mutex);
2165                         nr_truncates++;
2166                 } else {
2167                         ext4_msg(sb, KERN_DEBUG,
2168                                 "%s: deleting unreferenced inode %lu",
2169                                 __func__, inode->i_ino);
2170                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2171                                   inode->i_ino);
2172                         nr_orphans++;
2173                 }
2174                 iput(inode);  /* The delete magic happens here! */
2175         }
2176
2177 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2178
2179         if (nr_orphans)
2180                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2181                        PLURAL(nr_orphans));
2182         if (nr_truncates)
2183                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2184                        PLURAL(nr_truncates));
2185 #ifdef CONFIG_QUOTA
2186         /* Turn quotas off */
2187         for (i = 0; i < MAXQUOTAS; i++) {
2188                 if (sb_dqopt(sb)->files[i])
2189                         dquot_quota_off(sb, i);
2190         }
2191 #endif
2192         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2193 }
2194
2195 /*
2196  * Maximal extent format file size.
2197  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2198  * extent format containers, within a sector_t, and within i_blocks
2199  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2200  * so that won't be a limiting factor.
2201  *
2202  * However there is other limiting factor. We do store extents in the form
2203  * of starting block and length, hence the resulting length of the extent
2204  * covering maximum file size must fit into on-disk format containers as
2205  * well. Given that length is always by 1 unit bigger than max unit (because
2206  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2207  *
2208  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2209  */
2210 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2211 {
2212         loff_t res;
2213         loff_t upper_limit = MAX_LFS_FILESIZE;
2214
2215         /* small i_blocks in vfs inode? */
2216         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2217                 /*
2218                  * CONFIG_LBDAF is not enabled implies the inode
2219                  * i_block represent total blocks in 512 bytes
2220                  * 32 == size of vfs inode i_blocks * 8
2221                  */
2222                 upper_limit = (1LL << 32) - 1;
2223
2224                 /* total blocks in file system block size */
2225                 upper_limit >>= (blkbits - 9);
2226                 upper_limit <<= blkbits;
2227         }
2228
2229         /*
2230          * 32-bit extent-start container, ee_block. We lower the maxbytes
2231          * by one fs block, so ee_len can cover the extent of maximum file
2232          * size
2233          */
2234         res = (1LL << 32) - 1;
2235         res <<= blkbits;
2236
2237         /* Sanity check against vm- & vfs- imposed limits */
2238         if (res > upper_limit)
2239                 res = upper_limit;
2240
2241         return res;
2242 }
2243
2244 /*
2245  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2246  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2247  * We need to be 1 filesystem block less than the 2^48 sector limit.
2248  */
2249 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2250 {
2251         loff_t res = EXT4_NDIR_BLOCKS;
2252         int meta_blocks;
2253         loff_t upper_limit;
2254         /* This is calculated to be the largest file size for a dense, block
2255          * mapped file such that the file's total number of 512-byte sectors,
2256          * including data and all indirect blocks, does not exceed (2^48 - 1).
2257          *
2258          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2259          * number of 512-byte sectors of the file.
2260          */
2261
2262         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2263                 /*
2264                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2265                  * the inode i_block field represents total file blocks in
2266                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2267                  */
2268                 upper_limit = (1LL << 32) - 1;
2269
2270                 /* total blocks in file system block size */
2271                 upper_limit >>= (bits - 9);
2272
2273         } else {
2274                 /*
2275                  * We use 48 bit ext4_inode i_blocks
2276                  * With EXT4_HUGE_FILE_FL set the i_blocks
2277                  * represent total number of blocks in
2278                  * file system block size
2279                  */
2280                 upper_limit = (1LL << 48) - 1;
2281
2282         }
2283
2284         /* indirect blocks */
2285         meta_blocks = 1;
2286         /* double indirect blocks */
2287         meta_blocks += 1 + (1LL << (bits-2));
2288         /* tripple indirect blocks */
2289         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2290
2291         upper_limit -= meta_blocks;
2292         upper_limit <<= bits;
2293
2294         res += 1LL << (bits-2);
2295         res += 1LL << (2*(bits-2));
2296         res += 1LL << (3*(bits-2));
2297         res <<= bits;
2298         if (res > upper_limit)
2299                 res = upper_limit;
2300
2301         if (res > MAX_LFS_FILESIZE)
2302                 res = MAX_LFS_FILESIZE;
2303
2304         return res;
2305 }
2306
2307 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2308                                    ext4_fsblk_t logical_sb_block, int nr)
2309 {
2310         struct ext4_sb_info *sbi = EXT4_SB(sb);
2311         ext4_group_t bg, first_meta_bg;
2312         int has_super = 0;
2313
2314         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2315
2316         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2317             nr < first_meta_bg)
2318                 return logical_sb_block + nr + 1;
2319         bg = sbi->s_desc_per_block * nr;
2320         if (ext4_bg_has_super(sb, bg))
2321                 has_super = 1;
2322
2323         return (has_super + ext4_group_first_block_no(sb, bg));
2324 }
2325
2326 /**
2327  * ext4_get_stripe_size: Get the stripe size.
2328  * @sbi: In memory super block info
2329  *
2330  * If we have specified it via mount option, then
2331  * use the mount option value. If the value specified at mount time is
2332  * greater than the blocks per group use the super block value.
2333  * If the super block value is greater than blocks per group return 0.
2334  * Allocator needs it be less than blocks per group.
2335  *
2336  */
2337 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2338 {
2339         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2340         unsigned long stripe_width =
2341                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2342         int ret;
2343
2344         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2345                 ret = sbi->s_stripe;
2346         else if (stripe_width <= sbi->s_blocks_per_group)
2347                 ret = stripe_width;
2348         else if (stride <= sbi->s_blocks_per_group)
2349                 ret = stride;
2350         else
2351                 ret = 0;
2352
2353         /*
2354          * If the stripe width is 1, this makes no sense and
2355          * we set it to 0 to turn off stripe handling code.
2356          */
2357         if (ret <= 1)
2358                 ret = 0;
2359
2360         return ret;
2361 }
2362
2363 /* sysfs supprt */
2364
2365 struct ext4_attr {
2366         struct attribute attr;
2367         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2368         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2369                          const char *, size_t);
2370         int offset;
2371 };
2372
2373 static int parse_strtoull(const char *buf,
2374                 unsigned long long max, unsigned long long *value)
2375 {
2376         int ret;
2377
2378         ret = kstrtoull(skip_spaces(buf), 0, value);
2379         if (!ret && *value > max)
2380                 ret = -EINVAL;
2381         return ret;
2382 }
2383
2384 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2385                                               struct ext4_sb_info *sbi,
2386                                               char *buf)
2387 {
2388         return snprintf(buf, PAGE_SIZE, "%llu\n",
2389                 (s64) EXT4_C2B(sbi,
2390                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2391 }
2392
2393 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2394                                          struct ext4_sb_info *sbi, char *buf)
2395 {
2396         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2397
2398         if (!sb->s_bdev->bd_part)
2399                 return snprintf(buf, PAGE_SIZE, "0\n");
2400         return snprintf(buf, PAGE_SIZE, "%lu\n",
2401                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2402                          sbi->s_sectors_written_start) >> 1);
2403 }
2404
2405 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2406                                           struct ext4_sb_info *sbi, char *buf)
2407 {
2408         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2409
2410         if (!sb->s_bdev->bd_part)
2411                 return snprintf(buf, PAGE_SIZE, "0\n");
2412         return snprintf(buf, PAGE_SIZE, "%llu\n",
2413                         (unsigned long long)(sbi->s_kbytes_written +
2414                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2415                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2416 }
2417
2418 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2419                                           struct ext4_sb_info *sbi,
2420                                           const char *buf, size_t count)
2421 {
2422         unsigned long t;
2423         int ret;
2424
2425         ret = kstrtoul(skip_spaces(buf), 0, &t);
2426         if (ret)
2427                 return ret;
2428
2429         if (t && (!is_power_of_2(t) || t > 0x40000000))
2430                 return -EINVAL;
2431
2432         sbi->s_inode_readahead_blks = t;
2433         return count;
2434 }
2435
2436 static ssize_t sbi_ui_show(struct ext4_attr *a,
2437                            struct ext4_sb_info *sbi, char *buf)
2438 {
2439         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2440
2441         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2442 }
2443
2444 static ssize_t sbi_ui_store(struct ext4_attr *a,
2445                             struct ext4_sb_info *sbi,
2446                             const char *buf, size_t count)
2447 {
2448         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2449         unsigned long t;
2450         int ret;
2451
2452         ret = kstrtoul(skip_spaces(buf), 0, &t);
2453         if (ret)
2454                 return ret;
2455         *ui = t;
2456         return count;
2457 }
2458
2459 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2460                                   struct ext4_sb_info *sbi, char *buf)
2461 {
2462         return snprintf(buf, PAGE_SIZE, "%llu\n",
2463                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2464 }
2465
2466 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2467                                    struct ext4_sb_info *sbi,
2468                                    const char *buf, size_t count)
2469 {
2470         unsigned long long val;
2471         int ret;
2472
2473         if (parse_strtoull(buf, -1ULL, &val))
2474                 return -EINVAL;
2475         ret = ext4_reserve_clusters(sbi, val);
2476
2477         return ret ? ret : count;
2478 }
2479
2480 static ssize_t trigger_test_error(struct ext4_attr *a,
2481                                   struct ext4_sb_info *sbi,
2482                                   const char *buf, size_t count)
2483 {
2484         int len = count;
2485
2486         if (!capable(CAP_SYS_ADMIN))
2487                 return -EPERM;
2488
2489         if (len && buf[len-1] == '\n')
2490                 len--;
2491
2492         if (len)
2493                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2494         return count;
2495 }
2496
2497 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2498 static struct ext4_attr ext4_attr_##_name = {                   \
2499         .attr = {.name = __stringify(_name), .mode = _mode },   \
2500         .show   = _show,                                        \
2501         .store  = _store,                                       \
2502         .offset = offsetof(struct ext4_sb_info, _elname),       \
2503 }
2504 #define EXT4_ATTR(name, mode, show, store) \
2505 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2506
2507 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2508 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2509 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2510 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2511         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2512 #define ATTR_LIST(name) &ext4_attr_##name.attr
2513
2514 EXT4_RO_ATTR(delayed_allocation_blocks);
2515 EXT4_RO_ATTR(session_write_kbytes);
2516 EXT4_RO_ATTR(lifetime_write_kbytes);
2517 EXT4_RW_ATTR(reserved_clusters);
2518 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2519                  inode_readahead_blks_store, s_inode_readahead_blks);
2520 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2521 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2522 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2523 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2524 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2525 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2526 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2527 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2528 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2529 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2530
2531 static struct attribute *ext4_attrs[] = {
2532         ATTR_LIST(delayed_allocation_blocks),
2533         ATTR_LIST(session_write_kbytes),
2534         ATTR_LIST(lifetime_write_kbytes),
2535         ATTR_LIST(reserved_clusters),
2536         ATTR_LIST(inode_readahead_blks),
2537         ATTR_LIST(inode_goal),
2538         ATTR_LIST(mb_stats),
2539         ATTR_LIST(mb_max_to_scan),
2540         ATTR_LIST(mb_min_to_scan),
2541         ATTR_LIST(mb_order2_req),
2542         ATTR_LIST(mb_stream_req),
2543         ATTR_LIST(mb_group_prealloc),
2544         ATTR_LIST(max_writeback_mb_bump),
2545         ATTR_LIST(extent_max_zeroout_kb),
2546         ATTR_LIST(trigger_fs_error),
2547         NULL,
2548 };
2549
2550 /* Features this copy of ext4 supports */
2551 EXT4_INFO_ATTR(lazy_itable_init);
2552 EXT4_INFO_ATTR(batched_discard);
2553 EXT4_INFO_ATTR(meta_bg_resize);
2554
2555 static struct attribute *ext4_feat_attrs[] = {
2556         ATTR_LIST(lazy_itable_init),
2557         ATTR_LIST(batched_discard),
2558         ATTR_LIST(meta_bg_resize),
2559         NULL,
2560 };
2561
2562 static ssize_t ext4_attr_show(struct kobject *kobj,
2563                               struct attribute *attr, char *buf)
2564 {
2565         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2566                                                 s_kobj);
2567         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2568
2569         return a->show ? a->show(a, sbi, buf) : 0;
2570 }
2571
2572 static ssize_t ext4_attr_store(struct kobject *kobj,
2573                                struct attribute *attr,
2574                                const char *buf, size_t len)
2575 {
2576         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2577                                                 s_kobj);
2578         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2579
2580         return a->store ? a->store(a, sbi, buf, len) : 0;
2581 }
2582
2583 static void ext4_sb_release(struct kobject *kobj)
2584 {
2585         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2586                                                 s_kobj);
2587         complete(&sbi->s_kobj_unregister);
2588 }
2589
2590 static const struct sysfs_ops ext4_attr_ops = {
2591         .show   = ext4_attr_show,
2592         .store  = ext4_attr_store,
2593 };
2594
2595 static struct kobj_type ext4_ktype = {
2596         .default_attrs  = ext4_attrs,
2597         .sysfs_ops      = &ext4_attr_ops,
2598         .release        = ext4_sb_release,
2599 };
2600
2601 static void ext4_feat_release(struct kobject *kobj)
2602 {
2603         complete(&ext4_feat->f_kobj_unregister);
2604 }
2605
2606 static struct kobj_type ext4_feat_ktype = {
2607         .default_attrs  = ext4_feat_attrs,
2608         .sysfs_ops      = &ext4_attr_ops,
2609         .release        = ext4_feat_release,
2610 };
2611
2612 /*
2613  * Check whether this filesystem can be mounted based on
2614  * the features present and the RDONLY/RDWR mount requested.
2615  * Returns 1 if this filesystem can be mounted as requested,
2616  * 0 if it cannot be.
2617  */
2618 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2619 {
2620         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2621                 ext4_msg(sb, KERN_ERR,
2622                         "Couldn't mount because of "
2623                         "unsupported optional features (%x)",
2624                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2625                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2626                 return 0;
2627         }
2628
2629         if (readonly)
2630                 return 1;
2631
2632         /* Check that feature set is OK for a read-write mount */
2633         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2634                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2635                          "unsupported optional features (%x)",
2636                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2637                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2638                 return 0;
2639         }
2640         /*
2641          * Large file size enabled file system can only be mounted
2642          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2643          */
2644         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2645                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2646                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2647                                  "cannot be mounted RDWR without "
2648                                  "CONFIG_LBDAF");
2649                         return 0;
2650                 }
2651         }
2652         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2653             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2654                 ext4_msg(sb, KERN_ERR,
2655                          "Can't support bigalloc feature without "
2656                          "extents feature\n");
2657                 return 0;
2658         }
2659
2660 #ifndef CONFIG_QUOTA
2661         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2662             !readonly) {
2663                 ext4_msg(sb, KERN_ERR,
2664                          "Filesystem with quota feature cannot be mounted RDWR "
2665                          "without CONFIG_QUOTA");
2666                 return 0;
2667         }
2668 #endif  /* CONFIG_QUOTA */
2669         return 1;
2670 }
2671
2672 /*
2673  * This function is called once a day if we have errors logged
2674  * on the file system
2675  */
2676 static void print_daily_error_info(unsigned long arg)
2677 {
2678         struct super_block *sb = (struct super_block *) arg;
2679         struct ext4_sb_info *sbi;
2680         struct ext4_super_block *es;
2681
2682         sbi = EXT4_SB(sb);
2683         es = sbi->s_es;
2684
2685         if (es->s_error_count)
2686                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2687                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2688                          le32_to_cpu(es->s_error_count));
2689         if (es->s_first_error_time) {
2690                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2691                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2692                        (int) sizeof(es->s_first_error_func),
2693                        es->s_first_error_func,
2694                        le32_to_cpu(es->s_first_error_line));
2695                 if (es->s_first_error_ino)
2696                         printk(": inode %u",
2697                                le32_to_cpu(es->s_first_error_ino));
2698                 if (es->s_first_error_block)
2699                         printk(": block %llu", (unsigned long long)
2700                                le64_to_cpu(es->s_first_error_block));
2701                 printk("\n");
2702         }
2703         if (es->s_last_error_time) {
2704                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2705                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2706                        (int) sizeof(es->s_last_error_func),
2707                        es->s_last_error_func,
2708                        le32_to_cpu(es->s_last_error_line));
2709                 if (es->s_last_error_ino)
2710                         printk(": inode %u",
2711                                le32_to_cpu(es->s_last_error_ino));
2712                 if (es->s_last_error_block)
2713                         printk(": block %llu", (unsigned long long)
2714                                le64_to_cpu(es->s_last_error_block));
2715                 printk("\n");
2716         }
2717         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2718 }
2719
2720 /* Find next suitable group and run ext4_init_inode_table */
2721 static int ext4_run_li_request(struct ext4_li_request *elr)
2722 {
2723         struct ext4_group_desc *gdp = NULL;
2724         ext4_group_t group, ngroups;
2725         struct super_block *sb;
2726         unsigned long timeout = 0;
2727         int ret = 0;
2728
2729         sb = elr->lr_super;
2730         ngroups = EXT4_SB(sb)->s_groups_count;
2731
2732         sb_start_write(sb);
2733         for (group = elr->lr_next_group; group < ngroups; group++) {
2734                 gdp = ext4_get_group_desc(sb, group, NULL);
2735                 if (!gdp) {
2736                         ret = 1;
2737                         break;
2738                 }
2739
2740                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2741                         break;
2742         }
2743
2744         if (group >= ngroups)
2745                 ret = 1;
2746
2747         if (!ret) {
2748                 timeout = jiffies;
2749                 ret = ext4_init_inode_table(sb, group,
2750                                             elr->lr_timeout ? 0 : 1);
2751                 if (elr->lr_timeout == 0) {
2752                         timeout = (jiffies - timeout) *
2753                                   elr->lr_sbi->s_li_wait_mult;
2754                         elr->lr_timeout = timeout;
2755                 }
2756                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2757                 elr->lr_next_group = group + 1;
2758         }
2759         sb_end_write(sb);
2760
2761         return ret;
2762 }
2763
2764 /*
2765  * Remove lr_request from the list_request and free the
2766  * request structure. Should be called with li_list_mtx held
2767  */
2768 static void ext4_remove_li_request(struct ext4_li_request *elr)
2769 {
2770         struct ext4_sb_info *sbi;
2771
2772         if (!elr)
2773                 return;
2774
2775         sbi = elr->lr_sbi;
2776
2777         list_del(&elr->lr_request);
2778         sbi->s_li_request = NULL;
2779         kfree(elr);
2780 }
2781
2782 static void ext4_unregister_li_request(struct super_block *sb)
2783 {
2784         mutex_lock(&ext4_li_mtx);
2785         if (!ext4_li_info) {
2786                 mutex_unlock(&ext4_li_mtx);
2787                 return;
2788         }
2789
2790         mutex_lock(&ext4_li_info->li_list_mtx);
2791         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2792         mutex_unlock(&ext4_li_info->li_list_mtx);
2793         mutex_unlock(&ext4_li_mtx);
2794 }
2795
2796 static struct task_struct *ext4_lazyinit_task;
2797
2798 /*
2799  * This is the function where ext4lazyinit thread lives. It walks
2800  * through the request list searching for next scheduled filesystem.
2801  * When such a fs is found, run the lazy initialization request
2802  * (ext4_rn_li_request) and keep track of the time spend in this
2803  * function. Based on that time we compute next schedule time of
2804  * the request. When walking through the list is complete, compute
2805  * next waking time and put itself into sleep.
2806  */
2807 static int ext4_lazyinit_thread(void *arg)
2808 {
2809         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2810         struct list_head *pos, *n;
2811         struct ext4_li_request *elr;
2812         unsigned long next_wakeup, cur;
2813
2814         BUG_ON(NULL == eli);
2815
2816 cont_thread:
2817         while (true) {
2818                 next_wakeup = MAX_JIFFY_OFFSET;
2819
2820                 mutex_lock(&eli->li_list_mtx);
2821                 if (list_empty(&eli->li_request_list)) {
2822                         mutex_unlock(&eli->li_list_mtx);
2823                         goto exit_thread;
2824                 }
2825
2826                 list_for_each_safe(pos, n, &eli->li_request_list) {
2827                         elr = list_entry(pos, struct ext4_li_request,
2828                                          lr_request);
2829
2830                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2831                                 if (ext4_run_li_request(elr) != 0) {
2832                                         /* error, remove the lazy_init job */
2833                                         ext4_remove_li_request(elr);
2834                                         continue;
2835                                 }
2836                         }
2837
2838                         if (time_before(elr->lr_next_sched, next_wakeup))
2839                                 next_wakeup = elr->lr_next_sched;
2840                 }
2841                 mutex_unlock(&eli->li_list_mtx);
2842
2843                 try_to_freeze();
2844
2845                 cur = jiffies;
2846                 if ((time_after_eq(cur, next_wakeup)) ||
2847                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2848                         cond_resched();
2849                         continue;
2850                 }
2851
2852                 schedule_timeout_interruptible(next_wakeup - cur);
2853
2854                 if (kthread_should_stop()) {
2855                         ext4_clear_request_list();
2856                         goto exit_thread;
2857                 }
2858         }
2859
2860 exit_thread:
2861         /*
2862          * It looks like the request list is empty, but we need
2863          * to check it under the li_list_mtx lock, to prevent any
2864          * additions into it, and of course we should lock ext4_li_mtx
2865          * to atomically free the list and ext4_li_info, because at
2866          * this point another ext4 filesystem could be registering
2867          * new one.
2868          */
2869         mutex_lock(&ext4_li_mtx);
2870         mutex_lock(&eli->li_list_mtx);
2871         if (!list_empty(&eli->li_request_list)) {
2872                 mutex_unlock(&eli->li_list_mtx);
2873                 mutex_unlock(&ext4_li_mtx);
2874                 goto cont_thread;
2875         }
2876         mutex_unlock(&eli->li_list_mtx);
2877         kfree(ext4_li_info);
2878         ext4_li_info = NULL;
2879         mutex_unlock(&ext4_li_mtx);
2880
2881         return 0;
2882 }
2883
2884 static void ext4_clear_request_list(void)
2885 {
2886         struct list_head *pos, *n;
2887         struct ext4_li_request *elr;
2888
2889         mutex_lock(&ext4_li_info->li_list_mtx);
2890         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2891                 elr = list_entry(pos, struct ext4_li_request,
2892                                  lr_request);
2893                 ext4_remove_li_request(elr);
2894         }
2895         mutex_unlock(&ext4_li_info->li_list_mtx);
2896 }
2897
2898 static int ext4_run_lazyinit_thread(void)
2899 {
2900         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2901                                          ext4_li_info, "ext4lazyinit");
2902         if (IS_ERR(ext4_lazyinit_task)) {
2903                 int err = PTR_ERR(ext4_lazyinit_task);
2904                 ext4_clear_request_list();
2905                 kfree(ext4_li_info);
2906                 ext4_li_info = NULL;
2907                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2908                                  "initialization thread\n",
2909                                  err);
2910                 return err;
2911         }
2912         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2913         return 0;
2914 }
2915
2916 /*
2917  * Check whether it make sense to run itable init. thread or not.
2918  * If there is at least one uninitialized inode table, return
2919  * corresponding group number, else the loop goes through all
2920  * groups and return total number of groups.
2921  */
2922 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2923 {
2924         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2925         struct ext4_group_desc *gdp = NULL;
2926
2927         for (group = 0; group < ngroups; group++) {
2928                 gdp = ext4_get_group_desc(sb, group, NULL);
2929                 if (!gdp)
2930                         continue;
2931
2932                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2933                         break;
2934         }
2935
2936         return group;
2937 }
2938
2939 static int ext4_li_info_new(void)
2940 {
2941         struct ext4_lazy_init *eli = NULL;
2942
2943         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2944         if (!eli)
2945                 return -ENOMEM;
2946
2947         INIT_LIST_HEAD(&eli->li_request_list);
2948         mutex_init(&eli->li_list_mtx);
2949
2950         eli->li_state |= EXT4_LAZYINIT_QUIT;
2951
2952         ext4_li_info = eli;
2953
2954         return 0;
2955 }
2956
2957 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2958                                             ext4_group_t start)
2959 {
2960         struct ext4_sb_info *sbi = EXT4_SB(sb);
2961         struct ext4_li_request *elr;
2962         unsigned long rnd;
2963
2964         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2965         if (!elr)
2966                 return NULL;
2967
2968         elr->lr_super = sb;
2969         elr->lr_sbi = sbi;
2970         elr->lr_next_group = start;
2971
2972         /*
2973          * Randomize first schedule time of the request to
2974          * spread the inode table initialization requests
2975          * better.
2976          */
2977         get_random_bytes(&rnd, sizeof(rnd));
2978         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2979                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2980
2981         return elr;
2982 }
2983
2984 int ext4_register_li_request(struct super_block *sb,
2985                              ext4_group_t first_not_zeroed)
2986 {
2987         struct ext4_sb_info *sbi = EXT4_SB(sb);
2988         struct ext4_li_request *elr = NULL;
2989         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2990         int ret = 0;
2991
2992         mutex_lock(&ext4_li_mtx);
2993         if (sbi->s_li_request != NULL) {
2994                 /*
2995                  * Reset timeout so it can be computed again, because
2996                  * s_li_wait_mult might have changed.
2997                  */
2998                 sbi->s_li_request->lr_timeout = 0;
2999                 goto out;
3000         }
3001
3002         if (first_not_zeroed == ngroups ||
3003             (sb->s_flags & MS_RDONLY) ||
3004             !test_opt(sb, INIT_INODE_TABLE))
3005                 goto out;
3006
3007         elr = ext4_li_request_new(sb, first_not_zeroed);
3008         if (!elr) {
3009                 ret = -ENOMEM;
3010                 goto out;
3011         }
3012
3013         if (NULL == ext4_li_info) {
3014                 ret = ext4_li_info_new();
3015                 if (ret)
3016                         goto out;
3017         }
3018
3019         mutex_lock(&ext4_li_info->li_list_mtx);
3020         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3021         mutex_unlock(&ext4_li_info->li_list_mtx);
3022
3023         sbi->s_li_request = elr;
3024         /*
3025          * set elr to NULL here since it has been inserted to
3026          * the request_list and the removal and free of it is
3027          * handled by ext4_clear_request_list from now on.
3028          */
3029         elr = NULL;
3030
3031         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3032                 ret = ext4_run_lazyinit_thread();
3033                 if (ret)
3034                         goto out;
3035         }
3036 out:
3037         mutex_unlock(&ext4_li_mtx);
3038         if (ret)
3039                 kfree(elr);
3040         return ret;
3041 }
3042
3043 /*
3044  * We do not need to lock anything since this is called on
3045  * module unload.
3046  */
3047 static void ext4_destroy_lazyinit_thread(void)
3048 {
3049         /*
3050          * If thread exited earlier
3051          * there's nothing to be done.
3052          */
3053         if (!ext4_li_info || !ext4_lazyinit_task)
3054                 return;
3055
3056         kthread_stop(ext4_lazyinit_task);
3057 }
3058
3059 static int set_journal_csum_feature_set(struct super_block *sb)
3060 {
3061         int ret = 1;
3062         int compat, incompat;
3063         struct ext4_sb_info *sbi = EXT4_SB(sb);
3064
3065         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3066                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3067                 /* journal checksum v2 */
3068                 compat = 0;
3069                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3070         } else {
3071                 /* journal checksum v1 */
3072                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3073                 incompat = 0;
3074         }
3075
3076         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3077                 ret = jbd2_journal_set_features(sbi->s_journal,
3078                                 compat, 0,
3079                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3080                                 incompat);
3081         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3082                 ret = jbd2_journal_set_features(sbi->s_journal,
3083                                 compat, 0,
3084                                 incompat);
3085                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3086                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3087         } else {
3088                 jbd2_journal_clear_features(sbi->s_journal,
3089                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3090                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3091                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3092         }
3093
3094         return ret;
3095 }
3096
3097 /*
3098  * Note: calculating the overhead so we can be compatible with
3099  * historical BSD practice is quite difficult in the face of
3100  * clusters/bigalloc.  This is because multiple metadata blocks from
3101  * different block group can end up in the same allocation cluster.
3102  * Calculating the exact overhead in the face of clustered allocation
3103  * requires either O(all block bitmaps) in memory or O(number of block
3104  * groups**2) in time.  We will still calculate the superblock for
3105  * older file systems --- and if we come across with a bigalloc file
3106  * system with zero in s_overhead_clusters the estimate will be close to
3107  * correct especially for very large cluster sizes --- but for newer
3108  * file systems, it's better to calculate this figure once at mkfs
3109  * time, and store it in the superblock.  If the superblock value is
3110  * present (even for non-bigalloc file systems), we will use it.
3111  */
3112 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3113                           char *buf)
3114 {
3115         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3116         struct ext4_group_desc  *gdp;
3117         ext4_fsblk_t            first_block, last_block, b;
3118         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3119         int                     s, j, count = 0;
3120
3121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3122                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3123                         sbi->s_itb_per_group + 2);
3124
3125         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3126                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3127         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3128         for (i = 0; i < ngroups; i++) {
3129                 gdp = ext4_get_group_desc(sb, i, NULL);
3130                 b = ext4_block_bitmap(sb, gdp);
3131                 if (b >= first_block && b <= last_block) {
3132                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3133                         count++;
3134                 }
3135                 b = ext4_inode_bitmap(sb, gdp);
3136                 if (b >= first_block && b <= last_block) {
3137                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3138                         count++;
3139                 }
3140                 b = ext4_inode_table(sb, gdp);
3141                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3142                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3143                                 int c = EXT4_B2C(sbi, b - first_block);
3144                                 ext4_set_bit(c, buf);
3145                                 count++;
3146                         }
3147                 if (i != grp)
3148                         continue;
3149                 s = 0;
3150                 if (ext4_bg_has_super(sb, grp)) {
3151                         ext4_set_bit(s++, buf);
3152                         count++;
3153                 }
3154                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3155                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3156                         count++;
3157                 }
3158         }
3159         if (!count)
3160                 return 0;
3161         return EXT4_CLUSTERS_PER_GROUP(sb) -
3162                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3163 }
3164
3165 /*
3166  * Compute the overhead and stash it in sbi->s_overhead
3167  */
3168 int ext4_calculate_overhead(struct super_block *sb)
3169 {
3170         struct ext4_sb_info *sbi = EXT4_SB(sb);
3171         struct ext4_super_block *es = sbi->s_es;
3172         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3173         ext4_fsblk_t overhead = 0;
3174         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3175
3176         if (!buf)
3177                 return -ENOMEM;
3178
3179         /*
3180          * Compute the overhead (FS structures).  This is constant
3181          * for a given filesystem unless the number of block groups
3182          * changes so we cache the previous value until it does.
3183          */
3184
3185         /*
3186          * All of the blocks before first_data_block are overhead
3187          */
3188         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3189
3190         /*
3191          * Add the overhead found in each block group
3192          */
3193         for (i = 0; i < ngroups; i++) {
3194                 int blks;
3195
3196                 blks = count_overhead(sb, i, buf);
3197                 overhead += blks;
3198                 if (blks)
3199                         memset(buf, 0, PAGE_SIZE);
3200                 cond_resched();
3201         }
3202         /* Add the journal blocks as well */
3203         if (sbi->s_journal)
3204                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3205
3206         sbi->s_overhead = overhead;
3207         smp_wmb();
3208         free_page((unsigned long) buf);
3209         return 0;
3210 }
3211
3212
3213 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3214 {
3215         ext4_fsblk_t resv_clusters;
3216
3217         /*
3218          * There's no need to reserve anything when we aren't using extents.
3219          * The space estimates are exact, there are no unwritten extents,
3220          * hole punching doesn't need new metadata... This is needed especially
3221          * to keep ext2/3 backward compatibility.
3222          */
3223         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3224                 return 0;
3225         /*
3226          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3227          * This should cover the situations where we can not afford to run
3228          * out of space like for example punch hole, or converting
3229          * uninitialized extents in delalloc path. In most cases such
3230          * allocation would require 1, or 2 blocks, higher numbers are
3231          * very rare.
3232          */
3233         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3234                         EXT4_SB(sb)->s_cluster_bits;
3235
3236         do_div(resv_clusters, 50);
3237         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3238
3239         return resv_clusters;
3240 }
3241
3242
3243 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3244 {
3245         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3246                                 sbi->s_cluster_bits;
3247
3248         if (count >= clusters)
3249                 return -EINVAL;
3250
3251         atomic64_set(&sbi->s_resv_clusters, count);
3252         return 0;
3253 }
3254
3255 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3256 {
3257         char *orig_data = kstrdup(data, GFP_KERNEL);
3258         struct buffer_head *bh;
3259         struct ext4_super_block *es = NULL;
3260         struct ext4_sb_info *sbi;
3261         ext4_fsblk_t block;
3262         ext4_fsblk_t sb_block = get_sb_block(&data);
3263         ext4_fsblk_t logical_sb_block;
3264         unsigned long offset = 0;
3265         unsigned long journal_devnum = 0;
3266         unsigned long def_mount_opts;
3267         struct inode *root;
3268         char *cp;
3269         const char *descr;
3270         int ret = -ENOMEM;
3271         int blocksize, clustersize;
3272         unsigned int db_count;
3273         unsigned int i;
3274         int needs_recovery, has_huge_files, has_bigalloc;
3275         __u64 blocks_count;
3276         int err = 0;
3277         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3278         ext4_group_t first_not_zeroed;
3279
3280         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3281         if (!sbi)
3282                 goto out_free_orig;
3283
3284         sbi->s_blockgroup_lock =
3285                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3286         if (!sbi->s_blockgroup_lock) {
3287                 kfree(sbi);
3288                 goto out_free_orig;
3289         }
3290         sb->s_fs_info = sbi;
3291         sbi->s_sb = sb;
3292         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3293         sbi->s_sb_block = sb_block;
3294         if (sb->s_bdev->bd_part)
3295                 sbi->s_sectors_written_start =
3296                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3297
3298         /* Cleanup superblock name */
3299         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3300                 *cp = '!';
3301
3302         /* -EINVAL is default */
3303         ret = -EINVAL;
3304         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3305         if (!blocksize) {
3306                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3307                 goto out_fail;
3308         }
3309
3310         /*
3311          * The ext4 superblock will not be buffer aligned for other than 1kB
3312          * block sizes.  We need to calculate the offset from buffer start.
3313          */
3314         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3315                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3316                 offset = do_div(logical_sb_block, blocksize);
3317         } else {
3318                 logical_sb_block = sb_block;
3319         }
3320
3321         if (!(bh = sb_bread(sb, logical_sb_block))) {
3322                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3323                 goto out_fail;
3324         }
3325         /*
3326          * Note: s_es must be initialized as soon as possible because
3327          *       some ext4 macro-instructions depend on its value
3328          */
3329         es = (struct ext4_super_block *) (bh->b_data + offset);
3330         sbi->s_es = es;
3331         sb->s_magic = le16_to_cpu(es->s_magic);
3332         if (sb->s_magic != EXT4_SUPER_MAGIC)
3333                 goto cantfind_ext4;
3334         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3335
3336         /* Warn if metadata_csum and gdt_csum are both set. */
3337         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3338                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3339             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3340                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3341                              "redundant flags; please run fsck.");
3342
3343         /* Check for a known checksum algorithm */
3344         if (!ext4_verify_csum_type(sb, es)) {
3345                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3346                          "unknown checksum algorithm.");
3347                 silent = 1;
3348                 goto cantfind_ext4;
3349         }
3350
3351         /* Load the checksum driver */
3352         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3353                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3354                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3355                 if (IS_ERR(sbi->s_chksum_driver)) {
3356                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3357                         ret = PTR_ERR(sbi->s_chksum_driver);
3358                         sbi->s_chksum_driver = NULL;
3359                         goto failed_mount;
3360                 }
3361         }
3362
3363         /* Check superblock checksum */
3364         if (!ext4_superblock_csum_verify(sb, es)) {
3365                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3366                          "invalid superblock checksum.  Run e2fsck?");
3367                 silent = 1;
3368                 goto cantfind_ext4;
3369         }
3370
3371         /* Precompute checksum seed for all metadata */
3372         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3373                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3374                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3375                                                sizeof(es->s_uuid));
3376
3377         /* Set defaults before we parse the mount options */
3378         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3379         set_opt(sb, INIT_INODE_TABLE);
3380         if (def_mount_opts & EXT4_DEFM_DEBUG)
3381                 set_opt(sb, DEBUG);
3382         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3383                 set_opt(sb, GRPID);
3384         if (def_mount_opts & EXT4_DEFM_UID16)
3385                 set_opt(sb, NO_UID32);
3386         /* xattr user namespace & acls are now defaulted on */
3387         set_opt(sb, XATTR_USER);
3388 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3389         set_opt(sb, POSIX_ACL);
3390 #endif
3391         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3392                 set_opt(sb, JOURNAL_DATA);
3393         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3394                 set_opt(sb, ORDERED_DATA);
3395         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3396                 set_opt(sb, WRITEBACK_DATA);
3397
3398         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3399                 set_opt(sb, ERRORS_PANIC);
3400         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3401                 set_opt(sb, ERRORS_CONT);
3402         else
3403                 set_opt(sb, ERRORS_RO);
3404         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3405                 set_opt(sb, BLOCK_VALIDITY);
3406         if (def_mount_opts & EXT4_DEFM_DISCARD)
3407                 set_opt(sb, DISCARD);
3408
3409         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3410         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3411         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3412         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3413         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3414
3415         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3416                 set_opt(sb, BARRIER);
3417
3418         /*
3419          * enable delayed allocation by default
3420          * Use -o nodelalloc to turn it off
3421          */
3422         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3423             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3424                 set_opt(sb, DELALLOC);
3425
3426         /*
3427          * set default s_li_wait_mult for lazyinit, for the case there is
3428          * no mount option specified.
3429          */
3430         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3431
3432         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3433                            &journal_devnum, &journal_ioprio, 0)) {
3434                 ext4_msg(sb, KERN_WARNING,
3435                          "failed to parse options in superblock: %s",
3436                          sbi->s_es->s_mount_opts);
3437         }
3438         sbi->s_def_mount_opt = sbi->s_mount_opt;
3439         if (!parse_options((char *) data, sb, &journal_devnum,
3440                            &journal_ioprio, 0))
3441                 goto failed_mount;
3442
3443         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3444                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3445                             "with data=journal disables delayed "
3446                             "allocation and O_DIRECT support!\n");
3447                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3448                         ext4_msg(sb, KERN_ERR, "can't mount with "
3449                                  "both data=journal and delalloc");
3450                         goto failed_mount;
3451                 }
3452                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3453                         ext4_msg(sb, KERN_ERR, "can't mount with "
3454                                  "both data=journal and dioread_nolock");
3455                         goto failed_mount;
3456                 }
3457                 if (test_opt(sb, DELALLOC))
3458                         clear_opt(sb, DELALLOC);
3459         }
3460
3461         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3462                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3463
3464         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3465             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3466              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3467              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3468                 ext4_msg(sb, KERN_WARNING,
3469                        "feature flags set on rev 0 fs, "
3470                        "running e2fsck is recommended");
3471
3472         if (IS_EXT2_SB(sb)) {
3473                 if (ext2_feature_set_ok(sb))
3474                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3475                                  "using the ext4 subsystem");
3476                 else {
3477                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3478                                  "to feature incompatibilities");
3479                         goto failed_mount;
3480                 }
3481         }
3482
3483         if (IS_EXT3_SB(sb)) {
3484                 if (ext3_feature_set_ok(sb))
3485                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3486                                  "using the ext4 subsystem");
3487                 else {
3488                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3489                                  "to feature incompatibilities");
3490                         goto failed_mount;
3491                 }
3492         }
3493
3494         /*
3495          * Check feature flags regardless of the revision level, since we
3496          * previously didn't change the revision level when setting the flags,
3497          * so there is a chance incompat flags are set on a rev 0 filesystem.
3498          */
3499         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3500                 goto failed_mount;
3501
3502         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3503         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3504             blocksize > EXT4_MAX_BLOCK_SIZE) {
3505                 ext4_msg(sb, KERN_ERR,
3506                        "Unsupported filesystem blocksize %d", blocksize);
3507                 goto failed_mount;
3508         }
3509
3510         if (sb->s_blocksize != blocksize) {
3511                 /* Validate the filesystem blocksize */
3512                 if (!sb_set_blocksize(sb, blocksize)) {
3513                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3514                                         blocksize);
3515                         goto failed_mount;
3516                 }
3517
3518                 brelse(bh);
3519                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3520                 offset = do_div(logical_sb_block, blocksize);
3521                 bh = sb_bread(sb, logical_sb_block);
3522                 if (!bh) {
3523                         ext4_msg(sb, KERN_ERR,
3524                                "Can't read superblock on 2nd try");
3525                         goto failed_mount;
3526                 }
3527                 es = (struct ext4_super_block *)(bh->b_data + offset);
3528                 sbi->s_es = es;
3529                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3530                         ext4_msg(sb, KERN_ERR,
3531                                "Magic mismatch, very weird!");
3532                         goto failed_mount;
3533                 }
3534         }
3535
3536         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3537                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3538         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3539                                                       has_huge_files);
3540         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3541
3542         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3543                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3544                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3545         } else {
3546                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3547                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3548                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3549                     (!is_power_of_2(sbi->s_inode_size)) ||
3550                     (sbi->s_inode_size > blocksize)) {
3551                         ext4_msg(sb, KERN_ERR,
3552                                "unsupported inode size: %d",
3553                                sbi->s_inode_size);
3554                         goto failed_mount;
3555                 }
3556                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3557                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3558         }
3559
3560         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3561         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3562                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3563                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3564                     !is_power_of_2(sbi->s_desc_size)) {
3565                         ext4_msg(sb, KERN_ERR,
3566                                "unsupported descriptor size %lu",
3567                                sbi->s_desc_size);
3568                         goto failed_mount;
3569                 }
3570         } else
3571                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3572
3573         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3574         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3575         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3576                 goto cantfind_ext4;
3577
3578         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3579         if (sbi->s_inodes_per_block == 0)
3580                 goto cantfind_ext4;
3581         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3582                                         sbi->s_inodes_per_block;
3583         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3584         sbi->s_sbh = bh;
3585         sbi->s_mount_state = le16_to_cpu(es->s_state);
3586         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3587         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3588
3589         for (i = 0; i < 4; i++)
3590                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3591         sbi->s_def_hash_version = es->s_def_hash_version;
3592         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3593                 i = le32_to_cpu(es->s_flags);
3594                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3595                         sbi->s_hash_unsigned = 3;
3596                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3597 #ifdef __CHAR_UNSIGNED__
3598                         if (!(sb->s_flags & MS_RDONLY))
3599                                 es->s_flags |=
3600                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3601                         sbi->s_hash_unsigned = 3;
3602 #else
3603                         if (!(sb->s_flags & MS_RDONLY))
3604                                 es->s_flags |=
3605                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3606 #endif
3607                 }
3608         }
3609
3610         /* Handle clustersize */
3611         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3612         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3613                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3614         if (has_bigalloc) {
3615                 if (clustersize < blocksize) {
3616                         ext4_msg(sb, KERN_ERR,
3617                                  "cluster size (%d) smaller than "
3618                                  "block size (%d)", clustersize, blocksize);
3619                         goto failed_mount;
3620                 }
3621                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3622                         le32_to_cpu(es->s_log_block_size);
3623                 sbi->s_clusters_per_group =
3624                         le32_to_cpu(es->s_clusters_per_group);
3625                 if (sbi->s_clusters_per_group > blocksize * 8) {
3626                         ext4_msg(sb, KERN_ERR,
3627                                  "#clusters per group too big: %lu",
3628                                  sbi->s_clusters_per_group);
3629                         goto failed_mount;
3630                 }
3631                 if (sbi->s_blocks_per_group !=
3632                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3633                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3634                                  "clusters per group (%lu) inconsistent",
3635                                  sbi->s_blocks_per_group,
3636                                  sbi->s_clusters_per_group);
3637                         goto failed_mount;
3638                 }
3639         } else {
3640                 if (clustersize != blocksize) {
3641                         ext4_warning(sb, "fragment/cluster size (%d) != "
3642                                      "block size (%d)", clustersize,
3643                                      blocksize);
3644                         clustersize = blocksize;
3645                 }
3646                 if (sbi->s_blocks_per_group > blocksize * 8) {
3647                         ext4_msg(sb, KERN_ERR,
3648                                  "#blocks per group too big: %lu",
3649                                  sbi->s_blocks_per_group);
3650                         goto failed_mount;
3651                 }
3652                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3653                 sbi->s_cluster_bits = 0;
3654         }
3655         sbi->s_cluster_ratio = clustersize / blocksize;
3656
3657         if (sbi->s_inodes_per_group > blocksize * 8) {
3658                 ext4_msg(sb, KERN_ERR,
3659                        "#inodes per group too big: %lu",
3660                        sbi->s_inodes_per_group);
3661                 goto failed_mount;
3662         }
3663
3664         /* Do we have standard group size of clustersize * 8 blocks ? */
3665         if (sbi->s_blocks_per_group == clustersize << 3)
3666                 set_opt2(sb, STD_GROUP_SIZE);
3667
3668         /*
3669          * Test whether we have more sectors than will fit in sector_t,
3670          * and whether the max offset is addressable by the page cache.
3671          */
3672         err = generic_check_addressable(sb->s_blocksize_bits,
3673                                         ext4_blocks_count(es));
3674         if (err) {
3675                 ext4_msg(sb, KERN_ERR, "filesystem"
3676                          " too large to mount safely on this system");
3677                 if (sizeof(sector_t) < 8)
3678                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3679                 goto failed_mount;
3680         }
3681
3682         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3683                 goto cantfind_ext4;
3684
3685         /* check blocks count against device size */
3686         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3687         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3688                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3689                        "exceeds size of device (%llu blocks)",
3690                        ext4_blocks_count(es), blocks_count);
3691                 goto failed_mount;
3692         }
3693
3694         /*
3695          * It makes no sense for the first data block to be beyond the end
3696          * of the filesystem.
3697          */
3698         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3699                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3700                          "block %u is beyond end of filesystem (%llu)",
3701                          le32_to_cpu(es->s_first_data_block),
3702                          ext4_blocks_count(es));
3703                 goto failed_mount;
3704         }
3705         blocks_count = (ext4_blocks_count(es) -
3706                         le32_to_cpu(es->s_first_data_block) +
3707                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3708         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3709         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3710                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3711                        "(block count %llu, first data block %u, "
3712                        "blocks per group %lu)", sbi->s_groups_count,
3713                        ext4_blocks_count(es),
3714                        le32_to_cpu(es->s_first_data_block),
3715                        EXT4_BLOCKS_PER_GROUP(sb));
3716                 goto failed_mount;
3717         }
3718         sbi->s_groups_count = blocks_count;
3719         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3720                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3721         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3722                    EXT4_DESC_PER_BLOCK(sb);
3723         sbi->s_group_desc = ext4_kvmalloc(db_count *
3724                                           sizeof(struct buffer_head *),
3725                                           GFP_KERNEL);
3726         if (sbi->s_group_desc == NULL) {
3727                 ext4_msg(sb, KERN_ERR, "not enough memory");
3728                 ret = -ENOMEM;
3729                 goto failed_mount;
3730         }
3731
3732         if (ext4_proc_root)
3733                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3734
3735         if (sbi->s_proc)
3736                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3737                                  &ext4_seq_options_fops, sb);
3738
3739         bgl_lock_init(sbi->s_blockgroup_lock);
3740
3741         for (i = 0; i < db_count; i++) {
3742                 block = descriptor_loc(sb, logical_sb_block, i);
3743                 sbi->s_group_desc[i] = sb_bread(sb, block);
3744                 if (!sbi->s_group_desc[i]) {
3745                         ext4_msg(sb, KERN_ERR,
3746                                "can't read group descriptor %d", i);
3747                         db_count = i;
3748                         goto failed_mount2;
3749                 }
3750         }
3751         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3752                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3753                 goto failed_mount2;
3754         }
3755         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3756                 if (!ext4_fill_flex_info(sb)) {
3757                         ext4_msg(sb, KERN_ERR,
3758                                "unable to initialize "
3759                                "flex_bg meta info!");
3760                         goto failed_mount2;
3761                 }
3762
3763         sbi->s_gdb_count = db_count;
3764         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3765         spin_lock_init(&sbi->s_next_gen_lock);
3766
3767         init_timer(&sbi->s_err_report);
3768         sbi->s_err_report.function = print_daily_error_info;
3769         sbi->s_err_report.data = (unsigned long) sb;
3770
3771         /* Register extent status tree shrinker */
3772         ext4_es_register_shrinker(sb);
3773
3774         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3775                         ext4_count_free_clusters(sb));
3776         if (!err) {
3777                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3778                                 ext4_count_free_inodes(sb));
3779         }
3780         if (!err) {
3781                 err = percpu_counter_init(&sbi->s_dirs_counter,
3782                                 ext4_count_dirs(sb));
3783         }
3784         if (!err) {
3785                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3786         }
3787         if (!err) {
3788                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3789         }
3790         if (err) {
3791                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3792                 goto failed_mount3;
3793         }
3794
3795         sbi->s_stripe = ext4_get_stripe_size(sbi);
3796         sbi->s_max_writeback_mb_bump = 128;
3797         sbi->s_extent_max_zeroout_kb = 32;
3798
3799         /*
3800          * set up enough so that it can read an inode
3801          */
3802         if (!test_opt(sb, NOLOAD) &&
3803             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3804                 sb->s_op = &ext4_sops;
3805         else
3806                 sb->s_op = &ext4_nojournal_sops;
3807         sb->s_export_op = &ext4_export_ops;
3808         sb->s_xattr = ext4_xattr_handlers;
3809 #ifdef CONFIG_QUOTA
3810         sb->dq_op = &ext4_quota_operations;
3811         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3812                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3813         else
3814                 sb->s_qcop = &ext4_qctl_operations;
3815 #endif
3816         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3817
3818         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3819         mutex_init(&sbi->s_orphan_lock);
3820
3821         sb->s_root = NULL;
3822
3823         needs_recovery = (es->s_last_orphan != 0 ||
3824                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3825                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3826
3827         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3828             !(sb->s_flags & MS_RDONLY))
3829                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3830                         goto failed_mount3;
3831
3832         /*
3833          * The first inode we look at is the journal inode.  Don't try
3834          * root first: it may be modified in the journal!
3835          */
3836         if (!test_opt(sb, NOLOAD) &&
3837             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3838                 if (ext4_load_journal(sb, es, journal_devnum))
3839                         goto failed_mount3;
3840         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3841               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3842                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3843                        "suppressed and not mounted read-only");
3844                 goto failed_mount_wq;
3845         } else {
3846                 clear_opt(sb, DATA_FLAGS);
3847                 sbi->s_journal = NULL;
3848                 needs_recovery = 0;
3849                 goto no_journal;
3850         }
3851
3852         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3853             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3854                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3855                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3856                 goto failed_mount_wq;
3857         }
3858
3859         if (!set_journal_csum_feature_set(sb)) {
3860                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3861                          "feature set");
3862                 goto failed_mount_wq;
3863         }
3864
3865         /* We have now updated the journal if required, so we can
3866          * validate the data journaling mode. */
3867         switch (test_opt(sb, DATA_FLAGS)) {
3868         case 0:
3869                 /* No mode set, assume a default based on the journal
3870                  * capabilities: ORDERED_DATA if the journal can
3871                  * cope, else JOURNAL_DATA
3872                  */
3873                 if (jbd2_journal_check_available_features
3874                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3875                         set_opt(sb, ORDERED_DATA);
3876                 else
3877                         set_opt(sb, JOURNAL_DATA);
3878                 break;
3879
3880         case EXT4_MOUNT_ORDERED_DATA:
3881         case EXT4_MOUNT_WRITEBACK_DATA:
3882                 if (!jbd2_journal_check_available_features
3883                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3884                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3885                                "requested data journaling mode");
3886                         goto failed_mount_wq;
3887                 }
3888         default:
3889                 break;
3890         }
3891         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3892
3893         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3894
3895         /*
3896          * The journal may have updated the bg summary counts, so we
3897          * need to update the global counters.
3898          */
3899         percpu_counter_set(&sbi->s_freeclusters_counter,
3900                            ext4_count_free_clusters(sb));
3901         percpu_counter_set(&sbi->s_freeinodes_counter,
3902                            ext4_count_free_inodes(sb));
3903         percpu_counter_set(&sbi->s_dirs_counter,
3904                            ext4_count_dirs(sb));
3905         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3906
3907 no_journal:
3908         /*
3909          * Get the # of file system overhead blocks from the
3910          * superblock if present.
3911          */
3912         if (es->s_overhead_clusters)
3913                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3914         else {
3915                 err = ext4_calculate_overhead(sb);
3916                 if (err)
3917                         goto failed_mount_wq;
3918         }
3919
3920         /*
3921          * The maximum number of concurrent works can be high and
3922          * concurrency isn't really necessary.  Limit it to 1.
3923          */
3924         EXT4_SB(sb)->dio_unwritten_wq =
3925                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3926         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3927                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3928                 ret = -ENOMEM;
3929                 goto failed_mount_wq;
3930         }
3931
3932         /*
3933          * The jbd2_journal_load will have done any necessary log recovery,
3934          * so we can safely mount the rest of the filesystem now.
3935          */
3936
3937         root = ext4_iget(sb, EXT4_ROOT_INO);
3938         if (IS_ERR(root)) {
3939                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3940                 ret = PTR_ERR(root);
3941                 root = NULL;
3942                 goto failed_mount4;
3943         }
3944         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3945                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3946                 iput(root);
3947                 goto failed_mount4;
3948         }
3949         sb->s_root = d_make_root(root);
3950         if (!sb->s_root) {
3951                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3952                 ret = -ENOMEM;
3953                 goto failed_mount4;
3954         }
3955
3956         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3957                 sb->s_flags |= MS_RDONLY;
3958
3959         /* determine the minimum size of new large inodes, if present */
3960         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3961                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3962                                                      EXT4_GOOD_OLD_INODE_SIZE;
3963                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3964                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3965                         if (sbi->s_want_extra_isize <
3966                             le16_to_cpu(es->s_want_extra_isize))
3967                                 sbi->s_want_extra_isize =
3968                                         le16_to_cpu(es->s_want_extra_isize);
3969                         if (sbi->s_want_extra_isize <
3970                             le16_to_cpu(es->s_min_extra_isize))
3971                                 sbi->s_want_extra_isize =
3972                                         le16_to_cpu(es->s_min_extra_isize);
3973                 }
3974         }
3975         /* Check if enough inode space is available */
3976         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3977                                                         sbi->s_inode_size) {
3978                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3979                                                        EXT4_GOOD_OLD_INODE_SIZE;
3980                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3981                          "available");
3982         }
3983
3984         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
3985         if (err) {
3986                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
3987                          "reserved pool", ext4_calculate_resv_clusters(sb));
3988                 goto failed_mount4a;
3989         }
3990
3991         err = ext4_setup_system_zone(sb);
3992         if (err) {
3993                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3994                          "zone (%d)", err);
3995                 goto failed_mount4a;
3996         }
3997
3998         ext4_ext_init(sb);
3999         err = ext4_mb_init(sb);
4000         if (err) {
4001                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4002                          err);
4003                 goto failed_mount5;
4004         }
4005
4006         err = ext4_register_li_request(sb, first_not_zeroed);
4007         if (err)
4008                 goto failed_mount6;
4009
4010         sbi->s_kobj.kset = ext4_kset;
4011         init_completion(&sbi->s_kobj_unregister);
4012         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4013                                    "%s", sb->s_id);
4014         if (err)
4015                 goto failed_mount7;
4016
4017 #ifdef CONFIG_QUOTA
4018         /* Enable quota usage during mount. */
4019         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4020             !(sb->s_flags & MS_RDONLY)) {
4021                 err = ext4_enable_quotas(sb);
4022                 if (err)
4023                         goto failed_mount8;
4024         }
4025 #endif  /* CONFIG_QUOTA */
4026
4027         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4028         ext4_orphan_cleanup(sb, es);
4029         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4030         if (needs_recovery) {
4031                 ext4_msg(sb, KERN_INFO, "recovery complete");
4032                 ext4_mark_recovery_complete(sb, es);
4033         }
4034         if (EXT4_SB(sb)->s_journal) {
4035                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4036                         descr = " journalled data mode";
4037                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4038                         descr = " ordered data mode";
4039                 else
4040                         descr = " writeback data mode";
4041         } else
4042                 descr = "out journal";
4043
4044         if (test_opt(sb, DISCARD)) {
4045                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4046                 if (!blk_queue_discard(q))
4047                         ext4_msg(sb, KERN_WARNING,
4048                                  "mounting with \"discard\" option, but "
4049                                  "the device does not support discard");
4050         }
4051
4052         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4053                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4054                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4055
4056         if (es->s_error_count)
4057                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4058
4059         kfree(orig_data);
4060         return 0;
4061
4062 cantfind_ext4:
4063         if (!silent)
4064                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4065         goto failed_mount;
4066
4067 #ifdef CONFIG_QUOTA
4068 failed_mount8:
4069         kobject_del(&sbi->s_kobj);
4070 #endif
4071 failed_mount7:
4072         ext4_unregister_li_request(sb);
4073 failed_mount6:
4074         ext4_mb_release(sb);
4075 failed_mount5:
4076         ext4_ext_release(sb);
4077         ext4_release_system_zone(sb);
4078 failed_mount4a:
4079         dput(sb->s_root);
4080         sb->s_root = NULL;
4081 failed_mount4:
4082         ext4_msg(sb, KERN_ERR, "mount failed");
4083         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4084 failed_mount_wq:
4085         if (sbi->s_journal) {
4086                 jbd2_journal_destroy(sbi->s_journal);
4087                 sbi->s_journal = NULL;
4088         }
4089 failed_mount3:
4090         ext4_es_unregister_shrinker(sb);
4091         del_timer(&sbi->s_err_report);
4092         if (sbi->s_flex_groups)
4093                 ext4_kvfree(sbi->s_flex_groups);
4094         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4095         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4096         percpu_counter_destroy(&sbi->s_dirs_counter);
4097         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4098         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4099         if (sbi->s_mmp_tsk)
4100                 kthread_stop(sbi->s_mmp_tsk);
4101 failed_mount2:
4102         for (i = 0; i < db_count; i++)
4103                 brelse(sbi->s_group_desc[i]);
4104         ext4_kvfree(sbi->s_group_desc);
4105 failed_mount:
4106         if (sbi->s_chksum_driver)
4107                 crypto_free_shash(sbi->s_chksum_driver);
4108         if (sbi->s_proc) {
4109                 remove_proc_entry("options", sbi->s_proc);
4110                 remove_proc_entry(sb->s_id, ext4_proc_root);
4111         }
4112 #ifdef CONFIG_QUOTA
4113         for (i = 0; i < MAXQUOTAS; i++)
4114                 kfree(sbi->s_qf_names[i]);
4115 #endif
4116         ext4_blkdev_remove(sbi);
4117         brelse(bh);
4118 out_fail:
4119         sb->s_fs_info = NULL;
4120         kfree(sbi->s_blockgroup_lock);
4121         kfree(sbi);
4122 out_free_orig:
4123         kfree(orig_data);
4124         return err ? err : ret;
4125 }
4126
4127 /*
4128  * Setup any per-fs journal parameters now.  We'll do this both on
4129  * initial mount, once the journal has been initialised but before we've
4130  * done any recovery; and again on any subsequent remount.
4131  */
4132 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4133 {
4134         struct ext4_sb_info *sbi = EXT4_SB(sb);
4135
4136         journal->j_commit_interval = sbi->s_commit_interval;
4137         journal->j_min_batch_time = sbi->s_min_batch_time;
4138         journal->j_max_batch_time = sbi->s_max_batch_time;
4139
4140         write_lock(&journal->j_state_lock);
4141         if (test_opt(sb, BARRIER))
4142                 journal->j_flags |= JBD2_BARRIER;
4143         else
4144                 journal->j_flags &= ~JBD2_BARRIER;
4145         if (test_opt(sb, DATA_ERR_ABORT))
4146                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4147         else
4148                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4149         write_unlock(&journal->j_state_lock);
4150 }
4151
4152 static journal_t *ext4_get_journal(struct super_block *sb,
4153                                    unsigned int journal_inum)
4154 {
4155         struct inode *journal_inode;
4156         journal_t *journal;
4157
4158         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4159
4160         /* First, test for the existence of a valid inode on disk.  Bad
4161          * things happen if we iget() an unused inode, as the subsequent
4162          * iput() will try to delete it. */
4163
4164         journal_inode = ext4_iget(sb, journal_inum);
4165         if (IS_ERR(journal_inode)) {
4166                 ext4_msg(sb, KERN_ERR, "no journal found");
4167                 return NULL;
4168         }
4169         if (!journal_inode->i_nlink) {
4170                 make_bad_inode(journal_inode);
4171                 iput(journal_inode);
4172                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4173                 return NULL;
4174         }
4175
4176         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4177                   journal_inode, journal_inode->i_size);
4178         if (!S_ISREG(journal_inode->i_mode)) {
4179                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4180                 iput(journal_inode);
4181                 return NULL;
4182         }
4183
4184         journal = jbd2_journal_init_inode(journal_inode);
4185         if (!journal) {
4186                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4187                 iput(journal_inode);
4188                 return NULL;
4189         }
4190         journal->j_private = sb;
4191         ext4_init_journal_params(sb, journal);
4192         return journal;
4193 }
4194
4195 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4196                                        dev_t j_dev)
4197 {
4198         struct buffer_head *bh;
4199         journal_t *journal;
4200         ext4_fsblk_t start;
4201         ext4_fsblk_t len;
4202         int hblock, blocksize;
4203         ext4_fsblk_t sb_block;
4204         unsigned long offset;
4205         struct ext4_super_block *es;
4206         struct block_device *bdev;
4207
4208         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4209
4210         bdev = ext4_blkdev_get(j_dev, sb);
4211         if (bdev == NULL)
4212                 return NULL;
4213
4214         blocksize = sb->s_blocksize;
4215         hblock = bdev_logical_block_size(bdev);
4216         if (blocksize < hblock) {
4217                 ext4_msg(sb, KERN_ERR,
4218                         "blocksize too small for journal device");
4219                 goto out_bdev;
4220         }
4221
4222         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4223         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4224         set_blocksize(bdev, blocksize);
4225         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4226                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4227                        "external journal");
4228                 goto out_bdev;
4229         }
4230
4231         es = (struct ext4_super_block *) (bh->b_data + offset);
4232         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4233             !(le32_to_cpu(es->s_feature_incompat) &
4234               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4235                 ext4_msg(sb, KERN_ERR, "external journal has "
4236                                         "bad superblock");
4237                 brelse(bh);
4238                 goto out_bdev;
4239         }
4240
4241         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4242                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4243                 brelse(bh);
4244                 goto out_bdev;
4245         }
4246
4247         len = ext4_blocks_count(es);
4248         start = sb_block + 1;
4249         brelse(bh);     /* we're done with the superblock */
4250
4251         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4252                                         start, len, blocksize);
4253         if (!journal) {
4254                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4255                 goto out_bdev;
4256         }
4257         journal->j_private = sb;
4258         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4259         wait_on_buffer(journal->j_sb_buffer);
4260         if (!buffer_uptodate(journal->j_sb_buffer)) {
4261                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4262                 goto out_journal;
4263         }
4264         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4265                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4266                                         "user (unsupported) - %d",
4267                         be32_to_cpu(journal->j_superblock->s_nr_users));
4268                 goto out_journal;
4269         }
4270         EXT4_SB(sb)->journal_bdev = bdev;
4271         ext4_init_journal_params(sb, journal);
4272         return journal;
4273
4274 out_journal:
4275         jbd2_journal_destroy(journal);
4276 out_bdev:
4277         ext4_blkdev_put(bdev);
4278         return NULL;
4279 }
4280
4281 static int ext4_load_journal(struct super_block *sb,
4282                              struct ext4_super_block *es,
4283                              unsigned long journal_devnum)
4284 {
4285         journal_t *journal;
4286         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4287         dev_t journal_dev;
4288         int err = 0;
4289         int really_read_only;
4290
4291         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4292
4293         if (journal_devnum &&
4294             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4295                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4296                         "numbers have changed");
4297                 journal_dev = new_decode_dev(journal_devnum);
4298         } else
4299                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4300
4301         really_read_only = bdev_read_only(sb->s_bdev);
4302
4303         /*
4304          * Are we loading a blank journal or performing recovery after a
4305          * crash?  For recovery, we need to check in advance whether we
4306          * can get read-write access to the device.
4307          */
4308         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4309                 if (sb->s_flags & MS_RDONLY) {
4310                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4311                                         "required on readonly filesystem");
4312                         if (really_read_only) {
4313                                 ext4_msg(sb, KERN_ERR, "write access "
4314                                         "unavailable, cannot proceed");
4315                                 return -EROFS;
4316                         }
4317                         ext4_msg(sb, KERN_INFO, "write access will "
4318                                "be enabled during recovery");
4319                 }
4320         }
4321
4322         if (journal_inum && journal_dev) {
4323                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4324                        "and inode journals!");
4325                 return -EINVAL;
4326         }
4327
4328         if (journal_inum) {
4329                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4330                         return -EINVAL;
4331         } else {
4332                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4333                         return -EINVAL;
4334         }
4335
4336         if (!(journal->j_flags & JBD2_BARRIER))
4337                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4338
4339         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4340                 err = jbd2_journal_wipe(journal, !really_read_only);
4341         if (!err) {
4342                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4343                 if (save)
4344                         memcpy(save, ((char *) es) +
4345                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4346                 err = jbd2_journal_load(journal);
4347                 if (save)
4348                         memcpy(((char *) es) + EXT4_S_ERR_START,
4349                                save, EXT4_S_ERR_LEN);
4350                 kfree(save);
4351         }
4352
4353         if (err) {
4354                 ext4_msg(sb, KERN_ERR, "error loading journal");
4355                 jbd2_journal_destroy(journal);
4356                 return err;
4357         }
4358
4359         EXT4_SB(sb)->s_journal = journal;
4360         ext4_clear_journal_err(sb, es);
4361
4362         if (!really_read_only && journal_devnum &&
4363             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4364                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4365
4366                 /* Make sure we flush the recovery flag to disk. */
4367                 ext4_commit_super(sb, 1);
4368         }
4369
4370         return 0;
4371 }
4372
4373 static int ext4_commit_super(struct super_block *sb, int sync)
4374 {
4375         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4376         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4377         int error = 0;
4378
4379         if (!sbh || block_device_ejected(sb))
4380                 return error;
4381         if (buffer_write_io_error(sbh)) {
4382                 /*
4383                  * Oh, dear.  A previous attempt to write the
4384                  * superblock failed.  This could happen because the
4385                  * USB device was yanked out.  Or it could happen to
4386                  * be a transient write error and maybe the block will
4387                  * be remapped.  Nothing we can do but to retry the
4388                  * write and hope for the best.
4389                  */
4390                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4391                        "superblock detected");
4392                 clear_buffer_write_io_error(sbh);
4393                 set_buffer_uptodate(sbh);
4394         }
4395         /*
4396          * If the file system is mounted read-only, don't update the
4397          * superblock write time.  This avoids updating the superblock
4398          * write time when we are mounting the root file system
4399          * read/only but we need to replay the journal; at that point,
4400          * for people who are east of GMT and who make their clock
4401          * tick in localtime for Windows bug-for-bug compatibility,
4402          * the clock is set in the future, and this will cause e2fsck
4403          * to complain and force a full file system check.
4404          */
4405         if (!(sb->s_flags & MS_RDONLY))
4406                 es->s_wtime = cpu_to_le32(get_seconds());
4407         if (sb->s_bdev->bd_part)
4408                 es->s_kbytes_written =
4409                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4410                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4411                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4412         else
4413                 es->s_kbytes_written =
4414                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4415         ext4_free_blocks_count_set(es,
4416                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4417                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4418         es->s_free_inodes_count =
4419                 cpu_to_le32(percpu_counter_sum_positive(
4420                                 &EXT4_SB(sb)->s_freeinodes_counter));
4421         BUFFER_TRACE(sbh, "marking dirty");
4422         ext4_superblock_csum_set(sb);
4423         mark_buffer_dirty(sbh);
4424         if (sync) {
4425                 error = sync_dirty_buffer(sbh);
4426                 if (error)
4427                         return error;
4428
4429                 error = buffer_write_io_error(sbh);
4430                 if (error) {
4431                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4432                                "superblock");
4433                         clear_buffer_write_io_error(sbh);
4434                         set_buffer_uptodate(sbh);
4435                 }
4436         }
4437         return error;
4438 }
4439
4440 /*
4441  * Have we just finished recovery?  If so, and if we are mounting (or
4442  * remounting) the filesystem readonly, then we will end up with a
4443  * consistent fs on disk.  Record that fact.
4444  */
4445 static void ext4_mark_recovery_complete(struct super_block *sb,
4446                                         struct ext4_super_block *es)
4447 {
4448         journal_t *journal = EXT4_SB(sb)->s_journal;
4449
4450         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4451                 BUG_ON(journal != NULL);
4452                 return;
4453         }
4454         jbd2_journal_lock_updates(journal);
4455         if (jbd2_journal_flush(journal) < 0)
4456                 goto out;
4457
4458         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4459             sb->s_flags & MS_RDONLY) {
4460                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4461                 ext4_commit_super(sb, 1);
4462         }
4463
4464 out:
4465         jbd2_journal_unlock_updates(journal);
4466 }
4467
4468 /*
4469  * If we are mounting (or read-write remounting) a filesystem whose journal
4470  * has recorded an error from a previous lifetime, move that error to the
4471  * main filesystem now.
4472  */
4473 static void ext4_clear_journal_err(struct super_block *sb,
4474                                    struct ext4_super_block *es)
4475 {
4476         journal_t *journal;
4477         int j_errno;
4478         const char *errstr;
4479
4480         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4481
4482         journal = EXT4_SB(sb)->s_journal;
4483
4484         /*
4485          * Now check for any error status which may have been recorded in the
4486          * journal by a prior ext4_error() or ext4_abort()
4487          */
4488
4489         j_errno = jbd2_journal_errno(journal);
4490         if (j_errno) {
4491                 char nbuf[16];
4492
4493                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4494                 ext4_warning(sb, "Filesystem error recorded "
4495                              "from previous mount: %s", errstr);
4496                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4497
4498                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4499                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4500                 ext4_commit_super(sb, 1);
4501
4502                 jbd2_journal_clear_err(journal);
4503                 jbd2_journal_update_sb_errno(journal);
4504         }
4505 }
4506
4507 /*
4508  * Force the running and committing transactions to commit,
4509  * and wait on the commit.
4510  */
4511 int ext4_force_commit(struct super_block *sb)
4512 {
4513         journal_t *journal;
4514
4515         if (sb->s_flags & MS_RDONLY)
4516                 return 0;
4517
4518         journal = EXT4_SB(sb)->s_journal;
4519         return ext4_journal_force_commit(journal);
4520 }
4521
4522 static int ext4_sync_fs(struct super_block *sb, int wait)
4523 {
4524         int ret = 0;
4525         tid_t target;
4526         struct ext4_sb_info *sbi = EXT4_SB(sb);
4527
4528         trace_ext4_sync_fs(sb, wait);
4529         flush_workqueue(sbi->dio_unwritten_wq);
4530         /*
4531          * Writeback quota in non-journalled quota case - journalled quota has
4532          * no dirty dquots
4533          */
4534         dquot_writeback_dquots(sb, -1);
4535         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4536                 if (wait)
4537                         jbd2_log_wait_commit(sbi->s_journal, target);
4538         }
4539         return ret;
4540 }
4541
4542 /*
4543  * LVM calls this function before a (read-only) snapshot is created.  This
4544  * gives us a chance to flush the journal completely and mark the fs clean.
4545  *
4546  * Note that only this function cannot bring a filesystem to be in a clean
4547  * state independently. It relies on upper layer to stop all data & metadata
4548  * modifications.
4549  */
4550 static int ext4_freeze(struct super_block *sb)
4551 {
4552         int error = 0;
4553         journal_t *journal;
4554
4555         if (sb->s_flags & MS_RDONLY)
4556                 return 0;
4557
4558         journal = EXT4_SB(sb)->s_journal;
4559
4560         /* Now we set up the journal barrier. */
4561         jbd2_journal_lock_updates(journal);
4562
4563         /*
4564          * Don't clear the needs_recovery flag if we failed to flush
4565          * the journal.
4566          */
4567         error = jbd2_journal_flush(journal);
4568         if (error < 0)
4569                 goto out;
4570
4571         /* Journal blocked and flushed, clear needs_recovery flag. */
4572         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4573         error = ext4_commit_super(sb, 1);
4574 out:
4575         /* we rely on upper layer to stop further updates */
4576         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4577         return error;
4578 }
4579
4580 /*
4581  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4582  * flag here, even though the filesystem is not technically dirty yet.
4583  */
4584 static int ext4_unfreeze(struct super_block *sb)
4585 {
4586         if (sb->s_flags & MS_RDONLY)
4587                 return 0;
4588
4589         /* Reset the needs_recovery flag before the fs is unlocked. */
4590         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4591         ext4_commit_super(sb, 1);
4592         return 0;
4593 }
4594
4595 /*
4596  * Structure to save mount options for ext4_remount's benefit
4597  */
4598 struct ext4_mount_options {
4599         unsigned long s_mount_opt;
4600         unsigned long s_mount_opt2;
4601         kuid_t s_resuid;
4602         kgid_t s_resgid;
4603         unsigned long s_commit_interval;
4604         u32 s_min_batch_time, s_max_batch_time;
4605 #ifdef CONFIG_QUOTA
4606         int s_jquota_fmt;
4607         char *s_qf_names[MAXQUOTAS];
4608 #endif
4609 };
4610
4611 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4612 {
4613         struct ext4_super_block *es;
4614         struct ext4_sb_info *sbi = EXT4_SB(sb);
4615         unsigned long old_sb_flags;
4616         struct ext4_mount_options old_opts;
4617         int enable_quota = 0;
4618         ext4_group_t g;
4619         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4620         int err = 0;
4621 #ifdef CONFIG_QUOTA
4622         int i, j;
4623 #endif
4624         char *orig_data = kstrdup(data, GFP_KERNEL);
4625
4626         /* Store the original options */
4627         old_sb_flags = sb->s_flags;
4628         old_opts.s_mount_opt = sbi->s_mount_opt;
4629         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4630         old_opts.s_resuid = sbi->s_resuid;
4631         old_opts.s_resgid = sbi->s_resgid;
4632         old_opts.s_commit_interval = sbi->s_commit_interval;
4633         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4634         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4635 #ifdef CONFIG_QUOTA
4636         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4637         for (i = 0; i < MAXQUOTAS; i++)
4638                 if (sbi->s_qf_names[i]) {
4639                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4640                                                          GFP_KERNEL);
4641                         if (!old_opts.s_qf_names[i]) {
4642                                 for (j = 0; j < i; j++)
4643                                         kfree(old_opts.s_qf_names[j]);
4644                                 kfree(orig_data);
4645                                 return -ENOMEM;
4646                         }
4647                 } else
4648                         old_opts.s_qf_names[i] = NULL;
4649 #endif
4650         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4651                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4652
4653         /*
4654          * Allow the "check" option to be passed as a remount option.
4655          */
4656         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4657                 err = -EINVAL;
4658                 goto restore_opts;
4659         }
4660
4661         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4662                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4663                         ext4_msg(sb, KERN_ERR, "can't mount with "
4664                                  "both data=journal and delalloc");
4665                         err = -EINVAL;
4666                         goto restore_opts;
4667                 }
4668                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4669                         ext4_msg(sb, KERN_ERR, "can't mount with "
4670                                  "both data=journal and dioread_nolock");
4671                         err = -EINVAL;
4672                         goto restore_opts;
4673                 }
4674         }
4675
4676         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4677                 ext4_abort(sb, "Abort forced by user");
4678
4679         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4680                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4681
4682         es = sbi->s_es;
4683
4684         if (sbi->s_journal) {
4685                 ext4_init_journal_params(sb, sbi->s_journal);
4686                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4687         }
4688
4689         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4690                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4691                         err = -EROFS;
4692                         goto restore_opts;
4693                 }
4694
4695                 if (*flags & MS_RDONLY) {
4696                         err = dquot_suspend(sb, -1);
4697                         if (err < 0)
4698                                 goto restore_opts;
4699
4700                         /*
4701                          * First of all, the unconditional stuff we have to do
4702                          * to disable replay of the journal when we next remount
4703                          */
4704                         sb->s_flags |= MS_RDONLY;
4705
4706                         /*
4707                          * OK, test if we are remounting a valid rw partition
4708                          * readonly, and if so set the rdonly flag and then
4709                          * mark the partition as valid again.
4710                          */
4711                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4712                             (sbi->s_mount_state & EXT4_VALID_FS))
4713                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4714
4715                         if (sbi->s_journal)
4716                                 ext4_mark_recovery_complete(sb, es);
4717                 } else {
4718                         /* Make sure we can mount this feature set readwrite */
4719                         if (!ext4_feature_set_ok(sb, 0)) {
4720                                 err = -EROFS;
4721                                 goto restore_opts;
4722                         }
4723                         /*
4724                          * Make sure the group descriptor checksums
4725                          * are sane.  If they aren't, refuse to remount r/w.
4726                          */
4727                         for (g = 0; g < sbi->s_groups_count; g++) {
4728                                 struct ext4_group_desc *gdp =
4729                                         ext4_get_group_desc(sb, g, NULL);
4730
4731                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4732                                         ext4_msg(sb, KERN_ERR,
4733                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4734                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4735                                                le16_to_cpu(gdp->bg_checksum));
4736                                         err = -EINVAL;
4737                                         goto restore_opts;
4738                                 }
4739                         }
4740
4741                         /*
4742                          * If we have an unprocessed orphan list hanging
4743                          * around from a previously readonly bdev mount,
4744                          * require a full umount/remount for now.
4745                          */
4746                         if (es->s_last_orphan) {
4747                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4748                                        "remount RDWR because of unprocessed "
4749                                        "orphan inode list.  Please "
4750                                        "umount/remount instead");
4751                                 err = -EINVAL;
4752                                 goto restore_opts;
4753                         }
4754
4755                         /*
4756                          * Mounting a RDONLY partition read-write, so reread
4757                          * and store the current valid flag.  (It may have
4758                          * been changed by e2fsck since we originally mounted
4759                          * the partition.)
4760                          */
4761                         if (sbi->s_journal)
4762                                 ext4_clear_journal_err(sb, es);
4763                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4764                         if (!ext4_setup_super(sb, es, 0))
4765                                 sb->s_flags &= ~MS_RDONLY;
4766                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4767                                                      EXT4_FEATURE_INCOMPAT_MMP))
4768                                 if (ext4_multi_mount_protect(sb,
4769                                                 le64_to_cpu(es->s_mmp_block))) {
4770                                         err = -EROFS;
4771                                         goto restore_opts;
4772                                 }
4773                         enable_quota = 1;
4774                 }
4775         }
4776
4777         /*
4778          * Reinitialize lazy itable initialization thread based on
4779          * current settings
4780          */
4781         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4782                 ext4_unregister_li_request(sb);
4783         else {
4784                 ext4_group_t first_not_zeroed;
4785                 first_not_zeroed = ext4_has_uninit_itable(sb);
4786                 ext4_register_li_request(sb, first_not_zeroed);
4787         }
4788
4789         ext4_setup_system_zone(sb);
4790         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4791                 ext4_commit_super(sb, 1);
4792
4793 #ifdef CONFIG_QUOTA
4794         /* Release old quota file names */
4795         for (i = 0; i < MAXQUOTAS; i++)
4796                 kfree(old_opts.s_qf_names[i]);
4797         if (enable_quota) {
4798                 if (sb_any_quota_suspended(sb))
4799                         dquot_resume(sb, -1);
4800                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4801                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4802                         err = ext4_enable_quotas(sb);
4803                         if (err)
4804                                 goto restore_opts;
4805                 }
4806         }
4807 #endif
4808
4809         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4810         kfree(orig_data);
4811         return 0;
4812
4813 restore_opts:
4814         sb->s_flags = old_sb_flags;
4815         sbi->s_mount_opt = old_opts.s_mount_opt;
4816         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4817         sbi->s_resuid = old_opts.s_resuid;
4818         sbi->s_resgid = old_opts.s_resgid;
4819         sbi->s_commit_interval = old_opts.s_commit_interval;
4820         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4821         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4822 #ifdef CONFIG_QUOTA
4823         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4824         for (i = 0; i < MAXQUOTAS; i++) {
4825                 kfree(sbi->s_qf_names[i]);
4826                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4827         }
4828 #endif
4829         kfree(orig_data);
4830         return err;
4831 }
4832
4833 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4834 {
4835         struct super_block *sb = dentry->d_sb;
4836         struct ext4_sb_info *sbi = EXT4_SB(sb);
4837         struct ext4_super_block *es = sbi->s_es;
4838         ext4_fsblk_t overhead = 0, resv_blocks;
4839         u64 fsid;
4840         s64 bfree;
4841         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4842
4843         if (!test_opt(sb, MINIX_DF))
4844                 overhead = sbi->s_overhead;
4845
4846         buf->f_type = EXT4_SUPER_MAGIC;
4847         buf->f_bsize = sb->s_blocksize;
4848         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4849         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4850                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4851         /* prevent underflow in case that few free space is available */
4852         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4853         buf->f_bavail = buf->f_bfree -
4854                         (ext4_r_blocks_count(es) + resv_blocks);
4855         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4856                 buf->f_bavail = 0;
4857         buf->f_files = le32_to_cpu(es->s_inodes_count);
4858         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4859         buf->f_namelen = EXT4_NAME_LEN;
4860         fsid = le64_to_cpup((void *)es->s_uuid) ^
4861                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4862         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4863         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4864
4865         return 0;
4866 }
4867
4868 /* Helper function for writing quotas on sync - we need to start transaction
4869  * before quota file is locked for write. Otherwise the are possible deadlocks:
4870  * Process 1                         Process 2
4871  * ext4_create()                     quota_sync()
4872  *   jbd2_journal_start()                  write_dquot()
4873  *   dquot_initialize()                         down(dqio_mutex)
4874  *     down(dqio_mutex)                    jbd2_journal_start()
4875  *
4876  */
4877
4878 #ifdef CONFIG_QUOTA
4879
4880 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4881 {
4882         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4883 }
4884
4885 static int ext4_write_dquot(struct dquot *dquot)
4886 {
4887         int ret, err;
4888         handle_t *handle;
4889         struct inode *inode;
4890
4891         inode = dquot_to_inode(dquot);
4892         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4893                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4894         if (IS_ERR(handle))
4895                 return PTR_ERR(handle);
4896         ret = dquot_commit(dquot);
4897         err = ext4_journal_stop(handle);
4898         if (!ret)
4899                 ret = err;
4900         return ret;
4901 }
4902
4903 static int ext4_acquire_dquot(struct dquot *dquot)
4904 {
4905         int ret, err;
4906         handle_t *handle;
4907
4908         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4909                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4910         if (IS_ERR(handle))
4911                 return PTR_ERR(handle);
4912         ret = dquot_acquire(dquot);
4913         err = ext4_journal_stop(handle);
4914         if (!ret)
4915                 ret = err;
4916         return ret;
4917 }
4918
4919 static int ext4_release_dquot(struct dquot *dquot)
4920 {
4921         int ret, err;
4922         handle_t *handle;
4923
4924         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4925                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4926         if (IS_ERR(handle)) {
4927                 /* Release dquot anyway to avoid endless cycle in dqput() */
4928                 dquot_release(dquot);
4929                 return PTR_ERR(handle);
4930         }
4931         ret = dquot_release(dquot);
4932         err = ext4_journal_stop(handle);
4933         if (!ret)
4934                 ret = err;
4935         return ret;
4936 }
4937
4938 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4939 {
4940         struct super_block *sb = dquot->dq_sb;
4941         struct ext4_sb_info *sbi = EXT4_SB(sb);
4942
4943         /* Are we journaling quotas? */
4944         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4945             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4946                 dquot_mark_dquot_dirty(dquot);
4947                 return ext4_write_dquot(dquot);
4948         } else {
4949                 return dquot_mark_dquot_dirty(dquot);
4950         }
4951 }
4952
4953 static int ext4_write_info(struct super_block *sb, int type)
4954 {
4955         int ret, err;
4956         handle_t *handle;
4957
4958         /* Data block + inode block */
4959         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
4960         if (IS_ERR(handle))
4961                 return PTR_ERR(handle);
4962         ret = dquot_commit_info(sb, type);
4963         err = ext4_journal_stop(handle);
4964         if (!ret)
4965                 ret = err;
4966         return ret;
4967 }
4968
4969 /*
4970  * Turn on quotas during mount time - we need to find
4971  * the quota file and such...
4972  */
4973 static int ext4_quota_on_mount(struct super_block *sb, int type)
4974 {
4975         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4976                                         EXT4_SB(sb)->s_jquota_fmt, type);
4977 }
4978
4979 /*
4980  * Standard function to be called on quota_on
4981  */
4982 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4983                          struct path *path)
4984 {
4985         int err;
4986
4987         if (!test_opt(sb, QUOTA))
4988                 return -EINVAL;
4989
4990         /* Quotafile not on the same filesystem? */
4991         if (path->dentry->d_sb != sb)
4992                 return -EXDEV;
4993         /* Journaling quota? */
4994         if (EXT4_SB(sb)->s_qf_names[type]) {
4995                 /* Quotafile not in fs root? */
4996                 if (path->dentry->d_parent != sb->s_root)
4997                         ext4_msg(sb, KERN_WARNING,
4998                                 "Quota file not on filesystem root. "
4999                                 "Journaled quota will not work");
5000         }
5001
5002         /*
5003          * When we journal data on quota file, we have to flush journal to see
5004          * all updates to the file when we bypass pagecache...
5005          */
5006         if (EXT4_SB(sb)->s_journal &&
5007             ext4_should_journal_data(path->dentry->d_inode)) {
5008                 /*
5009                  * We don't need to lock updates but journal_flush() could
5010                  * otherwise be livelocked...
5011                  */
5012                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5013                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5014                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5015                 if (err)
5016                         return err;
5017         }
5018
5019         return dquot_quota_on(sb, type, format_id, path);
5020 }
5021
5022 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5023                              unsigned int flags)
5024 {
5025         int err;
5026         struct inode *qf_inode;
5027         unsigned long qf_inums[MAXQUOTAS] = {
5028                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5029                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5030         };
5031
5032         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5033
5034         if (!qf_inums[type])
5035                 return -EPERM;
5036
5037         qf_inode = ext4_iget(sb, qf_inums[type]);
5038         if (IS_ERR(qf_inode)) {
5039                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5040                 return PTR_ERR(qf_inode);
5041         }
5042
5043         /* Don't account quota for quota files to avoid recursion */
5044         qf_inode->i_flags |= S_NOQUOTA;
5045         err = dquot_enable(qf_inode, type, format_id, flags);
5046         iput(qf_inode);
5047
5048         return err;
5049 }
5050
5051 /* Enable usage tracking for all quota types. */
5052 static int ext4_enable_quotas(struct super_block *sb)
5053 {
5054         int type, err = 0;
5055         unsigned long qf_inums[MAXQUOTAS] = {
5056                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5057                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5058         };
5059
5060         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5061         for (type = 0; type < MAXQUOTAS; type++) {
5062                 if (qf_inums[type]) {
5063                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5064                                                 DQUOT_USAGE_ENABLED);
5065                         if (err) {
5066                                 ext4_warning(sb,
5067                                         "Failed to enable quota tracking "
5068                                         "(type=%d, err=%d). Please run "
5069                                         "e2fsck to fix.", type, err);
5070                                 return err;
5071                         }
5072                 }
5073         }
5074         return 0;
5075 }
5076
5077 /*
5078  * quota_on function that is used when QUOTA feature is set.
5079  */
5080 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5081                                  int format_id)
5082 {
5083         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5084                 return -EINVAL;
5085
5086         /*
5087          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5088          */
5089         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5090 }
5091
5092 static int ext4_quota_off(struct super_block *sb, int type)
5093 {
5094         struct inode *inode = sb_dqopt(sb)->files[type];
5095         handle_t *handle;
5096
5097         /* Force all delayed allocation blocks to be allocated.
5098          * Caller already holds s_umount sem */
5099         if (test_opt(sb, DELALLOC))
5100                 sync_filesystem(sb);
5101
5102         if (!inode)
5103                 goto out;
5104
5105         /* Update modification times of quota files when userspace can
5106          * start looking at them */
5107         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5108         if (IS_ERR(handle))
5109                 goto out;
5110         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5111         ext4_mark_inode_dirty(handle, inode);
5112         ext4_journal_stop(handle);
5113
5114 out:
5115         return dquot_quota_off(sb, type);
5116 }
5117
5118 /*
5119  * quota_off function that is used when QUOTA feature is set.
5120  */
5121 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5122 {
5123         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5124                 return -EINVAL;
5125
5126         /* Disable only the limits. */
5127         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5128 }
5129
5130 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5131  * acquiring the locks... As quota files are never truncated and quota code
5132  * itself serializes the operations (and no one else should touch the files)
5133  * we don't have to be afraid of races */
5134 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5135                                size_t len, loff_t off)
5136 {
5137         struct inode *inode = sb_dqopt(sb)->files[type];
5138         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5139         int err = 0;
5140         int offset = off & (sb->s_blocksize - 1);
5141         int tocopy;
5142         size_t toread;
5143         struct buffer_head *bh;
5144         loff_t i_size = i_size_read(inode);
5145
5146         if (off > i_size)
5147                 return 0;
5148         if (off+len > i_size)
5149                 len = i_size-off;
5150         toread = len;
5151         while (toread > 0) {
5152                 tocopy = sb->s_blocksize - offset < toread ?
5153                                 sb->s_blocksize - offset : toread;
5154                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5155                 if (err)
5156                         return err;
5157                 if (!bh)        /* A hole? */
5158                         memset(data, 0, tocopy);
5159                 else
5160                         memcpy(data, bh->b_data+offset, tocopy);
5161                 brelse(bh);
5162                 offset = 0;
5163                 toread -= tocopy;
5164                 data += tocopy;
5165                 blk++;
5166         }
5167         return len;
5168 }
5169
5170 /* Write to quotafile (we know the transaction is already started and has
5171  * enough credits) */
5172 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5173                                 const char *data, size_t len, loff_t off)
5174 {
5175         struct inode *inode = sb_dqopt(sb)->files[type];
5176         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5177         int err = 0;
5178         int offset = off & (sb->s_blocksize - 1);
5179         struct buffer_head *bh;
5180         handle_t *handle = journal_current_handle();
5181
5182         if (EXT4_SB(sb)->s_journal && !handle) {
5183                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5184                         " cancelled because transaction is not started",
5185                         (unsigned long long)off, (unsigned long long)len);
5186                 return -EIO;
5187         }
5188         /*
5189          * Since we account only one data block in transaction credits,
5190          * then it is impossible to cross a block boundary.
5191          */
5192         if (sb->s_blocksize - offset < len) {
5193                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5194                         " cancelled because not block aligned",
5195                         (unsigned long long)off, (unsigned long long)len);
5196                 return -EIO;
5197         }
5198
5199         bh = ext4_bread(handle, inode, blk, 1, &err);
5200         if (!bh)
5201                 goto out;
5202         err = ext4_journal_get_write_access(handle, bh);
5203         if (err) {
5204                 brelse(bh);
5205                 goto out;
5206         }
5207         lock_buffer(bh);
5208         memcpy(bh->b_data+offset, data, len);
5209         flush_dcache_page(bh->b_page);
5210         unlock_buffer(bh);
5211         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5212         brelse(bh);
5213 out:
5214         if (err)
5215                 return err;
5216         if (inode->i_size < off + len) {
5217                 i_size_write(inode, off + len);
5218                 EXT4_I(inode)->i_disksize = inode->i_size;
5219                 ext4_mark_inode_dirty(handle, inode);
5220         }
5221         return len;
5222 }
5223
5224 #endif
5225
5226 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5227                        const char *dev_name, void *data)
5228 {
5229         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5230 }
5231
5232 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5233 static inline void register_as_ext2(void)
5234 {
5235         int err = register_filesystem(&ext2_fs_type);
5236         if (err)
5237                 printk(KERN_WARNING
5238                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5239 }
5240
5241 static inline void unregister_as_ext2(void)
5242 {
5243         unregister_filesystem(&ext2_fs_type);
5244 }
5245
5246 static inline int ext2_feature_set_ok(struct super_block *sb)
5247 {
5248         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5249                 return 0;
5250         if (sb->s_flags & MS_RDONLY)
5251                 return 1;
5252         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5253                 return 0;
5254         return 1;
5255 }
5256 #else
5257 static inline void register_as_ext2(void) { }
5258 static inline void unregister_as_ext2(void) { }
5259 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5260 #endif
5261
5262 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5263 static inline void register_as_ext3(void)
5264 {
5265         int err = register_filesystem(&ext3_fs_type);
5266         if (err)
5267                 printk(KERN_WARNING
5268                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5269 }
5270
5271 static inline void unregister_as_ext3(void)
5272 {
5273         unregister_filesystem(&ext3_fs_type);
5274 }
5275
5276 static inline int ext3_feature_set_ok(struct super_block *sb)
5277 {
5278         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5279                 return 0;
5280         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5281                 return 0;
5282         if (sb->s_flags & MS_RDONLY)
5283                 return 1;
5284         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5285                 return 0;
5286         return 1;
5287 }
5288 #else
5289 static inline void register_as_ext3(void) { }
5290 static inline void unregister_as_ext3(void) { }
5291 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5292 #endif
5293
5294 static struct file_system_type ext4_fs_type = {
5295         .owner          = THIS_MODULE,
5296         .name           = "ext4",
5297         .mount          = ext4_mount,
5298         .kill_sb        = kill_block_super,
5299         .fs_flags       = FS_REQUIRES_DEV,
5300 };
5301 MODULE_ALIAS_FS("ext4");
5302
5303 static int __init ext4_init_feat_adverts(void)
5304 {
5305         struct ext4_features *ef;
5306         int ret = -ENOMEM;
5307
5308         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5309         if (!ef)
5310                 goto out;
5311
5312         ef->f_kobj.kset = ext4_kset;
5313         init_completion(&ef->f_kobj_unregister);
5314         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5315                                    "features");
5316         if (ret) {
5317                 kfree(ef);
5318                 goto out;
5319         }
5320
5321         ext4_feat = ef;
5322         ret = 0;
5323 out:
5324         return ret;
5325 }
5326
5327 static void ext4_exit_feat_adverts(void)
5328 {
5329         kobject_put(&ext4_feat->f_kobj);
5330         wait_for_completion(&ext4_feat->f_kobj_unregister);
5331         kfree(ext4_feat);
5332 }
5333
5334 /* Shared across all ext4 file systems */
5335 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5336 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5337
5338 static int __init ext4_init_fs(void)
5339 {
5340         int i, err;
5341
5342         ext4_li_info = NULL;
5343         mutex_init(&ext4_li_mtx);
5344
5345         /* Build-time check for flags consistency */
5346         ext4_check_flag_values();
5347
5348         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5349                 mutex_init(&ext4__aio_mutex[i]);
5350                 init_waitqueue_head(&ext4__ioend_wq[i]);
5351         }
5352
5353         err = ext4_init_es();
5354         if (err)
5355                 return err;
5356
5357         err = ext4_init_pageio();
5358         if (err)
5359                 goto out7;
5360
5361         err = ext4_init_system_zone();
5362         if (err)
5363                 goto out6;
5364         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5365         if (!ext4_kset) {
5366                 err = -ENOMEM;
5367                 goto out5;
5368         }
5369         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5370
5371         err = ext4_init_feat_adverts();
5372         if (err)
5373                 goto out4;
5374
5375         err = ext4_init_mballoc();
5376         if (err)
5377                 goto out3;
5378
5379         err = ext4_init_xattr();
5380         if (err)
5381                 goto out2;
5382         err = init_inodecache();
5383         if (err)
5384                 goto out1;
5385         register_as_ext3();
5386         register_as_ext2();
5387         err = register_filesystem(&ext4_fs_type);
5388         if (err)
5389                 goto out;
5390
5391         return 0;
5392 out:
5393         unregister_as_ext2();
5394         unregister_as_ext3();
5395         destroy_inodecache();
5396 out1:
5397         ext4_exit_xattr();
5398 out2:
5399         ext4_exit_mballoc();
5400 out3:
5401         ext4_exit_feat_adverts();
5402 out4:
5403         if (ext4_proc_root)
5404                 remove_proc_entry("fs/ext4", NULL);
5405         kset_unregister(ext4_kset);
5406 out5:
5407         ext4_exit_system_zone();
5408 out6:
5409         ext4_exit_pageio();
5410 out7:
5411         ext4_exit_es();
5412
5413         return err;
5414 }
5415
5416 static void __exit ext4_exit_fs(void)
5417 {
5418         ext4_destroy_lazyinit_thread();
5419         unregister_as_ext2();
5420         unregister_as_ext3();
5421         unregister_filesystem(&ext4_fs_type);
5422         destroy_inodecache();
5423         ext4_exit_xattr();
5424         ext4_exit_mballoc();
5425         ext4_exit_feat_adverts();
5426         remove_proc_entry("fs/ext4", NULL);
5427         kset_unregister(ext4_kset);
5428         ext4_exit_system_zone();
5429         ext4_exit_pageio();
5430         ext4_exit_es();
5431 }
5432
5433 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5434 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5435 MODULE_LICENSE("GPL");
5436 module_init(ext4_init_fs)
5437 module_exit(ext4_exit_fs)