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_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct proc_dir_entry *ext4_proc_root;
  57 static struct kset *ext4_kset;
  58 static struct ext4_lazy_init *ext4_li_info;
  59 static struct mutex ext4_li_mtx;
  60 static struct ext4_features *ext4_feat;
  61
  62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  63                              unsigned long journal_devnum);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static const char *ext4_decode_error(struct super_block *sb, int errno,
  71                                      char nbuf[16]);
  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 void ext4_write_super(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85
  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 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  95 #else
  96 #define IS_EXT2_SB(sb) (0)
  97 #endif
  98
  99
 100 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 101 static struct file_system_type ext3_fs_type = {
 102         .owner          = THIS_MODULE,
 103         .name           = "ext3",
 104         .mount          = ext4_mount,
 105         .kill_sb        = kill_block_super,
 106         .fs_flags       = FS_REQUIRES_DEV,
 107 };
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109 #else
 110 #define IS_EXT3_SB(sb) (0)
 111 #endif
 112
 113 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 114                                struct ext4_group_desc *bg)
 115 {
 116         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 117                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 118                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 119 }
 120
 121 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 122                                struct ext4_group_desc *bg)
 123 {
 124         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 125                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 126                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 127 }
 128
 129 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 130                               struct ext4_group_desc *bg)
 131 {
 132         return le32_to_cpu(bg->bg_inode_table_lo) |
 133                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 134                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 135 }
 136
 137 __u32 ext4_free_blks_count(struct super_block *sb,
 138                               struct ext4_group_desc *bg)
 139 {
 140         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 141                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 142                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 143 }
 144
 145 __u32 ext4_free_inodes_count(struct super_block *sb,
 146                               struct ext4_group_desc *bg)
 147 {
 148         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 149                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 150                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 151 }
 152
 153 __u32 ext4_used_dirs_count(struct super_block *sb,
 154                               struct ext4_group_desc *bg)
 155 {
 156         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 157                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 158                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 159 }
 160
 161 __u32 ext4_itable_unused_count(struct super_block *sb,
 162                               struct ext4_group_desc *bg)
 163 {
 164         return le16_to_cpu(bg->bg_itable_unused_lo) |
 165                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 166                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 167 }
 168
 169 void ext4_block_bitmap_set(struct super_block *sb,
 170                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 171 {
 172         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 173         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 174                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 175 }
 176
 177 void ext4_inode_bitmap_set(struct super_block *sb,
 178                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 179 {
 180         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 181         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 182                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 183 }
 184
 185 void ext4_inode_table_set(struct super_block *sb,
 186                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 187 {
 188         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 189         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 190                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 191 }
 192
 193 void ext4_free_blks_set(struct super_block *sb,
 194                           struct ext4_group_desc *bg, __u32 count)
 195 {
 196         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 197         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 198                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 199 }
 200
 201 void ext4_free_inodes_set(struct super_block *sb,
 202                           struct ext4_group_desc *bg, __u32 count)
 203 {
 204         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 205         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 206                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 207 }
 208
 209 void ext4_used_dirs_set(struct super_block *sb,
 210                           struct ext4_group_desc *bg, __u32 count)
 211 {
 212         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 213         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 214                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 215 }
 216
 217 void ext4_itable_unused_set(struct super_block *sb,
 218                           struct ext4_group_desc *bg, __u32 count)
 219 {
 220         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 221         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 222                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 223 }
 224
 225
 226 /* Just increment the non-pointer handle value */
 227 static handle_t *ext4_get_nojournal(void)
 228 {
 229         handle_t *handle = current->journal_info;
 230         unsigned long ref_cnt = (unsigned long)handle;
 231
 232         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 233
 234         ref_cnt++;
 235         handle = (handle_t *)ref_cnt;
 236
 237         current->journal_info = handle;
 238         return handle;
 239 }
 240
 241
 242 /* Decrement the non-pointer handle value */
 243 static void ext4_put_nojournal(handle_t *handle)
 244 {
 245         unsigned long ref_cnt = (unsigned long)handle;
 246
 247         BUG_ON(ref_cnt == 0);
 248
 249         ref_cnt--;
 250         handle = (handle_t *)ref_cnt;
 251
 252         current->journal_info = handle;
 253 }
 254
 255 /*
 256  * Wrappers for jbd2_journal_start/end.
 257  *
 258  * The only special thing we need to do here is to make sure that all
 259  * journal_end calls result in the superblock being marked dirty, so
 260  * that sync() will call the filesystem's write_super callback if
 261  * appropriate.
 262  *
 263  * To avoid j_barrier hold in userspace when a user calls freeze(),
 264  * ext4 prevents a new handle from being started by s_frozen, which
 265  * is in an upper layer.
 266  */
 267 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 268 {
 269         journal_t *journal;
 270         handle_t  *handle;
 271
 272         if (sb->s_flags & MS_RDONLY)
 273                 return ERR_PTR(-EROFS);
 274
 275         journal = EXT4_SB(sb)->s_journal;
 276         handle = ext4_journal_current_handle();
 277
 278         /*
 279          * If a handle has been started, it should be allowed to
 280          * finish, otherwise deadlock could happen between freeze
 281          * and others(e.g. truncate) due to the restart of the
 282          * journal handle if the filesystem is forzen and active
 283          * handles are not stopped.
 284          */
 285         if (!handle)
 286                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 287
 288         if (!journal)
 289                 return ext4_get_nojournal();
 290         /*
 291          * Special case here: if the journal has aborted behind our
 292          * backs (eg. EIO in the commit thread), then we still need to
 293          * take the FS itself readonly cleanly.
 294          */
 295         if (is_journal_aborted(journal)) {
 296                 ext4_abort(sb, "Detected aborted journal");
 297                 return ERR_PTR(-EROFS);
 298         }
 299         return jbd2_journal_start(journal, nblocks);
 300 }
 301
 302 /*
 303  * The only special thing we need to do here is to make sure that all
 304  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 305  * that sync() will call the filesystem's write_super callback if
 306  * appropriate.
 307  */
 308 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 309 {
 310         struct super_block *sb;
 311         int err;
 312         int rc;
 313
 314         if (!ext4_handle_valid(handle)) {
 315                 ext4_put_nojournal(handle);
 316                 return 0;
 317         }
 318         sb = handle->h_transaction->t_journal->j_private;
 319         err = handle->h_err;
 320         rc = jbd2_journal_stop(handle);
 321
 322         if (!err)
 323                 err = rc;
 324         if (err)
 325                 __ext4_std_error(sb, where, line, err);
 326         return err;
 327 }
 328
 329 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 330                                const char *err_fn, struct buffer_head *bh,
 331                                handle_t *handle, int err)
 332 {
 333         char nbuf[16];
 334         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 335
 336         BUG_ON(!ext4_handle_valid(handle));
 337
 338         if (bh)
 339                 BUFFER_TRACE(bh, "abort");
 340
 341         if (!handle->h_err)
 342                 handle->h_err = err;
 343
 344         if (is_handle_aborted(handle))
 345                 return;
 346
 347         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 348                caller, line, errstr, err_fn);
 349
 350         jbd2_journal_abort_handle(handle);
 351 }
 352
 353 static void __save_error_info(struct super_block *sb, const char *func,
 354                             unsigned int line)
 355 {
 356         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 357
 358         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 359         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 360         es->s_last_error_time = cpu_to_le32(get_seconds());
 361         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 362         es->s_last_error_line = cpu_to_le32(line);
 363         if (!es->s_first_error_time) {
 364                 es->s_first_error_time = es->s_last_error_time;
 365                 strncpy(es->s_first_error_func, func,
 366                         sizeof(es->s_first_error_func));
 367                 es->s_first_error_line = cpu_to_le32(line);
 368                 es->s_first_error_ino = es->s_last_error_ino;
 369                 es->s_first_error_block = es->s_last_error_block;
 370         }
 371         /*
 372          * Start the daily error reporting function if it hasn't been
 373          * started already
 374          */
 375         if (!es->s_error_count)
 376                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 377         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 378 }
 379
 380 static void save_error_info(struct super_block *sb, const char *func,
 381                             unsigned int line)
 382 {
 383         __save_error_info(sb, func, line);
 384         ext4_commit_super(sb, 1);
 385 }
 386
 387
 388 /* Deal with the reporting of failure conditions on a filesystem such as
 389  * inconsistencies detected or read IO failures.
 390  *
 391  * On ext2, we can store the error state of the filesystem in the
 392  * superblock.  That is not possible on ext4, because we may have other
 393  * write ordering constraints on the superblock which prevent us from
 394  * writing it out straight away; and given that the journal is about to
 395  * be aborted, we can't rely on the current, or future, transactions to
 396  * write out the superblock safely.
 397  *
 398  * We'll just use the jbd2_journal_abort() error code to record an error in
 399  * the journal instead.  On recovery, the journal will complain about
 400  * that error until we've noted it down and cleared it.
 401  */
 402
 403 static void ext4_handle_error(struct super_block *sb)
 404 {
 405         if (sb->s_flags & MS_RDONLY)
 406                 return;
 407
 408         if (!test_opt(sb, ERRORS_CONT)) {
 409                 journal_t *journal = EXT4_SB(sb)->s_journal;
 410
 411                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 412                 if (journal)
 413                         jbd2_journal_abort(journal, -EIO);
 414         }
 415         if (test_opt(sb, ERRORS_RO)) {
 416                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 417                 sb->s_flags |= MS_RDONLY;
 418         }
 419         if (test_opt(sb, ERRORS_PANIC))
 420                 panic("EXT4-fs (device %s): panic forced after error\n",
 421                         sb->s_id);
 422 }
 423
 424 void __ext4_error(struct super_block *sb, const char *function,
 425                   unsigned int line, const char *fmt, ...)
 426 {
 427         struct va_format vaf;
 428         va_list args;
 429
 430         va_start(args, fmt);
 431         vaf.fmt = fmt;
 432         vaf.va = &args;
 433         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 434                sb->s_id, function, line, current->comm, &vaf);
 435         va_end(args);
 436
 437         ext4_handle_error(sb);
 438 }
 439
 440 void ext4_error_inode(struct inode *inode, const char *function,
 441                       unsigned int line, ext4_fsblk_t block,
 442                       const char *fmt, ...)
 443 {
 444         va_list args;
 445         struct va_format vaf;
 446         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 447
 448         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 449         es->s_last_error_block = cpu_to_le64(block);
 450         save_error_info(inode->i_sb, function, line);
 451         va_start(args, fmt);
 452         vaf.fmt = fmt;
 453         vaf.va = &args;
 454         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 455                inode->i_sb->s_id, function, line, inode->i_ino);
 456         if (block)
 457                 printk(KERN_CONT "block %llu: ", block);
 458         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 459         va_end(args);
 460
 461         ext4_handle_error(inode->i_sb);
 462 }
 463
 464 void ext4_error_file(struct file *file, const char *function,
 465                      unsigned int line, ext4_fsblk_t block,
 466                      const char *fmt, ...)
 467 {
 468         va_list args;
 469         struct va_format vaf;
 470         struct ext4_super_block *es;
 471         struct inode *inode = file->f_dentry->d_inode;
 472         char pathname[80], *path;
 473
 474         es = EXT4_SB(inode->i_sb)->s_es;
 475         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 476         save_error_info(inode->i_sb, function, line);
 477         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 478         if (IS_ERR(path))
 479                 path = "(unknown)";
 480         printk(KERN_CRIT
 481                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 482                inode->i_sb->s_id, function, line, inode->i_ino);
 483         if (block)
 484                 printk(KERN_CONT "block %llu: ", block);
 485         va_start(args, fmt);
 486         vaf.fmt = fmt;
 487         vaf.va = &args;
 488         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 489         va_end(args);
 490
 491         ext4_handle_error(inode->i_sb);
 492 }
 493
 494 static const char *ext4_decode_error(struct super_block *sb, int errno,
 495                                      char nbuf[16])
 496 {
 497         char *errstr = NULL;
 498
 499         switch (errno) {
 500         case -EIO:
 501                 errstr = "IO failure";
 502                 break;
 503         case -ENOMEM:
 504                 errstr = "Out of memory";
 505                 break;
 506         case -EROFS:
 507                 if (!sb || (EXT4_SB(sb)->s_journal &&
 508                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 509                         errstr = "Journal has aborted";
 510                 else
 511                         errstr = "Readonly filesystem";
 512                 break;
 513         default:
 514                 /* If the caller passed in an extra buffer for unknown
 515                  * errors, textualise them now.  Else we just return
 516                  * NULL. */
 517                 if (nbuf) {
 518                         /* Check for truncated error codes... */
 519                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 520                                 errstr = nbuf;
 521                 }
 522                 break;
 523         }
 524
 525         return errstr;
 526 }
 527
 528 /* __ext4_std_error decodes expected errors from journaling functions
 529  * automatically and invokes the appropriate error response.  */
 530
 531 void __ext4_std_error(struct super_block *sb, const char *function,
 532                       unsigned int line, int errno)
 533 {
 534         char nbuf[16];
 535         const char *errstr;
 536
 537         /* Special case: if the error is EROFS, and we're not already
 538          * inside a transaction, then there's really no point in logging
 539          * an error. */
 540         if (errno == -EROFS && journal_current_handle() == NULL &&
 541             (sb->s_flags & MS_RDONLY))
 542                 return;
 543
 544         errstr = ext4_decode_error(sb, errno, nbuf);
 545         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 546                sb->s_id, function, line, errstr);
 547         save_error_info(sb, function, line);
 548
 549         ext4_handle_error(sb);
 550 }
 551
 552 /*
 553  * ext4_abort is a much stronger failure handler than ext4_error.  The
 554  * abort function may be used to deal with unrecoverable failures such
 555  * as journal IO errors or ENOMEM at a critical moment in log management.
 556  *
 557  * We unconditionally force the filesystem into an ABORT|READONLY state,
 558  * unless the error response on the fs has been set to panic in which
 559  * case we take the easy way out and panic immediately.
 560  */
 561
 562 void __ext4_abort(struct super_block *sb, const char *function,
 563                 unsigned int line, const char *fmt, ...)
 564 {
 565         va_list args;
 566
 567         save_error_info(sb, function, line);
 568         va_start(args, fmt);
 569         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 570                function, line);
 571         vprintk(fmt, args);
 572         printk("\n");
 573         va_end(args);
 574
 575         if ((sb->s_flags & MS_RDONLY) == 0) {
 576                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 577                 sb->s_flags |= MS_RDONLY;
 578                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 579                 if (EXT4_SB(sb)->s_journal)
 580                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 581                 save_error_info(sb, function, line);
 582         }
 583         if (test_opt(sb, ERRORS_PANIC))
 584                 panic("EXT4-fs panic from previous error\n");
 585 }
 586
 587 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 588 {
 589         struct va_format vaf;
 590         va_list args;
 591
 592         va_start(args, fmt);
 593         vaf.fmt = fmt;
 594         vaf.va = &args;
 595         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 596         va_end(args);
 597 }
 598
 599 void __ext4_warning(struct super_block *sb, const char *function,
 600                     unsigned int line, const char *fmt, ...)
 601 {
 602         struct va_format vaf;
 603         va_list args;
 604
 605         va_start(args, fmt);
 606         vaf.fmt = fmt;
 607         vaf.va = &args;
 608         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 609                sb->s_id, function, line, &vaf);
 610         va_end(args);
 611 }
 612
 613 void __ext4_grp_locked_error(const char *function, unsigned int line,
 614                              struct super_block *sb, ext4_group_t grp,
 615                              unsigned long ino, ext4_fsblk_t block,
 616                              const char *fmt, ...)
 617 __releases(bitlock)
 618 __acquires(bitlock)
 619 {
 620         struct va_format vaf;
 621         va_list args;
 622         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 623
 624         es->s_last_error_ino = cpu_to_le32(ino);
 625         es->s_last_error_block = cpu_to_le64(block);
 626         __save_error_info(sb, function, line);
 627
 628         va_start(args, fmt);
 629
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 633                sb->s_id, function, line, grp);
 634         if (ino)
 635                 printk(KERN_CONT "inode %lu: ", ino);
 636         if (block)
 637                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 638         printk(KERN_CONT "%pV\n", &vaf);
 639         va_end(args);
 640
 641         if (test_opt(sb, ERRORS_CONT)) {
 642                 ext4_commit_super(sb, 0);
 643                 return;
 644         }
 645
 646         ext4_unlock_group(sb, grp);
 647         ext4_handle_error(sb);
 648         /*
 649          * We only get here in the ERRORS_RO case; relocking the group
 650          * may be dangerous, but nothing bad will happen since the
 651          * filesystem will have already been marked read/only and the
 652          * journal has been aborted.  We return 1 as a hint to callers
 653          * who might what to use the return value from
 654          * ext4_grp_locked_error() to distinguish between the
 655          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 656          * aggressively from the ext4 function in question, with a
 657          * more appropriate error code.
 658          */
 659         ext4_lock_group(sb, grp);
 660         return;
 661 }
 662
 663 void ext4_update_dynamic_rev(struct super_block *sb)
 664 {
 665         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 666
 667         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 668                 return;
 669
 670         ext4_warning(sb,
 671                      "updating to rev %d because of new feature flag, "
 672                      "running e2fsck is recommended",
 673                      EXT4_DYNAMIC_REV);
 674
 675         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 676         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 677         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 678         /* leave es->s_feature_*compat flags alone */
 679         /* es->s_uuid will be set by e2fsck if empty */
 680
 681         /*
 682          * The rest of the superblock fields should be zero, and if not it
 683          * means they are likely already in use, so leave them alone.  We
 684          * can leave it up to e2fsck to clean up any inconsistencies there.
 685          */
 686 }
 687
 688 /*
 689  * Open the external journal device
 690  */
 691 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 692 {
 693         struct block_device *bdev;
 694         char b[BDEVNAME_SIZE];
 695
 696         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 697         if (IS_ERR(bdev))
 698                 goto fail;
 699         return bdev;
 700
 701 fail:
 702         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 703                         __bdevname(dev, b), PTR_ERR(bdev));
 704         return NULL;
 705 }
 706
 707 /*
 708  * Release the journal device
 709  */
 710 static int ext4_blkdev_put(struct block_device *bdev)
 711 {
 712         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 713 }
 714
 715 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 716 {
 717         struct block_device *bdev;
 718         int ret = -ENODEV;
 719
 720         bdev = sbi->journal_bdev;
 721         if (bdev) {
 722                 ret = ext4_blkdev_put(bdev);
 723                 sbi->journal_bdev = NULL;
 724         }
 725         return ret;
 726 }
 727
 728 static inline struct inode *orphan_list_entry(struct list_head *l)
 729 {
 730         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 731 }
 732
 733 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 734 {
 735         struct list_head *l;
 736
 737         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 738                  le32_to_cpu(sbi->s_es->s_last_orphan));
 739
 740         printk(KERN_ERR "sb_info orphan list:\n");
 741         list_for_each(l, &sbi->s_orphan) {
 742                 struct inode *inode = orphan_list_entry(l);
 743                 printk(KERN_ERR "  "
 744                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 745                        inode->i_sb->s_id, inode->i_ino, inode,
 746                        inode->i_mode, inode->i_nlink,
 747                        NEXT_ORPHAN(inode));
 748         }
 749 }
 750
 751 static void ext4_put_super(struct super_block *sb)
 752 {
 753         struct ext4_sb_info *sbi = EXT4_SB(sb);
 754         struct ext4_super_block *es = sbi->s_es;
 755         int i, err;
 756
 757         ext4_unregister_li_request(sb);
 758         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 759
 760         flush_workqueue(sbi->dio_unwritten_wq);
 761         destroy_workqueue(sbi->dio_unwritten_wq);
 762
 763         lock_super(sb);
 764         if (sb->s_dirt)
 765                 ext4_commit_super(sb, 1);
 766
 767         if (sbi->s_journal) {
 768                 err = jbd2_journal_destroy(sbi->s_journal);
 769                 sbi->s_journal = NULL;
 770                 if (err < 0)
 771                         ext4_abort(sb, "Couldn't clean up the journal");
 772         }
 773
 774         del_timer(&sbi->s_err_report);
 775         ext4_release_system_zone(sb);
 776         ext4_mb_release(sb);
 777         ext4_ext_release(sb);
 778         ext4_xattr_put_super(sb);
 779
 780         if (!(sb->s_flags & MS_RDONLY)) {
 781                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 782                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 783                 ext4_commit_super(sb, 1);
 784         }
 785         if (sbi->s_proc) {
 786                 remove_proc_entry(sb->s_id, ext4_proc_root);
 787         }
 788         kobject_del(&sbi->s_kobj);
 789
 790         for (i = 0; i < sbi->s_gdb_count; i++)
 791                 brelse(sbi->s_group_desc[i]);
 792         kfree(sbi->s_group_desc);
 793         if (is_vmalloc_addr(sbi->s_flex_groups))
 794                 vfree(sbi->s_flex_groups);
 795         else
 796                 kfree(sbi->s_flex_groups);
 797         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 798         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 799         percpu_counter_destroy(&sbi->s_dirs_counter);
 800         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 801         brelse(sbi->s_sbh);
 802 #ifdef CONFIG_QUOTA
 803         for (i = 0; i < MAXQUOTAS; i++)
 804                 kfree(sbi->s_qf_names[i]);
 805 #endif
 806
 807         /* Debugging code just in case the in-memory inode orphan list
 808          * isn't empty.  The on-disk one can be non-empty if we've
 809          * detected an error and taken the fs readonly, but the
 810          * in-memory list had better be clean by this point. */
 811         if (!list_empty(&sbi->s_orphan))
 812                 dump_orphan_list(sb, sbi);
 813         J_ASSERT(list_empty(&sbi->s_orphan));
 814
 815         invalidate_bdev(sb->s_bdev);
 816         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 817                 /*
 818                  * Invalidate the journal device's buffers.  We don't want them
 819                  * floating about in memory - the physical journal device may
 820                  * hotswapped, and it breaks the `ro-after' testing code.
 821                  */
 822                 sync_blockdev(sbi->journal_bdev);
 823                 invalidate_bdev(sbi->journal_bdev);
 824                 ext4_blkdev_remove(sbi);
 825         }
 826         if (sbi->s_mmp_tsk)
 827                 kthread_stop(sbi->s_mmp_tsk);
 828         sb->s_fs_info = NULL;
 829         /*
 830          * Now that we are completely done shutting down the
 831          * superblock, we need to actually destroy the kobject.
 832          */
 833         unlock_super(sb);
 834         kobject_put(&sbi->s_kobj);
 835         wait_for_completion(&sbi->s_kobj_unregister);
 836         kfree(sbi->s_blockgroup_lock);
 837         kfree(sbi);
 838 }
 839
 840 static struct kmem_cache *ext4_inode_cachep;
 841
 842 /*
 843  * Called inside transaction, so use GFP_NOFS
 844  */
 845 static struct inode *ext4_alloc_inode(struct super_block *sb)
 846 {
 847         struct ext4_inode_info *ei;
 848
 849         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 850         if (!ei)
 851                 return NULL;
 852
 853         ei->vfs_inode.i_version = 1;
 854         ei->vfs_inode.i_data.writeback_index = 0;
 855         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 856         INIT_LIST_HEAD(&ei->i_prealloc_list);
 857         spin_lock_init(&ei->i_prealloc_lock);
 858         ei->i_reserved_data_blocks = 0;
 859         ei->i_reserved_meta_blocks = 0;
 860         ei->i_allocated_meta_blocks = 0;
 861         ei->i_da_metadata_calc_len = 0;
 862         spin_lock_init(&(ei->i_block_reservation_lock));
 863 #ifdef CONFIG_QUOTA
 864         ei->i_reserved_quota = 0;
 865 #endif
 866         ei->jinode = NULL;
 867         INIT_LIST_HEAD(&ei->i_completed_io_list);
 868         spin_lock_init(&ei->i_completed_io_lock);
 869         ei->cur_aio_dio = NULL;
 870         ei->i_sync_tid = 0;
 871         ei->i_datasync_tid = 0;
 872         atomic_set(&ei->i_ioend_count, 0);
 873         atomic_set(&ei->i_aiodio_unwritten, 0);
 874
 875         return &ei->vfs_inode;
 876 }
 877
 878 static int ext4_drop_inode(struct inode *inode)
 879 {
 880         int drop = generic_drop_inode(inode);
 881
 882         trace_ext4_drop_inode(inode, drop);
 883         return drop;
 884 }
 885
 886 static void ext4_i_callback(struct rcu_head *head)
 887 {
 888         struct inode *inode = container_of(head, struct inode, i_rcu);
 889         INIT_LIST_HEAD(&inode->i_dentry);
 890         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 891 }
 892
 893 static void ext4_destroy_inode(struct inode *inode)
 894 {
 895         ext4_ioend_wait(inode);
 896         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 897                 ext4_msg(inode->i_sb, KERN_ERR,
 898                          "Inode %lu (%p): orphan list check failed!",
 899                          inode->i_ino, EXT4_I(inode));
 900                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 901                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 902                                 true);
 903                 dump_stack();
 904         }
 905         call_rcu(&inode->i_rcu, ext4_i_callback);
 906 }
 907
 908 static void init_once(void *foo)
 909 {
 910         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 911
 912         INIT_LIST_HEAD(&ei->i_orphan);
 913 #ifdef CONFIG_EXT4_FS_XATTR
 914         init_rwsem(&ei->xattr_sem);
 915 #endif
 916         init_rwsem(&ei->i_data_sem);
 917         inode_init_once(&ei->vfs_inode);
 918 }
 919
 920 static int init_inodecache(void)
 921 {
 922         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 923                                              sizeof(struct ext4_inode_info),
 924                                              0, (SLAB_RECLAIM_ACCOUNT|
 925                                                 SLAB_MEM_SPREAD),
 926                                              init_once);
 927         if (ext4_inode_cachep == NULL)
 928                 return -ENOMEM;
 929         return 0;
 930 }
 931
 932 static void destroy_inodecache(void)
 933 {
 934         kmem_cache_destroy(ext4_inode_cachep);
 935 }
 936
 937 void ext4_clear_inode(struct inode *inode)
 938 {
 939         invalidate_inode_buffers(inode);
 940         end_writeback(inode);
 941         dquot_drop(inode);
 942         ext4_discard_preallocations(inode);
 943         if (EXT4_I(inode)->jinode) {
 944                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 945                                                EXT4_I(inode)->jinode);
 946                 jbd2_free_inode(EXT4_I(inode)->jinode);
 947                 EXT4_I(inode)->jinode = NULL;
 948         }
 949 }
 950
 951 static inline void ext4_show_quota_options(struct seq_file *seq,
 952                                            struct super_block *sb)
 953 {
 954 #if defined(CONFIG_QUOTA)
 955         struct ext4_sb_info *sbi = EXT4_SB(sb);
 956
 957         if (sbi->s_jquota_fmt) {
 958                 char *fmtname = "";
 959
 960                 switch (sbi->s_jquota_fmt) {
 961                 case QFMT_VFS_OLD:
 962                         fmtname = "vfsold";
 963                         break;
 964                 case QFMT_VFS_V0:
 965                         fmtname = "vfsv0";
 966                         break;
 967                 case QFMT_VFS_V1:
 968                         fmtname = "vfsv1";
 969                         break;
 970                 }
 971                 seq_printf(seq, ",jqfmt=%s", fmtname);
 972         }
 973
 974         if (sbi->s_qf_names[USRQUOTA])
 975                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 976
 977         if (sbi->s_qf_names[GRPQUOTA])
 978                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 979
 980         if (test_opt(sb, USRQUOTA))
 981                 seq_puts(seq, ",usrquota");
 982
 983         if (test_opt(sb, GRPQUOTA))
 984                 seq_puts(seq, ",grpquota");
 985 #endif
 986 }
 987
 988 /*
 989  * Show an option if
 990  *  - it's set to a non-default value OR
 991  *  - if the per-sb default is different from the global default
 992  */
 993 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 994 {
 995         int def_errors;
 996         unsigned long def_mount_opts;
 997         struct super_block *sb = vfs->mnt_sb;
 998         struct ext4_sb_info *sbi = EXT4_SB(sb);
 999         struct ext4_super_block *es = sbi->s_es;
1000
1001         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1002         def_errors     = le16_to_cpu(es->s_errors);
1003
1004         if (sbi->s_sb_block != 1)
1005                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1006         if (test_opt(sb, MINIX_DF))
1007                 seq_puts(seq, ",minixdf");
1008         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1009                 seq_puts(seq, ",grpid");
1010         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1011                 seq_puts(seq, ",nogrpid");
1012         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1013             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1014                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1015         }
1016         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1017             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1018                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1019         }
1020         if (test_opt(sb, ERRORS_RO)) {
1021                 if (def_errors == EXT4_ERRORS_PANIC ||
1022                     def_errors == EXT4_ERRORS_CONTINUE) {
1023                         seq_puts(seq, ",errors=remount-ro");
1024                 }
1025         }
1026         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1027                 seq_puts(seq, ",errors=continue");
1028         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1029                 seq_puts(seq, ",errors=panic");
1030         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1031                 seq_puts(seq, ",nouid32");
1032         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1033                 seq_puts(seq, ",debug");
1034         if (test_opt(sb, OLDALLOC))
1035                 seq_puts(seq, ",oldalloc");
1036 #ifdef CONFIG_EXT4_FS_XATTR
1037         if (test_opt(sb, XATTR_USER))
1038                 seq_puts(seq, ",user_xattr");
1039         if (!test_opt(sb, XATTR_USER))
1040                 seq_puts(seq, ",nouser_xattr");
1041 #endif
1042 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1043         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1044                 seq_puts(seq, ",acl");
1045         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1046                 seq_puts(seq, ",noacl");
1047 #endif
1048         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1049                 seq_printf(seq, ",commit=%u",
1050                            (unsigned) (sbi->s_commit_interval / HZ));
1051         }
1052         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1053                 seq_printf(seq, ",min_batch_time=%u",
1054                            (unsigned) sbi->s_min_batch_time);
1055         }
1056         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1057                 seq_printf(seq, ",max_batch_time=%u",
1058                            (unsigned) sbi->s_min_batch_time);
1059         }
1060
1061         /*
1062          * We're changing the default of barrier mount option, so
1063          * let's always display its mount state so it's clear what its
1064          * status is.
1065          */
1066         seq_puts(seq, ",barrier=");
1067         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1068         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1069                 seq_puts(seq, ",journal_async_commit");
1070         else if (test_opt(sb, JOURNAL_CHECKSUM))
1071                 seq_puts(seq, ",journal_checksum");
1072         if (test_opt(sb, I_VERSION))
1073                 seq_puts(seq, ",i_version");
1074         if (!test_opt(sb, DELALLOC) &&
1075             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1076                 seq_puts(seq, ",nodelalloc");
1077
1078         if (!test_opt(sb, MBLK_IO_SUBMIT))
1079                 seq_puts(seq, ",nomblk_io_submit");
1080         if (sbi->s_stripe)
1081                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1082         /*
1083          * journal mode get enabled in different ways
1084          * So just print the value even if we didn't specify it
1085          */
1086         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1087                 seq_puts(seq, ",data=journal");
1088         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1089                 seq_puts(seq, ",data=ordered");
1090         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1091                 seq_puts(seq, ",data=writeback");
1092
1093         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1094                 seq_printf(seq, ",inode_readahead_blks=%u",
1095                            sbi->s_inode_readahead_blks);
1096
1097         if (test_opt(sb, DATA_ERR_ABORT))
1098                 seq_puts(seq, ",data_err=abort");
1099
1100         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1101                 seq_puts(seq, ",noauto_da_alloc");
1102
1103         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1104                 seq_puts(seq, ",discard");
1105
1106         if (test_opt(sb, NOLOAD))
1107                 seq_puts(seq, ",norecovery");
1108
1109         if (test_opt(sb, DIOREAD_NOLOCK))
1110                 seq_puts(seq, ",dioread_nolock");
1111
1112         if (test_opt(sb, BLOCK_VALIDITY) &&
1113             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1114                 seq_puts(seq, ",block_validity");
1115
1116         if (!test_opt(sb, INIT_INODE_TABLE))
1117                 seq_puts(seq, ",noinit_inode_table");
1118         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1119                 seq_printf(seq, ",init_inode_table=%u",
1120                            (unsigned) sbi->s_li_wait_mult);
1121
1122         ext4_show_quota_options(seq, sb);
1123
1124         return 0;
1125 }
1126
1127 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1128                                         u64 ino, u32 generation)
1129 {
1130         struct inode *inode;
1131
1132         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1133                 return ERR_PTR(-ESTALE);
1134         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1135                 return ERR_PTR(-ESTALE);
1136
1137         /* iget isn't really right if the inode is currently unallocated!!
1138          *
1139          * ext4_read_inode will return a bad_inode if the inode had been
1140          * deleted, so we should be safe.
1141          *
1142          * Currently we don't know the generation for parent directory, so
1143          * a generation of 0 means "accept any"
1144          */
1145         inode = ext4_iget(sb, ino);
1146         if (IS_ERR(inode))
1147                 return ERR_CAST(inode);
1148         if (generation && inode->i_generation != generation) {
1149                 iput(inode);
1150                 return ERR_PTR(-ESTALE);
1151         }
1152
1153         return inode;
1154 }
1155
1156 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1157                                         int fh_len, int fh_type)
1158 {
1159         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1160                                     ext4_nfs_get_inode);
1161 }
1162
1163 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1164                                         int fh_len, int fh_type)
1165 {
1166         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1167                                     ext4_nfs_get_inode);
1168 }
1169
1170 /*
1171  * Try to release metadata pages (indirect blocks, directories) which are
1172  * mapped via the block device.  Since these pages could have journal heads
1173  * which would prevent try_to_free_buffers() from freeing them, we must use
1174  * jbd2 layer's try_to_free_buffers() function to release them.
1175  */
1176 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1177                                  gfp_t wait)
1178 {
1179         journal_t *journal = EXT4_SB(sb)->s_journal;
1180
1181         WARN_ON(PageChecked(page));
1182         if (!page_has_buffers(page))
1183                 return 0;
1184         if (journal)
1185                 return jbd2_journal_try_to_free_buffers(journal, page,
1186                                                         wait & ~__GFP_WAIT);
1187         return try_to_free_buffers(page);
1188 }
1189
1190 #ifdef CONFIG_QUOTA
1191 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1192 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1193
1194 static int ext4_write_dquot(struct dquot *dquot);
1195 static int ext4_acquire_dquot(struct dquot *dquot);
1196 static int ext4_release_dquot(struct dquot *dquot);
1197 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1198 static int ext4_write_info(struct super_block *sb, int type);
1199 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1200                          struct path *path);
1201 static int ext4_quota_off(struct super_block *sb, int type);
1202 static int ext4_quota_on_mount(struct super_block *sb, int type);
1203 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1204                                size_t len, loff_t off);
1205 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1206                                 const char *data, size_t len, loff_t off);
1207
1208 static const struct dquot_operations ext4_quota_operations = {
1209         .get_reserved_space = ext4_get_reserved_space,
1210         .write_dquot    = ext4_write_dquot,
1211         .acquire_dquot  = ext4_acquire_dquot,
1212         .release_dquot  = ext4_release_dquot,
1213         .mark_dirty     = ext4_mark_dquot_dirty,
1214         .write_info     = ext4_write_info,
1215         .alloc_dquot    = dquot_alloc,
1216         .destroy_dquot  = dquot_destroy,
1217 };
1218
1219 static const struct quotactl_ops ext4_qctl_operations = {
1220         .quota_on       = ext4_quota_on,
1221         .quota_off      = ext4_quota_off,
1222         .quota_sync     = dquot_quota_sync,
1223         .get_info       = dquot_get_dqinfo,
1224         .set_info       = dquot_set_dqinfo,
1225         .get_dqblk      = dquot_get_dqblk,
1226         .set_dqblk      = dquot_set_dqblk
1227 };
1228 #endif
1229
1230 static const struct super_operations ext4_sops = {
1231         .alloc_inode    = ext4_alloc_inode,
1232         .destroy_inode  = ext4_destroy_inode,
1233         .write_inode    = ext4_write_inode,
1234         .dirty_inode    = ext4_dirty_inode,
1235         .drop_inode     = ext4_drop_inode,
1236         .evict_inode    = ext4_evict_inode,
1237         .put_super      = ext4_put_super,
1238         .sync_fs        = ext4_sync_fs,
1239         .freeze_fs      = ext4_freeze,
1240         .unfreeze_fs    = ext4_unfreeze,
1241         .statfs         = ext4_statfs,
1242         .remount_fs     = ext4_remount,
1243         .show_options   = ext4_show_options,
1244 #ifdef CONFIG_QUOTA
1245         .quota_read     = ext4_quota_read,
1246         .quota_write    = ext4_quota_write,
1247 #endif
1248         .bdev_try_to_free_page = bdev_try_to_free_page,
1249 };
1250
1251 static const struct super_operations ext4_nojournal_sops = {
1252         .alloc_inode    = ext4_alloc_inode,
1253         .destroy_inode  = ext4_destroy_inode,
1254         .write_inode    = ext4_write_inode,
1255         .dirty_inode    = ext4_dirty_inode,
1256         .drop_inode     = ext4_drop_inode,
1257         .evict_inode    = ext4_evict_inode,
1258         .write_super    = ext4_write_super,
1259         .put_super      = ext4_put_super,
1260         .statfs         = ext4_statfs,
1261         .remount_fs     = ext4_remount,
1262         .show_options   = ext4_show_options,
1263 #ifdef CONFIG_QUOTA
1264         .quota_read     = ext4_quota_read,
1265         .quota_write    = ext4_quota_write,
1266 #endif
1267         .bdev_try_to_free_page = bdev_try_to_free_page,
1268 };
1269
1270 static const struct export_operations ext4_export_ops = {
1271         .fh_to_dentry = ext4_fh_to_dentry,
1272         .fh_to_parent = ext4_fh_to_parent,
1273         .get_parent = ext4_get_parent,
1274 };
1275
1276 enum {
1277         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1278         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1279         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1280         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1281         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1282         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1283         Opt_journal_update, Opt_journal_dev,
1284         Opt_journal_checksum, Opt_journal_async_commit,
1285         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1286         Opt_data_err_abort, Opt_data_err_ignore,
1287         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1288         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1289         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1290         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1291         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1292         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1293         Opt_inode_readahead_blks, Opt_journal_ioprio,
1294         Opt_dioread_nolock, Opt_dioread_lock,
1295         Opt_discard, Opt_nodiscard,
1296         Opt_init_inode_table, Opt_noinit_inode_table,
1297 };
1298
1299 static const match_table_t tokens = {
1300         {Opt_bsd_df, "bsddf"},
1301         {Opt_minix_df, "minixdf"},
1302         {Opt_grpid, "grpid"},
1303         {Opt_grpid, "bsdgroups"},
1304         {Opt_nogrpid, "nogrpid"},
1305         {Opt_nogrpid, "sysvgroups"},
1306         {Opt_resgid, "resgid=%u"},
1307         {Opt_resuid, "resuid=%u"},
1308         {Opt_sb, "sb=%u"},
1309         {Opt_err_cont, "errors=continue"},
1310         {Opt_err_panic, "errors=panic"},
1311         {Opt_err_ro, "errors=remount-ro"},
1312         {Opt_nouid32, "nouid32"},
1313         {Opt_debug, "debug"},
1314         {Opt_oldalloc, "oldalloc"},
1315         {Opt_orlov, "orlov"},
1316         {Opt_user_xattr, "user_xattr"},
1317         {Opt_nouser_xattr, "nouser_xattr"},
1318         {Opt_acl, "acl"},
1319         {Opt_noacl, "noacl"},
1320         {Opt_noload, "noload"},
1321         {Opt_noload, "norecovery"},
1322         {Opt_nobh, "nobh"},
1323         {Opt_bh, "bh"},
1324         {Opt_commit, "commit=%u"},
1325         {Opt_min_batch_time, "min_batch_time=%u"},
1326         {Opt_max_batch_time, "max_batch_time=%u"},
1327         {Opt_journal_update, "journal=update"},
1328         {Opt_journal_dev, "journal_dev=%u"},
1329         {Opt_journal_checksum, "journal_checksum"},
1330         {Opt_journal_async_commit, "journal_async_commit"},
1331         {Opt_abort, "abort"},
1332         {Opt_data_journal, "data=journal"},
1333         {Opt_data_ordered, "data=ordered"},
1334         {Opt_data_writeback, "data=writeback"},
1335         {Opt_data_err_abort, "data_err=abort"},
1336         {Opt_data_err_ignore, "data_err=ignore"},
1337         {Opt_offusrjquota, "usrjquota="},
1338         {Opt_usrjquota, "usrjquota=%s"},
1339         {Opt_offgrpjquota, "grpjquota="},
1340         {Opt_grpjquota, "grpjquota=%s"},
1341         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1342         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1343         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1344         {Opt_grpquota, "grpquota"},
1345         {Opt_noquota, "noquota"},
1346         {Opt_quota, "quota"},
1347         {Opt_usrquota, "usrquota"},
1348         {Opt_barrier, "barrier=%u"},
1349         {Opt_barrier, "barrier"},
1350         {Opt_nobarrier, "nobarrier"},
1351         {Opt_i_version, "i_version"},
1352         {Opt_stripe, "stripe=%u"},
1353         {Opt_resize, "resize"},
1354         {Opt_delalloc, "delalloc"},
1355         {Opt_nodelalloc, "nodelalloc"},
1356         {Opt_mblk_io_submit, "mblk_io_submit"},
1357         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1358         {Opt_block_validity, "block_validity"},
1359         {Opt_noblock_validity, "noblock_validity"},
1360         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1361         {Opt_journal_ioprio, "journal_ioprio=%u"},
1362         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1363         {Opt_auto_da_alloc, "auto_da_alloc"},
1364         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1365         {Opt_dioread_nolock, "dioread_nolock"},
1366         {Opt_dioread_lock, "dioread_lock"},
1367         {Opt_discard, "discard"},
1368         {Opt_nodiscard, "nodiscard"},
1369         {Opt_init_inode_table, "init_itable=%u"},
1370         {Opt_init_inode_table, "init_itable"},
1371         {Opt_noinit_inode_table, "noinit_itable"},
1372         {Opt_err, NULL},
1373 };
1374
1375 static ext4_fsblk_t get_sb_block(void **data)
1376 {
1377         ext4_fsblk_t    sb_block;
1378         char            *options = (char *) *data;
1379
1380         if (!options || strncmp(options, "sb=", 3) != 0)
1381                 return 1;       /* Default location */
1382
1383         options += 3;
1384         /* TODO: use simple_strtoll with >32bit ext4 */
1385         sb_block = simple_strtoul(options, &options, 0);
1386         if (*options && *options != ',') {
1387                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1388                        (char *) *data);
1389                 return 1;
1390         }
1391         if (*options == ',')
1392                 options++;
1393         *data = (void *) options;
1394
1395         return sb_block;
1396 }
1397
1398 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1399 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1400         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1401
1402 #ifdef CONFIG_QUOTA
1403 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1404 {
1405         struct ext4_sb_info *sbi = EXT4_SB(sb);
1406         char *qname;
1407
1408         if (sb_any_quota_loaded(sb) &&
1409                 !sbi->s_qf_names[qtype]) {
1410                 ext4_msg(sb, KERN_ERR,
1411                         "Cannot change journaled "
1412                         "quota options when quota turned on");
1413                 return 0;
1414         }
1415         qname = match_strdup(args);
1416         if (!qname) {
1417                 ext4_msg(sb, KERN_ERR,
1418                         "Not enough memory for storing quotafile name");
1419                 return 0;
1420         }
1421         if (sbi->s_qf_names[qtype] &&
1422                 strcmp(sbi->s_qf_names[qtype], qname)) {
1423                 ext4_msg(sb, KERN_ERR,
1424                         "%s quota file already specified", QTYPE2NAME(qtype));
1425                 kfree(qname);
1426                 return 0;
1427         }
1428         sbi->s_qf_names[qtype] = qname;
1429         if (strchr(sbi->s_qf_names[qtype], '/')) {
1430                 ext4_msg(sb, KERN_ERR,
1431                         "quotafile must be on filesystem root");
1432                 kfree(sbi->s_qf_names[qtype]);
1433                 sbi->s_qf_names[qtype] = NULL;
1434                 return 0;
1435         }
1436         set_opt(sb, QUOTA);
1437         return 1;
1438 }
1439
1440 static int clear_qf_name(struct super_block *sb, int qtype)
1441 {
1442
1443         struct ext4_sb_info *sbi = EXT4_SB(sb);
1444
1445         if (sb_any_quota_loaded(sb) &&
1446                 sbi->s_qf_names[qtype]) {
1447                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1448                         " when quota turned on");
1449                 return 0;
1450         }
1451         /*
1452          * The space will be released later when all options are confirmed
1453          * to be correct
1454          */
1455         sbi->s_qf_names[qtype] = NULL;
1456         return 1;
1457 }
1458 #endif
1459
1460 static int parse_options(char *options, struct super_block *sb,
1461                          unsigned long *journal_devnum,
1462                          unsigned int *journal_ioprio,
1463                          ext4_fsblk_t *n_blocks_count, int is_remount)
1464 {
1465         struct ext4_sb_info *sbi = EXT4_SB(sb);
1466         char *p;
1467         substring_t args[MAX_OPT_ARGS];
1468         int data_opt = 0;
1469         int option;
1470 #ifdef CONFIG_QUOTA
1471         int qfmt;
1472 #endif
1473
1474         if (!options)
1475                 return 1;
1476
1477         while ((p = strsep(&options, ",")) != NULL) {
1478                 int token;
1479                 if (!*p)
1480                         continue;
1481
1482                 /*
1483                  * Initialize args struct so we know whether arg was
1484                  * found; some options take optional arguments.
1485                  */
1486                 args[0].to = args[0].from = NULL;
1487                 token = match_token(p, tokens, args);
1488                 switch (token) {
1489                 case Opt_bsd_df:
1490                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1491                         clear_opt(sb, MINIX_DF);
1492                         break;
1493                 case Opt_minix_df:
1494                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1495                         set_opt(sb, MINIX_DF);
1496
1497                         break;
1498                 case Opt_grpid:
1499                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1500                         set_opt(sb, GRPID);
1501
1502                         break;
1503                 case Opt_nogrpid:
1504                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1505                         clear_opt(sb, GRPID);
1506
1507                         break;
1508                 case Opt_resuid:
1509                         if (match_int(&args[0], &option))
1510                                 return 0;
1511                         sbi->s_resuid = option;
1512                         break;
1513                 case Opt_resgid:
1514                         if (match_int(&args[0], &option))
1515                                 return 0;
1516                         sbi->s_resgid = option;
1517                         break;
1518                 case Opt_sb:
1519                         /* handled by get_sb_block() instead of here */
1520                         /* *sb_block = match_int(&args[0]); */
1521                         break;
1522                 case Opt_err_panic:
1523                         clear_opt(sb, ERRORS_CONT);
1524                         clear_opt(sb, ERRORS_RO);
1525                         set_opt(sb, ERRORS_PANIC);
1526                         break;
1527                 case Opt_err_ro:
1528                         clear_opt(sb, ERRORS_CONT);
1529                         clear_opt(sb, ERRORS_PANIC);
1530                         set_opt(sb, ERRORS_RO);
1531                         break;
1532                 case Opt_err_cont:
1533                         clear_opt(sb, ERRORS_RO);
1534                         clear_opt(sb, ERRORS_PANIC);
1535                         set_opt(sb, ERRORS_CONT);
1536                         break;
1537                 case Opt_nouid32:
1538                         set_opt(sb, NO_UID32);
1539                         break;
1540                 case Opt_debug:
1541                         set_opt(sb, DEBUG);
1542                         break;
1543                 case Opt_oldalloc:
1544                         set_opt(sb, OLDALLOC);
1545                         break;
1546                 case Opt_orlov:
1547                         clear_opt(sb, OLDALLOC);
1548                         break;
1549 #ifdef CONFIG_EXT4_FS_XATTR
1550                 case Opt_user_xattr:
1551                         set_opt(sb, XATTR_USER);
1552                         break;
1553                 case Opt_nouser_xattr:
1554                         clear_opt(sb, XATTR_USER);
1555                         break;
1556 #else
1557                 case Opt_user_xattr:
1558                 case Opt_nouser_xattr:
1559                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1560                         break;
1561 #endif
1562 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1563                 case Opt_acl:
1564                         set_opt(sb, POSIX_ACL);
1565                         break;
1566                 case Opt_noacl:
1567                         clear_opt(sb, POSIX_ACL);
1568                         break;
1569 #else
1570                 case Opt_acl:
1571                 case Opt_noacl:
1572                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1573                         break;
1574 #endif
1575                 case Opt_journal_update:
1576                         /* @@@ FIXME */
1577                         /* Eventually we will want to be able to create
1578                            a journal file here.  For now, only allow the
1579                            user to specify an existing inode to be the
1580                            journal file. */
1581                         if (is_remount) {
1582                                 ext4_msg(sb, KERN_ERR,
1583                                          "Cannot specify journal on remount");
1584                                 return 0;
1585                         }
1586                         set_opt(sb, UPDATE_JOURNAL);
1587                         break;
1588                 case Opt_journal_dev:
1589                         if (is_remount) {
1590                                 ext4_msg(sb, KERN_ERR,
1591                                         "Cannot specify journal on remount");
1592                                 return 0;
1593                         }
1594                         if (match_int(&args[0], &option))
1595                                 return 0;
1596                         *journal_devnum = option;
1597                         break;
1598                 case Opt_journal_checksum:
1599                         set_opt(sb, JOURNAL_CHECKSUM);
1600                         break;
1601                 case Opt_journal_async_commit:
1602                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1603                         set_opt(sb, JOURNAL_CHECKSUM);
1604                         break;
1605                 case Opt_noload:
1606                         set_opt(sb, NOLOAD);
1607                         break;
1608                 case Opt_commit:
1609                         if (match_int(&args[0], &option))
1610                                 return 0;
1611                         if (option < 0)
1612                                 return 0;
1613                         if (option == 0)
1614                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1615                         sbi->s_commit_interval = HZ * option;
1616                         break;
1617                 case Opt_max_batch_time:
1618                         if (match_int(&args[0], &option))
1619                                 return 0;
1620                         if (option < 0)
1621                                 return 0;
1622                         if (option == 0)
1623                                 option = EXT4_DEF_MAX_BATCH_TIME;
1624                         sbi->s_max_batch_time = option;
1625                         break;
1626                 case Opt_min_batch_time:
1627                         if (match_int(&args[0], &option))
1628                                 return 0;
1629                         if (option < 0)
1630                                 return 0;
1631                         sbi->s_min_batch_time = option;
1632                         break;
1633                 case Opt_data_journal:
1634                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1635                         goto datacheck;
1636                 case Opt_data_ordered:
1637                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1638                         goto datacheck;
1639                 case Opt_data_writeback:
1640                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1641                 datacheck:
1642                         if (is_remount) {
1643                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1644                                         ext4_msg(sb, KERN_ERR,
1645                                                 "Cannot change data mode on remount");
1646                                         return 0;
1647                                 }
1648                         } else {
1649                                 clear_opt(sb, DATA_FLAGS);
1650                                 sbi->s_mount_opt |= data_opt;
1651                         }
1652                         break;
1653                 case Opt_data_err_abort:
1654                         set_opt(sb, DATA_ERR_ABORT);
1655                         break;
1656                 case Opt_data_err_ignore:
1657                         clear_opt(sb, DATA_ERR_ABORT);
1658                         break;
1659 #ifdef CONFIG_QUOTA
1660                 case Opt_usrjquota:
1661                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1662                                 return 0;
1663                         break;
1664                 case Opt_grpjquota:
1665                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1666                                 return 0;
1667                         break;
1668                 case Opt_offusrjquota:
1669                         if (!clear_qf_name(sb, USRQUOTA))
1670                                 return 0;
1671                         break;
1672                 case Opt_offgrpjquota:
1673                         if (!clear_qf_name(sb, GRPQUOTA))
1674                                 return 0;
1675                         break;
1676
1677                 case Opt_jqfmt_vfsold:
1678                         qfmt = QFMT_VFS_OLD;
1679                         goto set_qf_format;
1680                 case Opt_jqfmt_vfsv0:
1681                         qfmt = QFMT_VFS_V0;
1682                         goto set_qf_format;
1683                 case Opt_jqfmt_vfsv1:
1684                         qfmt = QFMT_VFS_V1;
1685 set_qf_format:
1686                         if (sb_any_quota_loaded(sb) &&
1687                             sbi->s_jquota_fmt != qfmt) {
1688                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1689                                         "journaled quota options when "
1690                                         "quota turned on");
1691                                 return 0;
1692                         }
1693                         sbi->s_jquota_fmt = qfmt;
1694                         break;
1695                 case Opt_quota:
1696                 case Opt_usrquota:
1697                         set_opt(sb, QUOTA);
1698                         set_opt(sb, USRQUOTA);
1699                         break;
1700                 case Opt_grpquota:
1701                         set_opt(sb, QUOTA);
1702                         set_opt(sb, GRPQUOTA);
1703                         break;
1704                 case Opt_noquota:
1705                         if (sb_any_quota_loaded(sb)) {
1706                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1707                                         "options when quota turned on");
1708                                 return 0;
1709                         }
1710                         clear_opt(sb, QUOTA);
1711                         clear_opt(sb, USRQUOTA);
1712                         clear_opt(sb, GRPQUOTA);
1713                         break;
1714 #else
1715                 case Opt_quota:
1716                 case Opt_usrquota:
1717                 case Opt_grpquota:
1718                         ext4_msg(sb, KERN_ERR,
1719                                 "quota options not supported");
1720                         break;
1721                 case Opt_usrjquota:
1722                 case Opt_grpjquota:
1723                 case Opt_offusrjquota:
1724                 case Opt_offgrpjquota:
1725                 case Opt_jqfmt_vfsold:
1726                 case Opt_jqfmt_vfsv0:
1727                 case Opt_jqfmt_vfsv1:
1728                         ext4_msg(sb, KERN_ERR,
1729                                 "journaled quota options not supported");
1730                         break;
1731                 case Opt_noquota:
1732                         break;
1733 #endif
1734                 case Opt_abort:
1735                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1736                         break;
1737                 case Opt_nobarrier:
1738                         clear_opt(sb, BARRIER);
1739                         break;
1740                 case Opt_barrier:
1741                         if (args[0].from) {
1742                                 if (match_int(&args[0], &option))
1743                                         return 0;
1744                         } else
1745                                 option = 1;     /* No argument, default to 1 */
1746                         if (option)
1747                                 set_opt(sb, BARRIER);
1748                         else
1749                                 clear_opt(sb, BARRIER);
1750                         break;
1751                 case Opt_ignore:
1752                         break;
1753                 case Opt_resize:
1754                         if (!is_remount) {
1755                                 ext4_msg(sb, KERN_ERR,
1756                                         "resize option only available "
1757                                         "for remount");
1758                                 return 0;
1759                         }
1760                         if (match_int(&args[0], &option) != 0)
1761                                 return 0;
1762                         *n_blocks_count = option;
1763                         break;
1764                 case Opt_nobh:
1765                         ext4_msg(sb, KERN_WARNING,
1766                                  "Ignoring deprecated nobh option");
1767                         break;
1768                 case Opt_bh:
1769                         ext4_msg(sb, KERN_WARNING,
1770                                  "Ignoring deprecated bh option");
1771                         break;
1772                 case Opt_i_version:
1773                         set_opt(sb, I_VERSION);
1774                         sb->s_flags |= MS_I_VERSION;
1775                         break;
1776                 case Opt_nodelalloc:
1777                         clear_opt(sb, DELALLOC);
1778                         break;
1779                 case Opt_mblk_io_submit:
1780                         set_opt(sb, MBLK_IO_SUBMIT);
1781                         break;
1782                 case Opt_nomblk_io_submit:
1783                         clear_opt(sb, MBLK_IO_SUBMIT);
1784                         break;
1785                 case Opt_stripe:
1786                         if (match_int(&args[0], &option))
1787                                 return 0;
1788                         if (option < 0)
1789                                 return 0;
1790                         sbi->s_stripe = option;
1791                         break;
1792                 case Opt_delalloc:
1793                         set_opt(sb, DELALLOC);
1794                         break;
1795                 case Opt_block_validity:
1796                         set_opt(sb, BLOCK_VALIDITY);
1797                         break;
1798                 case Opt_noblock_validity:
1799                         clear_opt(sb, BLOCK_VALIDITY);
1800                         break;
1801                 case Opt_inode_readahead_blks:
1802                         if (match_int(&args[0], &option))
1803                                 return 0;
1804                         if (option < 0 || option > (1 << 30))
1805                                 return 0;
1806                         if (option && !is_power_of_2(option)) {
1807                                 ext4_msg(sb, KERN_ERR,
1808                                          "EXT4-fs: inode_readahead_blks"
1809                                          " must be a power of 2");
1810                                 return 0;
1811                         }
1812                         sbi->s_inode_readahead_blks = option;
1813                         break;
1814                 case Opt_journal_ioprio:
1815                         if (match_int(&args[0], &option))
1816                                 return 0;
1817                         if (option < 0 || option > 7)
1818                                 break;
1819                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1820                                                             option);
1821                         break;
1822                 case Opt_noauto_da_alloc:
1823                         set_opt(sb, NO_AUTO_DA_ALLOC);
1824                         break;
1825                 case Opt_auto_da_alloc:
1826                         if (args[0].from) {
1827                                 if (match_int(&args[0], &option))
1828                                         return 0;
1829                         } else
1830                                 option = 1;     /* No argument, default to 1 */
1831                         if (option)
1832                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1833                         else
1834                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1835                         break;
1836                 case Opt_discard:
1837                         set_opt(sb, DISCARD);
1838                         break;
1839                 case Opt_nodiscard:
1840                         clear_opt(sb, DISCARD);
1841                         break;
1842                 case Opt_dioread_nolock:
1843                         set_opt(sb, DIOREAD_NOLOCK);
1844                         break;
1845                 case Opt_dioread_lock:
1846                         clear_opt(sb, DIOREAD_NOLOCK);
1847                         break;
1848                 case Opt_init_inode_table:
1849                         set_opt(sb, INIT_INODE_TABLE);
1850                         if (args[0].from) {
1851                                 if (match_int(&args[0], &option))
1852                                         return 0;
1853                         } else
1854                                 option = EXT4_DEF_LI_WAIT_MULT;
1855                         if (option < 0)
1856                                 return 0;
1857                         sbi->s_li_wait_mult = option;
1858                         break;
1859                 case Opt_noinit_inode_table:
1860                         clear_opt(sb, INIT_INODE_TABLE);
1861                         break;
1862                 default:
1863                         ext4_msg(sb, KERN_ERR,
1864                                "Unrecognized mount option \"%s\" "
1865                                "or missing value", p);
1866                         return 0;
1867                 }
1868         }
1869 #ifdef CONFIG_QUOTA
1870         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1871                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1872                         clear_opt(sb, USRQUOTA);
1873
1874                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1875                         clear_opt(sb, GRPQUOTA);
1876
1877                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1878                         ext4_msg(sb, KERN_ERR, "old and new quota "
1879                                         "format mixing");
1880                         return 0;
1881                 }
1882
1883                 if (!sbi->s_jquota_fmt) {
1884                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1885                                         "not specified");
1886                         return 0;
1887                 }
1888         } else {
1889                 if (sbi->s_jquota_fmt) {
1890                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1891                                         "specified with no journaling "
1892                                         "enabled");
1893                         return 0;
1894                 }
1895         }
1896 #endif
1897         return 1;
1898 }
1899
1900 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1901                             int read_only)
1902 {
1903         struct ext4_sb_info *sbi = EXT4_SB(sb);
1904         int res = 0;
1905
1906         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1907                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1908                          "forcing read-only mode");
1909                 res = MS_RDONLY;
1910         }
1911         if (read_only)
1912                 return res;
1913         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1914                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1915                          "running e2fsck is recommended");
1916         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1917                 ext4_msg(sb, KERN_WARNING,
1918                          "warning: mounting fs with errors, "
1919                          "running e2fsck is recommended");
1920         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1921                  le16_to_cpu(es->s_mnt_count) >=
1922                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1923                 ext4_msg(sb, KERN_WARNING,
1924                          "warning: maximal mount count reached, "
1925                          "running e2fsck is recommended");
1926         else if (le32_to_cpu(es->s_checkinterval) &&
1927                 (le32_to_cpu(es->s_lastcheck) +
1928                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1929                 ext4_msg(sb, KERN_WARNING,
1930                          "warning: checktime reached, "
1931                          "running e2fsck is recommended");
1932         if (!sbi->s_journal)
1933                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1934         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1935                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1936         le16_add_cpu(&es->s_mnt_count, 1);
1937         es->s_mtime = cpu_to_le32(get_seconds());
1938         ext4_update_dynamic_rev(sb);
1939         if (sbi->s_journal)
1940                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1941
1942         ext4_commit_super(sb, 1);
1943         if (test_opt(sb, DEBUG))
1944                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1945                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1946                         sb->s_blocksize,
1947                         sbi->s_groups_count,
1948                         EXT4_BLOCKS_PER_GROUP(sb),
1949                         EXT4_INODES_PER_GROUP(sb),
1950                         sbi->s_mount_opt, sbi->s_mount_opt2);
1951
1952         cleancache_init_fs(sb);
1953         return res;
1954 }
1955
1956 static int ext4_fill_flex_info(struct super_block *sb)
1957 {
1958         struct ext4_sb_info *sbi = EXT4_SB(sb);
1959         struct ext4_group_desc *gdp = NULL;
1960         ext4_group_t flex_group_count;
1961         ext4_group_t flex_group;
1962         int groups_per_flex = 0;
1963         size_t size;
1964         int i;
1965
1966         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1967         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1968
1969         if (groups_per_flex < 2) {
1970                 sbi->s_log_groups_per_flex = 0;
1971                 return 1;
1972         }
1973
1974         /* We allocate both existing and potentially added groups */
1975         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1976                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1977                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1978         size = flex_group_count * sizeof(struct flex_groups);
1979         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1980         if (sbi->s_flex_groups == NULL) {
1981                 sbi->s_flex_groups = vzalloc(size);
1982                 if (sbi->s_flex_groups == NULL) {
1983                         ext4_msg(sb, KERN_ERR,
1984                                  "not enough memory for %u flex groups",
1985                                  flex_group_count);
1986                         goto failed;
1987                 }
1988         }
1989
1990         for (i = 0; i < sbi->s_groups_count; i++) {
1991                 gdp = ext4_get_group_desc(sb, i, NULL);
1992
1993                 flex_group = ext4_flex_group(sbi, i);
1994                 atomic_add(ext4_free_inodes_count(sb, gdp),
1995                            &sbi->s_flex_groups[flex_group].free_inodes);
1996                 atomic_add(ext4_free_blks_count(sb, gdp),
1997                            &sbi->s_flex_groups[flex_group].free_blocks);
1998                 atomic_add(ext4_used_dirs_count(sb, gdp),
1999                            &sbi->s_flex_groups[flex_group].used_dirs);
2000         }
2001
2002         return 1;
2003 failed:
2004         return 0;
2005 }
2006
2007 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2008                             struct ext4_group_desc *gdp)
2009 {
2010         __u16 crc = 0;
2011
2012         if (sbi->s_es->s_feature_ro_compat &
2013             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2014                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2015                 __le32 le_group = cpu_to_le32(block_group);
2016
2017                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2018                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2019                 crc = crc16(crc, (__u8 *)gdp, offset);
2020                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2021                 /* for checksum of struct ext4_group_desc do the rest...*/
2022                 if ((sbi->s_es->s_feature_incompat &
2023                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2024                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2025                         crc = crc16(crc, (__u8 *)gdp + offset,
2026                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2027                                         offset);
2028         }
2029
2030         return cpu_to_le16(crc);
2031 }
2032
2033 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2034                                 struct ext4_group_desc *gdp)
2035 {
2036         if ((sbi->s_es->s_feature_ro_compat &
2037              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2038             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2039                 return 0;
2040
2041         return 1;
2042 }
2043
2044 /* Called at mount-time, super-block is locked */
2045 static int ext4_check_descriptors(struct super_block *sb,
2046                                   ext4_group_t *first_not_zeroed)
2047 {
2048         struct ext4_sb_info *sbi = EXT4_SB(sb);
2049         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2050         ext4_fsblk_t last_block;
2051         ext4_fsblk_t block_bitmap;
2052         ext4_fsblk_t inode_bitmap;
2053         ext4_fsblk_t inode_table;
2054         int flexbg_flag = 0;
2055         ext4_group_t i, grp = sbi->s_groups_count;
2056
2057         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2058                 flexbg_flag = 1;
2059
2060         ext4_debug("Checking group descriptors");
2061
2062         for (i = 0; i < sbi->s_groups_count; i++) {
2063                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2064
2065                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2066                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2067                 else
2068                         last_block = first_block +
2069                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2070
2071                 if ((grp == sbi->s_groups_count) &&
2072                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2073                         grp = i;
2074
2075                 block_bitmap = ext4_block_bitmap(sb, gdp);
2076                 if (block_bitmap < first_block || block_bitmap > last_block) {
2077                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2078                                "Block bitmap for group %u not in group "
2079                                "(block %llu)!", i, block_bitmap);
2080                         return 0;
2081                 }
2082                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2083                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2084                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2085                                "Inode bitmap for group %u not in group "
2086                                "(block %llu)!", i, inode_bitmap);
2087                         return 0;
2088                 }
2089                 inode_table = ext4_inode_table(sb, gdp);
2090                 if (inode_table < first_block ||
2091                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2092                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2093                                "Inode table for group %u not in group "
2094                                "(block %llu)!", i, inode_table);
2095                         return 0;
2096                 }
2097                 ext4_lock_group(sb, i);
2098                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2099                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2100                                  "Checksum for group %u failed (%u!=%u)",
2101                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2102                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2103                         if (!(sb->s_flags & MS_RDONLY)) {
2104                                 ext4_unlock_group(sb, i);
2105                                 return 0;
2106                         }
2107                 }
2108                 ext4_unlock_group(sb, i);
2109                 if (!flexbg_flag)
2110                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2111         }
2112         if (NULL != first_not_zeroed)
2113                 *first_not_zeroed = grp;
2114
2115         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2116         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2117         return 1;
2118 }
2119
2120 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2121  * the superblock) which were deleted from all directories, but held open by
2122  * a process at the time of a crash.  We walk the list and try to delete these
2123  * inodes at recovery time (only with a read-write filesystem).
2124  *
2125  * In order to keep the orphan inode chain consistent during traversal (in
2126  * case of crash during recovery), we link each inode into the superblock
2127  * orphan list_head and handle it the same way as an inode deletion during
2128  * normal operation (which journals the operations for us).
2129  *
2130  * We only do an iget() and an iput() on each inode, which is very safe if we
2131  * accidentally point at an in-use or already deleted inode.  The worst that
2132  * can happen in this case is that we get a "bit already cleared" message from
2133  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2134  * e2fsck was run on this filesystem, and it must have already done the orphan
2135  * inode cleanup for us, so we can safely abort without any further action.
2136  */
2137 static void ext4_orphan_cleanup(struct super_block *sb,
2138                                 struct ext4_super_block *es)
2139 {
2140         unsigned int s_flags = sb->s_flags;
2141         int nr_orphans = 0, nr_truncates = 0;
2142 #ifdef CONFIG_QUOTA
2143         int i;
2144 #endif
2145         if (!es->s_last_orphan) {
2146                 jbd_debug(4, "no orphan inodes to clean up\n");
2147                 return;
2148         }
2149
2150         if (bdev_read_only(sb->s_bdev)) {
2151                 ext4_msg(sb, KERN_ERR, "write access "
2152                         "unavailable, skipping orphan cleanup");
2153                 return;
2154         }
2155
2156         /* Check if feature set would not allow a r/w mount */
2157         if (!ext4_feature_set_ok(sb, 0)) {
2158                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2159                          "unknown ROCOMPAT features");
2160                 return;
2161         }
2162
2163         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2164                 if (es->s_last_orphan)
2165                         jbd_debug(1, "Errors on filesystem, "
2166                                   "clearing orphan list.\n");
2167                 es->s_last_orphan = 0;
2168                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2169                 return;
2170         }
2171
2172         if (s_flags & MS_RDONLY) {
2173                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2174                 sb->s_flags &= ~MS_RDONLY;
2175         }
2176 #ifdef CONFIG_QUOTA
2177         /* Needed for iput() to work correctly and not trash data */
2178         sb->s_flags |= MS_ACTIVE;
2179         /* Turn on quotas so that they are updated correctly */
2180         for (i = 0; i < MAXQUOTAS; i++) {
2181                 if (EXT4_SB(sb)->s_qf_names[i]) {
2182                         int ret = ext4_quota_on_mount(sb, i);
2183                         if (ret < 0)
2184                                 ext4_msg(sb, KERN_ERR,
2185                                         "Cannot turn on journaled "
2186                                         "quota: error %d", ret);
2187                 }
2188         }
2189 #endif
2190
2191         while (es->s_last_orphan) {
2192                 struct inode *inode;
2193
2194                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2195                 if (IS_ERR(inode)) {
2196                         es->s_last_orphan = 0;
2197                         break;
2198                 }
2199
2200                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2201                 dquot_initialize(inode);
2202                 if (inode->i_nlink) {
2203                         ext4_msg(sb, KERN_DEBUG,
2204                                 "%s: truncating inode %lu to %lld bytes",
2205                                 __func__, inode->i_ino, inode->i_size);
2206                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2207                                   inode->i_ino, inode->i_size);
2208                         ext4_truncate(inode);
2209                         nr_truncates++;
2210                 } else {
2211                         ext4_msg(sb, KERN_DEBUG,
2212                                 "%s: deleting unreferenced inode %lu",
2213                                 __func__, inode->i_ino);
2214                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2215                                   inode->i_ino);
2216                         nr_orphans++;
2217                 }
2218                 iput(inode);  /* The delete magic happens here! */
2219         }
2220
2221 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2222
2223         if (nr_orphans)
2224                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2225                        PLURAL(nr_orphans));
2226         if (nr_truncates)
2227                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2228                        PLURAL(nr_truncates));
2229 #ifdef CONFIG_QUOTA
2230         /* Turn quotas off */
2231         for (i = 0; i < MAXQUOTAS; i++) {
2232                 if (sb_dqopt(sb)->files[i])
2233                         dquot_quota_off(sb, i);
2234         }
2235 #endif
2236         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2237 }
2238
2239 /*
2240  * Maximal extent format file size.
2241  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2242  * extent format containers, within a sector_t, and within i_blocks
2243  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2244  * so that won't be a limiting factor.
2245  *
2246  * However there is other limiting factor. We do store extents in the form
2247  * of starting block and length, hence the resulting length of the extent
2248  * covering maximum file size must fit into on-disk format containers as
2249  * well. Given that length is always by 1 unit bigger than max unit (because
2250  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2251  *
2252  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2253  */
2254 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2255 {
2256         loff_t res;
2257         loff_t upper_limit = MAX_LFS_FILESIZE;
2258
2259         /* small i_blocks in vfs inode? */
2260         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2261                 /*
2262                  * CONFIG_LBDAF is not enabled implies the inode
2263                  * i_block represent total blocks in 512 bytes
2264                  * 32 == size of vfs inode i_blocks * 8
2265                  */
2266                 upper_limit = (1LL << 32) - 1;
2267
2268                 /* total blocks in file system block size */
2269                 upper_limit >>= (blkbits - 9);
2270                 upper_limit <<= blkbits;
2271         }
2272
2273         /*
2274          * 32-bit extent-start container, ee_block. We lower the maxbytes
2275          * by one fs block, so ee_len can cover the extent of maximum file
2276          * size
2277          */
2278         res = (1LL << 32) - 1;
2279         res <<= blkbits;
2280
2281         /* Sanity check against vm- & vfs- imposed limits */
2282         if (res > upper_limit)
2283                 res = upper_limit;
2284
2285         return res;
2286 }
2287
2288 /*
2289  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2290  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2291  * We need to be 1 filesystem block less than the 2^48 sector limit.
2292  */
2293 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2294 {
2295         loff_t res = EXT4_NDIR_BLOCKS;
2296         int meta_blocks;
2297         loff_t upper_limit;
2298         /* This is calculated to be the largest file size for a dense, block
2299          * mapped file such that the file's total number of 512-byte sectors,
2300          * including data and all indirect blocks, does not exceed (2^48 - 1).
2301          *
2302          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2303          * number of 512-byte sectors of the file.
2304          */
2305
2306         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2307                 /*
2308                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2309                  * the inode i_block field represents total file blocks in
2310                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2311                  */
2312                 upper_limit = (1LL << 32) - 1;
2313
2314                 /* total blocks in file system block size */
2315                 upper_limit >>= (bits - 9);
2316
2317         } else {
2318                 /*
2319                  * We use 48 bit ext4_inode i_blocks
2320                  * With EXT4_HUGE_FILE_FL set the i_blocks
2321                  * represent total number of blocks in
2322                  * file system block size
2323                  */
2324                 upper_limit = (1LL << 48) - 1;
2325
2326         }
2327
2328         /* indirect blocks */
2329         meta_blocks = 1;
2330         /* double indirect blocks */
2331         meta_blocks += 1 + (1LL << (bits-2));
2332         /* tripple indirect blocks */
2333         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2334
2335         upper_limit -= meta_blocks;
2336         upper_limit <<= bits;
2337
2338         res += 1LL << (bits-2);
2339         res += 1LL << (2*(bits-2));
2340         res += 1LL << (3*(bits-2));
2341         res <<= bits;
2342         if (res > upper_limit)
2343                 res = upper_limit;
2344
2345         if (res > MAX_LFS_FILESIZE)
2346                 res = MAX_LFS_FILESIZE;
2347
2348         return res;
2349 }
2350
2351 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2352                                    ext4_fsblk_t logical_sb_block, int nr)
2353 {
2354         struct ext4_sb_info *sbi = EXT4_SB(sb);
2355         ext4_group_t bg, first_meta_bg;
2356         int has_super = 0;
2357
2358         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2359
2360         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2361             nr < first_meta_bg)
2362                 return logical_sb_block + nr + 1;
2363         bg = sbi->s_desc_per_block * nr;
2364         if (ext4_bg_has_super(sb, bg))
2365                 has_super = 1;
2366
2367         return (has_super + ext4_group_first_block_no(sb, bg));
2368 }
2369
2370 /**
2371  * ext4_get_stripe_size: Get the stripe size.
2372  * @sbi: In memory super block info
2373  *
2374  * If we have specified it via mount option, then
2375  * use the mount option value. If the value specified at mount time is
2376  * greater than the blocks per group use the super block value.
2377  * If the super block value is greater than blocks per group return 0.
2378  * Allocator needs it be less than blocks per group.
2379  *
2380  */
2381 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2382 {
2383         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2384         unsigned long stripe_width =
2385                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2386
2387         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2388                 return sbi->s_stripe;
2389
2390         if (stripe_width <= sbi->s_blocks_per_group)
2391                 return stripe_width;
2392
2393         if (stride <= sbi->s_blocks_per_group)
2394                 return stride;
2395
2396         return 0;
2397 }
2398
2399 /* sysfs supprt */
2400
2401 struct ext4_attr {
2402         struct attribute attr;
2403         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2404         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2405                          const char *, size_t);
2406         int offset;
2407 };
2408
2409 static int parse_strtoul(const char *buf,
2410                 unsigned long max, unsigned long *value)
2411 {
2412         char *endp;
2413
2414         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2415         endp = skip_spaces(endp);
2416         if (*endp || *value > max)
2417                 return -EINVAL;
2418
2419         return 0;
2420 }
2421
2422 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2423                                               struct ext4_sb_info *sbi,
2424                                               char *buf)
2425 {
2426         return snprintf(buf, PAGE_SIZE, "%llu\n",
2427                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2428 }
2429
2430 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2431                                          struct ext4_sb_info *sbi, char *buf)
2432 {
2433         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2434
2435         if (!sb->s_bdev->bd_part)
2436                 return snprintf(buf, PAGE_SIZE, "0\n");
2437         return snprintf(buf, PAGE_SIZE, "%lu\n",
2438                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2439                          sbi->s_sectors_written_start) >> 1);
2440 }
2441
2442 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2443                                           struct ext4_sb_info *sbi, char *buf)
2444 {
2445         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2446
2447         if (!sb->s_bdev->bd_part)
2448                 return snprintf(buf, PAGE_SIZE, "0\n");
2449         return snprintf(buf, PAGE_SIZE, "%llu\n",
2450                         (unsigned long long)(sbi->s_kbytes_written +
2451                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2452                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2453 }
2454
2455 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2456                                       struct ext4_sb_info *sbi, char *buf)
2457 {
2458         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2459 }
2460
2461 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2462                                         struct ext4_sb_info *sbi, char *buf)
2463 {
2464         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2465 }
2466
2467 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2468                                           struct ext4_sb_info *sbi,
2469                                           const char *buf, size_t count)
2470 {
2471         unsigned long t;
2472
2473         if (parse_strtoul(buf, 0x40000000, &t))
2474                 return -EINVAL;
2475
2476         if (t && !is_power_of_2(t))
2477                 return -EINVAL;
2478
2479         sbi->s_inode_readahead_blks = t;
2480         return count;
2481 }
2482
2483 static ssize_t sbi_ui_show(struct ext4_attr *a,
2484                            struct ext4_sb_info *sbi, char *buf)
2485 {
2486         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2487
2488         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2489 }
2490
2491 static ssize_t sbi_ui_store(struct ext4_attr *a,
2492                             struct ext4_sb_info *sbi,
2493                             const char *buf, size_t count)
2494 {
2495         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2496         unsigned long t;
2497
2498         if (parse_strtoul(buf, 0xffffffff, &t))
2499                 return -EINVAL;
2500         *ui = t;
2501         return count;
2502 }
2503
2504 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2505 static struct ext4_attr ext4_attr_##_name = {                   \
2506         .attr = {.name = __stringify(_name), .mode = _mode },   \
2507         .show   = _show,                                        \
2508         .store  = _store,                                       \
2509         .offset = offsetof(struct ext4_sb_info, _elname),       \
2510 }
2511 #define EXT4_ATTR(name, mode, show, store) \
2512 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2513
2514 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2515 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2516 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2517 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2518         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2519 #define ATTR_LIST(name) &ext4_attr_##name.attr
2520
2521 EXT4_RO_ATTR(delayed_allocation_blocks);
2522 EXT4_RO_ATTR(session_write_kbytes);
2523 EXT4_RO_ATTR(lifetime_write_kbytes);
2524 EXT4_RO_ATTR(extent_cache_hits);
2525 EXT4_RO_ATTR(extent_cache_misses);
2526 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2527                  inode_readahead_blks_store, s_inode_readahead_blks);
2528 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2529 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2530 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2531 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2532 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2533 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2534 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2535 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2536
2537 static struct attribute *ext4_attrs[] = {
2538         ATTR_LIST(delayed_allocation_blocks),
2539         ATTR_LIST(session_write_kbytes),
2540         ATTR_LIST(lifetime_write_kbytes),
2541         ATTR_LIST(extent_cache_hits),
2542         ATTR_LIST(extent_cache_misses),
2543         ATTR_LIST(inode_readahead_blks),
2544         ATTR_LIST(inode_goal),
2545         ATTR_LIST(mb_stats),
2546         ATTR_LIST(mb_max_to_scan),
2547         ATTR_LIST(mb_min_to_scan),
2548         ATTR_LIST(mb_order2_req),
2549         ATTR_LIST(mb_stream_req),
2550         ATTR_LIST(mb_group_prealloc),
2551         ATTR_LIST(max_writeback_mb_bump),
2552         NULL,
2553 };
2554
2555 /* Features this copy of ext4 supports */
2556 EXT4_INFO_ATTR(lazy_itable_init);
2557 EXT4_INFO_ATTR(batched_discard);
2558
2559 static struct attribute *ext4_feat_attrs[] = {
2560         ATTR_LIST(lazy_itable_init),
2561         ATTR_LIST(batched_discard),
2562         NULL,
2563 };
2564
2565 static ssize_t ext4_attr_show(struct kobject *kobj,
2566                               struct attribute *attr, char *buf)
2567 {
2568         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2569                                                 s_kobj);
2570         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2571
2572         return a->show ? a->show(a, sbi, buf) : 0;
2573 }
2574
2575 static ssize_t ext4_attr_store(struct kobject *kobj,
2576                                struct attribute *attr,
2577                                const char *buf, size_t len)
2578 {
2579         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2580                                                 s_kobj);
2581         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2582
2583         return a->store ? a->store(a, sbi, buf, len) : 0;
2584 }
2585
2586 static void ext4_sb_release(struct kobject *kobj)
2587 {
2588         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2589                                                 s_kobj);
2590         complete(&sbi->s_kobj_unregister);
2591 }
2592
2593 static const struct sysfs_ops ext4_attr_ops = {
2594         .show   = ext4_attr_show,
2595         .store  = ext4_attr_store,
2596 };
2597
2598 static struct kobj_type ext4_ktype = {
2599         .default_attrs  = ext4_attrs,
2600         .sysfs_ops      = &ext4_attr_ops,
2601         .release        = ext4_sb_release,
2602 };
2603
2604 static void ext4_feat_release(struct kobject *kobj)
2605 {
2606         complete(&ext4_feat->f_kobj_unregister);
2607 }
2608
2609 static struct kobj_type ext4_feat_ktype = {
2610         .default_attrs  = ext4_feat_attrs,
2611         .sysfs_ops      = &ext4_attr_ops,
2612         .release        = ext4_feat_release,
2613 };
2614
2615 /*
2616  * Check whether this filesystem can be mounted based on
2617  * the features present and the RDONLY/RDWR mount requested.
2618  * Returns 1 if this filesystem can be mounted as requested,
2619  * 0 if it cannot be.
2620  */
2621 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2622 {
2623         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2624                 ext4_msg(sb, KERN_ERR,
2625                         "Couldn't mount because of "
2626                         "unsupported optional features (%x)",
2627                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2628                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2629                 return 0;
2630         }
2631
2632         if (readonly)
2633                 return 1;
2634
2635         /* Check that feature set is OK for a read-write mount */
2636         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2637                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2638                          "unsupported optional features (%x)",
2639                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2640                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2641                 return 0;
2642         }
2643         /*
2644          * Large file size enabled file system can only be mounted
2645          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2646          */
2647         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2648                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2649                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2650                                  "cannot be mounted RDWR without "
2651                                  "CONFIG_LBDAF");
2652                         return 0;
2653                 }
2654         }
2655         return 1;
2656 }
2657
2658 /*
2659  * This function is called once a day if we have errors logged
2660  * on the file system
2661  */
2662 static void print_daily_error_info(unsigned long arg)
2663 {
2664         struct super_block *sb = (struct super_block *) arg;
2665         struct ext4_sb_info *sbi;
2666         struct ext4_super_block *es;
2667
2668         sbi = EXT4_SB(sb);
2669         es = sbi->s_es;
2670
2671         if (es->s_error_count)
2672                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2673                          le32_to_cpu(es->s_error_count));
2674         if (es->s_first_error_time) {
2675                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2676                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2677                        (int) sizeof(es->s_first_error_func),
2678                        es->s_first_error_func,
2679                        le32_to_cpu(es->s_first_error_line));
2680                 if (es->s_first_error_ino)
2681                         printk(": inode %u",
2682                                le32_to_cpu(es->s_first_error_ino));
2683                 if (es->s_first_error_block)
2684                         printk(": block %llu", (unsigned long long)
2685                                le64_to_cpu(es->s_first_error_block));
2686                 printk("\n");
2687         }
2688         if (es->s_last_error_time) {
2689                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2690                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2691                        (int) sizeof(es->s_last_error_func),
2692                        es->s_last_error_func,
2693                        le32_to_cpu(es->s_last_error_line));
2694                 if (es->s_last_error_ino)
2695                         printk(": inode %u",
2696                                le32_to_cpu(es->s_last_error_ino));
2697                 if (es->s_last_error_block)
2698                         printk(": block %llu", (unsigned long long)
2699                                le64_to_cpu(es->s_last_error_block));
2700                 printk("\n");
2701         }
2702         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2703 }
2704
2705 /* Find next suitable group and run ext4_init_inode_table */
2706 static int ext4_run_li_request(struct ext4_li_request *elr)
2707 {
2708         struct ext4_group_desc *gdp = NULL;
2709         ext4_group_t group, ngroups;
2710         struct super_block *sb;
2711         unsigned long timeout = 0;
2712         int ret = 0;
2713
2714         sb = elr->lr_super;
2715         ngroups = EXT4_SB(sb)->s_groups_count;
2716
2717         for (group = elr->lr_next_group; group < ngroups; group++) {
2718                 gdp = ext4_get_group_desc(sb, group, NULL);
2719                 if (!gdp) {
2720                         ret = 1;
2721                         break;
2722                 }
2723
2724                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2725                         break;
2726         }
2727
2728         if (group == ngroups)
2729                 ret = 1;
2730
2731         if (!ret) {
2732                 timeout = jiffies;
2733                 ret = ext4_init_inode_table(sb, group,
2734                                             elr->lr_timeout ? 0 : 1);
2735                 if (elr->lr_timeout == 0) {
2736                         timeout = (jiffies - timeout) *
2737                                   elr->lr_sbi->s_li_wait_mult;
2738                         elr->lr_timeout = timeout;
2739                 }
2740                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2741                 elr->lr_next_group = group + 1;
2742         }
2743
2744         return ret;
2745 }
2746
2747 /*
2748  * Remove lr_request from the list_request and free the
2749  * request structure. Should be called with li_list_mtx held
2750  */
2751 static void ext4_remove_li_request(struct ext4_li_request *elr)
2752 {
2753         struct ext4_sb_info *sbi;
2754
2755         if (!elr)
2756                 return;
2757
2758         sbi = elr->lr_sbi;
2759
2760         list_del(&elr->lr_request);
2761         sbi->s_li_request = NULL;
2762         kfree(elr);
2763 }
2764
2765 static void ext4_unregister_li_request(struct super_block *sb)
2766 {
2767         mutex_lock(&ext4_li_mtx);
2768         if (!ext4_li_info) {
2769                 mutex_unlock(&ext4_li_mtx);
2770                 return;
2771         }
2772
2773         mutex_lock(&ext4_li_info->li_list_mtx);
2774         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2775         mutex_unlock(&ext4_li_info->li_list_mtx);
2776         mutex_unlock(&ext4_li_mtx);
2777 }
2778
2779 static struct task_struct *ext4_lazyinit_task;
2780
2781 /*
2782  * This is the function where ext4lazyinit thread lives. It walks
2783  * through the request list searching for next scheduled filesystem.
2784  * When such a fs is found, run the lazy initialization request
2785  * (ext4_rn_li_request) and keep track of the time spend in this
2786  * function. Based on that time we compute next schedule time of
2787  * the request. When walking through the list is complete, compute
2788  * next waking time and put itself into sleep.
2789  */
2790 static int ext4_lazyinit_thread(void *arg)
2791 {
2792         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2793         struct list_head *pos, *n;
2794         struct ext4_li_request *elr;
2795         unsigned long next_wakeup, cur;
2796
2797         BUG_ON(NULL == eli);
2798
2799 cont_thread:
2800         while (true) {
2801                 next_wakeup = MAX_JIFFY_OFFSET;
2802
2803                 mutex_lock(&eli->li_list_mtx);
2804                 if (list_empty(&eli->li_request_list)) {
2805                         mutex_unlock(&eli->li_list_mtx);
2806                         goto exit_thread;
2807                 }
2808
2809                 list_for_each_safe(pos, n, &eli->li_request_list) {
2810                         elr = list_entry(pos, struct ext4_li_request,
2811                                          lr_request);
2812
2813                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2814                                 if (ext4_run_li_request(elr) != 0) {
2815                                         /* error, remove the lazy_init job */
2816                                         ext4_remove_li_request(elr);
2817                                         continue;
2818                                 }
2819                         }
2820
2821                         if (time_before(elr->lr_next_sched, next_wakeup))
2822                                 next_wakeup = elr->lr_next_sched;
2823                 }
2824                 mutex_unlock(&eli->li_list_mtx);
2825
2826                 if (freezing(current))
2827                         refrigerator();
2828
2829                 cur = jiffies;
2830                 if ((time_after_eq(cur, next_wakeup)) ||
2831                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2832                         cond_resched();
2833                         continue;
2834                 }
2835
2836                 schedule_timeout_interruptible(next_wakeup - cur);
2837
2838                 if (kthread_should_stop()) {
2839                         ext4_clear_request_list();
2840                         goto exit_thread;
2841                 }
2842         }
2843
2844 exit_thread:
2845         /*
2846          * It looks like the request list is empty, but we need
2847          * to check it under the li_list_mtx lock, to prevent any
2848          * additions into it, and of course we should lock ext4_li_mtx
2849          * to atomically free the list and ext4_li_info, because at
2850          * this point another ext4 filesystem could be registering
2851          * new one.
2852          */
2853         mutex_lock(&ext4_li_mtx);
2854         mutex_lock(&eli->li_list_mtx);
2855         if (!list_empty(&eli->li_request_list)) {
2856                 mutex_unlock(&eli->li_list_mtx);
2857                 mutex_unlock(&ext4_li_mtx);
2858                 goto cont_thread;
2859         }
2860         mutex_unlock(&eli->li_list_mtx);
2861         kfree(ext4_li_info);
2862         ext4_li_info = NULL;
2863         mutex_unlock(&ext4_li_mtx);
2864
2865         return 0;
2866 }
2867
2868 static void ext4_clear_request_list(void)
2869 {
2870         struct list_head *pos, *n;
2871         struct ext4_li_request *elr;
2872
2873         mutex_lock(&ext4_li_info->li_list_mtx);
2874         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2875                 elr = list_entry(pos, struct ext4_li_request,
2876                                  lr_request);
2877                 ext4_remove_li_request(elr);
2878         }
2879         mutex_unlock(&ext4_li_info->li_list_mtx);
2880 }
2881
2882 static int ext4_run_lazyinit_thread(void)
2883 {
2884         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2885                                          ext4_li_info, "ext4lazyinit");
2886         if (IS_ERR(ext4_lazyinit_task)) {
2887                 int err = PTR_ERR(ext4_lazyinit_task);
2888                 ext4_clear_request_list();
2889                 kfree(ext4_li_info);
2890                 ext4_li_info = NULL;
2891                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2892                                  "initialization thread\n",
2893                                  err);
2894                 return err;
2895         }
2896         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2897         return 0;
2898 }
2899
2900 /*
2901  * Check whether it make sense to run itable init. thread or not.
2902  * If there is at least one uninitialized inode table, return
2903  * corresponding group number, else the loop goes through all
2904  * groups and return total number of groups.
2905  */
2906 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2907 {
2908         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2909         struct ext4_group_desc *gdp = NULL;
2910
2911         for (group = 0; group < ngroups; group++) {
2912                 gdp = ext4_get_group_desc(sb, group, NULL);
2913                 if (!gdp)
2914                         continue;
2915
2916                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2917                         break;
2918         }
2919
2920         return group;
2921 }
2922
2923 static int ext4_li_info_new(void)
2924 {
2925         struct ext4_lazy_init *eli = NULL;
2926
2927         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2928         if (!eli)
2929                 return -ENOMEM;
2930
2931         INIT_LIST_HEAD(&eli->li_request_list);
2932         mutex_init(&eli->li_list_mtx);
2933
2934         eli->li_state |= EXT4_LAZYINIT_QUIT;
2935
2936         ext4_li_info = eli;
2937
2938         return 0;
2939 }
2940
2941 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2942                                             ext4_group_t start)
2943 {
2944         struct ext4_sb_info *sbi = EXT4_SB(sb);
2945         struct ext4_li_request *elr;
2946         unsigned long rnd;
2947
2948         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2949         if (!elr)
2950                 return NULL;
2951
2952         elr->lr_super = sb;
2953         elr->lr_sbi = sbi;
2954         elr->lr_next_group = start;
2955
2956         /*
2957          * Randomize first schedule time of the request to
2958          * spread the inode table initialization requests
2959          * better.
2960          */
2961         get_random_bytes(&rnd, sizeof(rnd));
2962         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2963                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2964
2965         return elr;
2966 }
2967
2968 static int ext4_register_li_request(struct super_block *sb,
2969                                     ext4_group_t first_not_zeroed)
2970 {
2971         struct ext4_sb_info *sbi = EXT4_SB(sb);
2972         struct ext4_li_request *elr;
2973         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2974         int ret = 0;
2975
2976         if (sbi->s_li_request != NULL) {
2977                 /*
2978                  * Reset timeout so it can be computed again, because
2979                  * s_li_wait_mult might have changed.
2980                  */
2981                 sbi->s_li_request->lr_timeout = 0;
2982                 return 0;
2983         }
2984
2985         if (first_not_zeroed == ngroups ||
2986             (sb->s_flags & MS_RDONLY) ||
2987             !test_opt(sb, INIT_INODE_TABLE))
2988                 return 0;
2989
2990         elr = ext4_li_request_new(sb, first_not_zeroed);
2991         if (!elr)
2992                 return -ENOMEM;
2993
2994         mutex_lock(&ext4_li_mtx);
2995
2996         if (NULL == ext4_li_info) {
2997                 ret = ext4_li_info_new();
2998                 if (ret)
2999                         goto out;
3000         }
3001
3002         mutex_lock(&ext4_li_info->li_list_mtx);
3003         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3004         mutex_unlock(&ext4_li_info->li_list_mtx);
3005
3006         sbi->s_li_request = elr;
3007         /*
3008          * set elr to NULL here since it has been inserted to
3009          * the request_list and the removal and free of it is
3010          * handled by ext4_clear_request_list from now on.
3011          */
3012         elr = NULL;
3013
3014         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3015                 ret = ext4_run_lazyinit_thread();
3016                 if (ret)
3017                         goto out;
3018         }
3019 out:
3020         mutex_unlock(&ext4_li_mtx);
3021         if (ret)
3022                 kfree(elr);
3023         return ret;
3024 }
3025
3026 /*
3027  * We do not need to lock anything since this is called on
3028  * module unload.
3029  */
3030 static void ext4_destroy_lazyinit_thread(void)
3031 {
3032         /*
3033          * If thread exited earlier
3034          * there's nothing to be done.
3035          */
3036         if (!ext4_li_info || !ext4_lazyinit_task)
3037                 return;
3038
3039         kthread_stop(ext4_lazyinit_task);
3040 }
3041
3042 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3043                                 __releases(kernel_lock)
3044                                 __acquires(kernel_lock)
3045 {
3046         char *orig_data = kstrdup(data, GFP_KERNEL);
3047         struct buffer_head *bh;
3048         struct ext4_super_block *es = NULL;
3049         struct ext4_sb_info *sbi;
3050         ext4_fsblk_t block;
3051         ext4_fsblk_t sb_block = get_sb_block(&data);
3052         ext4_fsblk_t logical_sb_block;
3053         unsigned long offset = 0;
3054         unsigned long journal_devnum = 0;
3055         unsigned long def_mount_opts;
3056         struct inode *root;
3057         char *cp;
3058         const char *descr;
3059         int ret = -ENOMEM;
3060         int blocksize;
3061         unsigned int db_count;
3062         unsigned int i;
3063         int needs_recovery, has_huge_files;
3064         __u64 blocks_count;
3065         int err;
3066         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3067         ext4_group_t first_not_zeroed;
3068
3069         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3070         if (!sbi)
3071                 goto out_free_orig;
3072
3073         sbi->s_blockgroup_lock =
3074                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3075         if (!sbi->s_blockgroup_lock) {
3076                 kfree(sbi);
3077                 goto out_free_orig;
3078         }
3079         sb->s_fs_info = sbi;
3080         sbi->s_mount_opt = 0;
3081         sbi->s_resuid = EXT4_DEF_RESUID;
3082         sbi->s_resgid = EXT4_DEF_RESGID;
3083         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3084         sbi->s_sb_block = sb_block;
3085         if (sb->s_bdev->bd_part)
3086                 sbi->s_sectors_written_start =
3087                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3088
3089         /* Cleanup superblock name */
3090         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3091                 *cp = '!';
3092
3093         ret = -EINVAL;
3094         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3095         if (!blocksize) {
3096                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3097                 goto out_fail;
3098         }
3099
3100         /*
3101          * The ext4 superblock will not be buffer aligned for other than 1kB
3102          * block sizes.  We need to calculate the offset from buffer start.
3103          */
3104         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3105                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3106                 offset = do_div(logical_sb_block, blocksize);
3107         } else {
3108                 logical_sb_block = sb_block;
3109         }
3110
3111         if (!(bh = sb_bread(sb, logical_sb_block))) {
3112                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3113                 goto out_fail;
3114         }
3115         /*
3116          * Note: s_es must be initialized as soon as possible because
3117          *       some ext4 macro-instructions depend on its value
3118          */
3119         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3120         sbi->s_es = es;
3121         sb->s_magic = le16_to_cpu(es->s_magic);
3122         if (sb->s_magic != EXT4_SUPER_MAGIC)
3123                 goto cantfind_ext4;
3124         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3125
3126         /* Set defaults before we parse the mount options */
3127         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3128         set_opt(sb, INIT_INODE_TABLE);
3129         if (def_mount_opts & EXT4_DEFM_DEBUG)
3130                 set_opt(sb, DEBUG);
3131         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3132                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3133                         "2.6.38");
3134                 set_opt(sb, GRPID);
3135         }
3136         if (def_mount_opts & EXT4_DEFM_UID16)
3137                 set_opt(sb, NO_UID32);
3138         /* xattr user namespace & acls are now defaulted on */
3139 #ifdef CONFIG_EXT4_FS_XATTR
3140         set_opt(sb, XATTR_USER);
3141 #endif
3142 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3143         set_opt(sb, POSIX_ACL);
3144 #endif
3145         set_opt(sb, MBLK_IO_SUBMIT);
3146         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3147                 set_opt(sb, JOURNAL_DATA);
3148         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3149                 set_opt(sb, ORDERED_DATA);
3150         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3151                 set_opt(sb, WRITEBACK_DATA);
3152
3153         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3154                 set_opt(sb, ERRORS_PANIC);
3155         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3156                 set_opt(sb, ERRORS_CONT);
3157         else
3158                 set_opt(sb, ERRORS_RO);
3159         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3160                 set_opt(sb, BLOCK_VALIDITY);
3161         if (def_mount_opts & EXT4_DEFM_DISCARD)
3162                 set_opt(sb, DISCARD);
3163
3164         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3165         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3166         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3167         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3168         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3169
3170         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3171                 set_opt(sb, BARRIER);
3172
3173         /*
3174          * enable delayed allocation by default
3175          * Use -o nodelalloc to turn it off
3176          */
3177         if (!IS_EXT3_SB(sb) &&
3178             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3179                 set_opt(sb, DELALLOC);
3180
3181         /*
3182          * set default s_li_wait_mult for lazyinit, for the case there is
3183          * no mount option specified.
3184          */
3185         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3186
3187         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3188                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3189                 ext4_msg(sb, KERN_WARNING,
3190                          "failed to parse options in superblock: %s",
3191                          sbi->s_es->s_mount_opts);
3192         }
3193         if (!parse_options((char *) data, sb, &journal_devnum,
3194                            &journal_ioprio, NULL, 0))
3195                 goto failed_mount;
3196
3197         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3198                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3199
3200         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3201             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3202              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3203              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3204                 ext4_msg(sb, KERN_WARNING,
3205                        "feature flags set on rev 0 fs, "
3206                        "running e2fsck is recommended");
3207
3208         if (IS_EXT2_SB(sb)) {
3209                 if (ext2_feature_set_ok(sb))
3210                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3211                                  "using the ext4 subsystem");
3212                 else {
3213                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3214                                  "to feature incompatibilities");
3215                         goto failed_mount;
3216                 }
3217         }
3218
3219         if (IS_EXT3_SB(sb)) {
3220                 if (ext3_feature_set_ok(sb))
3221                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3222                                  "using the ext4 subsystem");
3223                 else {
3224                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3225                                  "to feature incompatibilities");
3226                         goto failed_mount;
3227                 }
3228         }
3229
3230         /*
3231          * Check feature flags regardless of the revision level, since we
3232          * previously didn't change the revision level when setting the flags,
3233          * so there is a chance incompat flags are set on a rev 0 filesystem.
3234          */
3235         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3236                 goto failed_mount;
3237
3238         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3239
3240         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3241             blocksize > EXT4_MAX_BLOCK_SIZE) {
3242                 ext4_msg(sb, KERN_ERR,
3243                        "Unsupported filesystem blocksize %d", blocksize);
3244                 goto failed_mount;
3245         }
3246
3247         if (sb->s_blocksize != blocksize) {
3248                 /* Validate the filesystem blocksize */
3249                 if (!sb_set_blocksize(sb, blocksize)) {
3250                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3251                                         blocksize);
3252                         goto failed_mount;
3253                 }
3254
3255                 brelse(bh);
3256                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3257                 offset = do_div(logical_sb_block, blocksize);
3258                 bh = sb_bread(sb, logical_sb_block);
3259                 if (!bh) {
3260                         ext4_msg(sb, KERN_ERR,
3261                                "Can't read superblock on 2nd try");
3262                         goto failed_mount;
3263                 }
3264                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3265                 sbi->s_es = es;
3266                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3267                         ext4_msg(sb, KERN_ERR,
3268                                "Magic mismatch, very weird!");
3269                         goto failed_mount;
3270                 }
3271         }
3272
3273         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3274                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3275         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3276                                                       has_huge_files);
3277         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3278
3279         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3280                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3281                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3282         } else {
3283                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3284                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3285                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3286                     (!is_power_of_2(sbi->s_inode_size)) ||
3287                     (sbi->s_inode_size > blocksize)) {
3288                         ext4_msg(sb, KERN_ERR,
3289                                "unsupported inode size: %d",
3290                                sbi->s_inode_size);
3291                         goto failed_mount;
3292                 }
3293                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3294                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3295         }
3296
3297         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3298         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3299                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3300                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3301                     !is_power_of_2(sbi->s_desc_size)) {
3302                         ext4_msg(sb, KERN_ERR,
3303                                "unsupported descriptor size %lu",
3304                                sbi->s_desc_size);
3305                         goto failed_mount;
3306                 }
3307         } else
3308                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3309
3310         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3311         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3312         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3313                 goto cantfind_ext4;
3314
3315         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3316         if (sbi->s_inodes_per_block == 0)
3317                 goto cantfind_ext4;
3318         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3319                                         sbi->s_inodes_per_block;
3320         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3321         sbi->s_sbh = bh;
3322         sbi->s_mount_state = le16_to_cpu(es->s_state);
3323         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3324         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3325
3326         for (i = 0; i < 4; i++)
3327                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3328         sbi->s_def_hash_version = es->s_def_hash_version;
3329         i = le32_to_cpu(es->s_flags);
3330         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3331                 sbi->s_hash_unsigned = 3;
3332         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3333 #ifdef __CHAR_UNSIGNED__
3334                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3335                 sbi->s_hash_unsigned = 3;
3336 #else
3337                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3338 #endif
3339                 sb->s_dirt = 1;
3340         }
3341
3342         if (sbi->s_blocks_per_group > blocksize * 8) {
3343                 ext4_msg(sb, KERN_ERR,
3344                        "#blocks per group too big: %lu",
3345                        sbi->s_blocks_per_group);
3346                 goto failed_mount;
3347         }
3348         if (sbi->s_inodes_per_group > blocksize * 8) {
3349                 ext4_msg(sb, KERN_ERR,
3350                        "#inodes per group too big: %lu",
3351                        sbi->s_inodes_per_group);
3352                 goto failed_mount;
3353         }
3354
3355         /*
3356          * Test whether we have more sectors than will fit in sector_t,
3357          * and whether the max offset is addressable by the page cache.
3358          */
3359         err = generic_check_addressable(sb->s_blocksize_bits,
3360                                         ext4_blocks_count(es));
3361         if (err) {
3362                 ext4_msg(sb, KERN_ERR, "filesystem"
3363                          " too large to mount safely on this system");
3364                 if (sizeof(sector_t) < 8)
3365                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3366                 ret = err;
3367                 goto failed_mount;
3368         }
3369
3370         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3371                 goto cantfind_ext4;
3372
3373         /* check blocks count against device size */
3374         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3375         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3376                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3377                        "exceeds size of device (%llu blocks)",
3378                        ext4_blocks_count(es), blocks_count);
3379                 goto failed_mount;
3380         }
3381
3382         /*
3383          * It makes no sense for the first data block to be beyond the end
3384          * of the filesystem.
3385          */
3386         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3387                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3388                          "block %u is beyond end of filesystem (%llu)",
3389                          le32_to_cpu(es->s_first_data_block),
3390                          ext4_blocks_count(es));
3391                 goto failed_mount;
3392         }
3393         blocks_count = (ext4_blocks_count(es) -
3394                         le32_to_cpu(es->s_first_data_block) +
3395                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3396         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3397         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3398                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3399                        "(block count %llu, first data block %u, "
3400                        "blocks per group %lu)", sbi->s_groups_count,
3401                        ext4_blocks_count(es),
3402                        le32_to_cpu(es->s_first_data_block),
3403                        EXT4_BLOCKS_PER_GROUP(sb));
3404                 goto failed_mount;
3405         }
3406         sbi->s_groups_count = blocks_count;
3407         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3408                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3409         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3410                    EXT4_DESC_PER_BLOCK(sb);
3411         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3412                                     GFP_KERNEL);
3413         if (sbi->s_group_desc == NULL) {
3414                 ext4_msg(sb, KERN_ERR, "not enough memory");
3415                 goto failed_mount;
3416         }
3417
3418 #ifdef CONFIG_PROC_FS
3419         if (ext4_proc_root)
3420                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3421 #endif
3422
3423         bgl_lock_init(sbi->s_blockgroup_lock);
3424
3425         for (i = 0; i < db_count; i++) {
3426                 block = descriptor_loc(sb, logical_sb_block, i);
3427                 sbi->s_group_desc[i] = sb_bread(sb, block);
3428                 if (!sbi->s_group_desc[i]) {
3429                         ext4_msg(sb, KERN_ERR,
3430                                "can't read group descriptor %d", i);
3431                         db_count = i;
3432                         goto failed_mount2;
3433                 }
3434         }
3435         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3436                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3437                 goto failed_mount2;
3438         }
3439         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3440                 if (!ext4_fill_flex_info(sb)) {
3441                         ext4_msg(sb, KERN_ERR,
3442                                "unable to initialize "
3443                                "flex_bg meta info!");
3444                         goto failed_mount2;
3445                 }
3446
3447         sbi->s_gdb_count = db_count;
3448         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3449         spin_lock_init(&sbi->s_next_gen_lock);
3450
3451         init_timer(&sbi->s_err_report);
3452         sbi->s_err_report.function = print_daily_error_info;
3453         sbi->s_err_report.data = (unsigned long) sb;
3454
3455         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3456                         ext4_count_free_blocks(sb));
3457         if (!err) {
3458                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3459                                 ext4_count_free_inodes(sb));
3460         }
3461         if (!err) {
3462                 err = percpu_counter_init(&sbi->s_dirs_counter,
3463                                 ext4_count_dirs(sb));
3464         }
3465         if (!err) {
3466                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3467         }
3468         if (err) {
3469                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3470                 goto failed_mount3;
3471         }
3472
3473         sbi->s_stripe = ext4_get_stripe_size(sbi);
3474         sbi->s_max_writeback_mb_bump = 128;
3475
3476         /*
3477          * set up enough so that it can read an inode
3478          */
3479         if (!test_opt(sb, NOLOAD) &&
3480             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3481                 sb->s_op = &ext4_sops;
3482         else
3483                 sb->s_op = &ext4_nojournal_sops;
3484         sb->s_export_op = &ext4_export_ops;
3485         sb->s_xattr = ext4_xattr_handlers;
3486 #ifdef CONFIG_QUOTA
3487         sb->s_qcop = &ext4_qctl_operations;
3488         sb->dq_op = &ext4_quota_operations;
3489 #endif
3490         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3491
3492         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3493         mutex_init(&sbi->s_orphan_lock);
3494         mutex_init(&sbi->s_resize_lock);
3495
3496         sb->s_root = NULL;
3497
3498         needs_recovery = (es->s_last_orphan != 0 ||
3499                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3500                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3501
3502         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3503             !(sb->s_flags & MS_RDONLY))
3504                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3505                         goto failed_mount3;
3506
3507         /*
3508          * The first inode we look at is the journal inode.  Don't try
3509          * root first: it may be modified in the journal!
3510          */
3511         if (!test_opt(sb, NOLOAD) &&
3512             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3513                 if (ext4_load_journal(sb, es, journal_devnum))
3514                         goto failed_mount3;
3515         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3516               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3517                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3518                        "suppressed and not mounted read-only");
3519                 goto failed_mount_wq;
3520         } else {
3521                 clear_opt(sb, DATA_FLAGS);
3522                 sbi->s_journal = NULL;
3523                 needs_recovery = 0;
3524                 goto no_journal;
3525         }
3526
3527         if (ext4_blocks_count(es) > 0xffffffffULL &&
3528             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3529                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3530                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3531                 goto failed_mount_wq;
3532         }
3533
3534         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3535                 jbd2_journal_set_features(sbi->s_journal,
3536                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3537                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3538         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3539                 jbd2_journal_set_features(sbi->s_journal,
3540                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3541                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3542                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3543         } else {
3544                 jbd2_journal_clear_features(sbi->s_journal,
3545                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3546                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3547         }
3548
3549         /* We have now updated the journal if required, so we can
3550          * validate the data journaling mode. */
3551         switch (test_opt(sb, DATA_FLAGS)) {
3552         case 0:
3553                 /* No mode set, assume a default based on the journal
3554                  * capabilities: ORDERED_DATA if the journal can
3555                  * cope, else JOURNAL_DATA
3556                  */
3557                 if (jbd2_journal_check_available_features
3558                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3559                         set_opt(sb, ORDERED_DATA);
3560                 else
3561                         set_opt(sb, JOURNAL_DATA);
3562                 break;
3563
3564         case EXT4_MOUNT_ORDERED_DATA:
3565         case EXT4_MOUNT_WRITEBACK_DATA:
3566                 if (!jbd2_journal_check_available_features
3567                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3568                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3569                                "requested data journaling mode");
3570                         goto failed_mount_wq;
3571                 }
3572         default:
3573                 break;
3574         }
3575         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3576
3577         /*
3578          * The journal may have updated the bg summary counts, so we
3579          * need to update the global counters.
3580          */
3581         percpu_counter_set(&sbi->s_freeblocks_counter,
3582                            ext4_count_free_blocks(sb));
3583         percpu_counter_set(&sbi->s_freeinodes_counter,
3584                            ext4_count_free_inodes(sb));
3585         percpu_counter_set(&sbi->s_dirs_counter,
3586                            ext4_count_dirs(sb));
3587         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3588
3589 no_journal:
3590         /*
3591          * The maximum number of concurrent works can be high and
3592          * concurrency isn't really necessary.  Limit it to 1.
3593          */
3594         EXT4_SB(sb)->dio_unwritten_wq =
3595                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3596         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3597                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3598                 goto failed_mount_wq;
3599         }
3600
3601         /*
3602          * The jbd2_journal_load will have done any necessary log recovery,
3603          * so we can safely mount the rest of the filesystem now.
3604          */
3605
3606         root = ext4_iget(sb, EXT4_ROOT_INO);
3607         if (IS_ERR(root)) {
3608                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3609                 ret = PTR_ERR(root);
3610                 root = NULL;
3611                 goto failed_mount4;
3612         }
3613         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3614                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3615                 goto failed_mount4;
3616         }
3617         sb->s_root = d_alloc_root(root);
3618         if (!sb->s_root) {
3619                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3620                 ret = -ENOMEM;
3621                 goto failed_mount4;
3622         }
3623
3624         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3625
3626         /* determine the minimum size of new large inodes, if present */
3627         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3628                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3629                                                      EXT4_GOOD_OLD_INODE_SIZE;
3630                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3631                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3632                         if (sbi->s_want_extra_isize <
3633                             le16_to_cpu(es->s_want_extra_isize))
3634                                 sbi->s_want_extra_isize =
3635                                         le16_to_cpu(es->s_want_extra_isize);
3636                         if (sbi->s_want_extra_isize <
3637                             le16_to_cpu(es->s_min_extra_isize))
3638                                 sbi->s_want_extra_isize =
3639                                         le16_to_cpu(es->s_min_extra_isize);
3640                 }
3641         }
3642         /* Check if enough inode space is available */
3643         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3644                                                         sbi->s_inode_size) {
3645                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3646                                                        EXT4_GOOD_OLD_INODE_SIZE;
3647                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3648                          "available");
3649         }
3650
3651         if (test_opt(sb, DELALLOC) &&
3652             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3653                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3654                          "requested data journaling mode");
3655                 clear_opt(sb, DELALLOC);
3656         }
3657         if (test_opt(sb, DIOREAD_NOLOCK)) {
3658                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3659                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3660                                 "option - requested data journaling mode");
3661                         clear_opt(sb, DIOREAD_NOLOCK);
3662                 }
3663                 if (sb->s_blocksize < PAGE_SIZE) {
3664                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3665                                 "option - block size is too small");
3666                         clear_opt(sb, DIOREAD_NOLOCK);
3667                 }
3668         }
3669
3670         err = ext4_setup_system_zone(sb);
3671         if (err) {
3672                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3673                          "zone (%d)", err);
3674                 goto failed_mount4;
3675         }
3676
3677         ext4_ext_init(sb);
3678         err = ext4_mb_init(sb, needs_recovery);
3679         if (err) {
3680                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3681                          err);
3682                 goto failed_mount4;
3683         }
3684
3685         err = ext4_register_li_request(sb, first_not_zeroed);
3686         if (err)
3687                 goto failed_mount4;
3688
3689         sbi->s_kobj.kset = ext4_kset;
3690         init_completion(&sbi->s_kobj_unregister);
3691         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3692                                    "%s", sb->s_id);
3693         if (err) {
3694                 ext4_mb_release(sb);
3695                 ext4_ext_release(sb);
3696                 goto failed_mount4;
3697         };
3698
3699         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3700         ext4_orphan_cleanup(sb, es);
3701         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3702         if (needs_recovery) {
3703                 ext4_msg(sb, KERN_INFO, "recovery complete");
3704                 ext4_mark_recovery_complete(sb, es);
3705         }
3706         if (EXT4_SB(sb)->s_journal) {
3707                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3708                         descr = " journalled data mode";
3709                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3710                         descr = " ordered data mode";
3711                 else
3712                         descr = " writeback data mode";
3713         } else
3714                 descr = "out journal";
3715
3716         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3717                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3718                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3719
3720         if (es->s_error_count)
3721                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3722
3723         kfree(orig_data);
3724         return 0;
3725
3726 cantfind_ext4:
3727         if (!silent)
3728                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3729         goto failed_mount;
3730
3731 failed_mount4:
3732         iput(root);
3733         sb->s_root = NULL;
3734         ext4_msg(sb, KERN_ERR, "mount failed");
3735         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3736 failed_mount_wq:
3737         ext4_release_system_zone(sb);
3738         if (sbi->s_journal) {
3739                 jbd2_journal_destroy(sbi->s_journal);
3740                 sbi->s_journal = NULL;
3741         }
3742 failed_mount3:
3743         del_timer(&sbi->s_err_report);
3744         if (sbi->s_flex_groups) {
3745                 if (is_vmalloc_addr(sbi->s_flex_groups))
3746                         vfree(sbi->s_flex_groups);
3747                 else
3748                         kfree(sbi->s_flex_groups);
3749         }
3750         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3751         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3752         percpu_counter_destroy(&sbi->s_dirs_counter);
3753         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3754         if (sbi->s_mmp_tsk)
3755                 kthread_stop(sbi->s_mmp_tsk);
3756 failed_mount2:
3757         for (i = 0; i < db_count; i++)
3758                 brelse(sbi->s_group_desc[i]);
3759         kfree(sbi->s_group_desc);
3760 failed_mount:
3761         if (sbi->s_proc) {
3762                 remove_proc_entry(sb->s_id, ext4_proc_root);
3763         }
3764 #ifdef CONFIG_QUOTA
3765         for (i = 0; i < MAXQUOTAS; i++)
3766                 kfree(sbi->s_qf_names[i]);
3767 #endif
3768         ext4_blkdev_remove(sbi);
3769         brelse(bh);
3770 out_fail:
3771         sb->s_fs_info = NULL;
3772         kfree(sbi->s_blockgroup_lock);
3773         kfree(sbi);
3774 out_free_orig:
3775         kfree(orig_data);
3776         return ret;
3777 }
3778
3779 /*
3780  * Setup any per-fs journal parameters now.  We'll do this both on
3781  * initial mount, once the journal has been initialised but before we've
3782  * done any recovery; and again on any subsequent remount.
3783  */
3784 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3785 {
3786         struct ext4_sb_info *sbi = EXT4_SB(sb);
3787
3788         journal->j_commit_interval = sbi->s_commit_interval;
3789         journal->j_min_batch_time = sbi->s_min_batch_time;
3790         journal->j_max_batch_time = sbi->s_max_batch_time;
3791
3792         write_lock(&journal->j_state_lock);
3793         if (test_opt(sb, BARRIER))
3794                 journal->j_flags |= JBD2_BARRIER;
3795         else
3796                 journal->j_flags &= ~JBD2_BARRIER;
3797         if (test_opt(sb, DATA_ERR_ABORT))
3798                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3799         else
3800                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3801         write_unlock(&journal->j_state_lock);
3802 }
3803
3804 static journal_t *ext4_get_journal(struct super_block *sb,
3805                                    unsigned int journal_inum)
3806 {
3807         struct inode *journal_inode;
3808         journal_t *journal;
3809
3810         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3811
3812         /* First, test for the existence of a valid inode on disk.  Bad
3813          * things happen if we iget() an unused inode, as the subsequent
3814          * iput() will try to delete it. */
3815
3816         journal_inode = ext4_iget(sb, journal_inum);
3817         if (IS_ERR(journal_inode)) {
3818                 ext4_msg(sb, KERN_ERR, "no journal found");
3819                 return NULL;
3820         }
3821         if (!journal_inode->i_nlink) {
3822                 make_bad_inode(journal_inode);
3823                 iput(journal_inode);
3824                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3825                 return NULL;
3826         }
3827
3828         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3829                   journal_inode, journal_inode->i_size);
3830         if (!S_ISREG(journal_inode->i_mode)) {
3831                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3832                 iput(journal_inode);
3833                 return NULL;
3834         }
3835
3836         journal = jbd2_journal_init_inode(journal_inode);
3837         if (!journal) {
3838                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3839                 iput(journal_inode);
3840                 return NULL;
3841         }
3842         journal->j_private = sb;
3843         ext4_init_journal_params(sb, journal);
3844         return journal;
3845 }
3846
3847 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3848                                        dev_t j_dev)
3849 {
3850         struct buffer_head *bh;
3851         journal_t *journal;
3852         ext4_fsblk_t start;
3853         ext4_fsblk_t len;
3854         int hblock, blocksize;
3855         ext4_fsblk_t sb_block;
3856         unsigned long offset;
3857         struct ext4_super_block *es;
3858         struct block_device *bdev;
3859
3860         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3861
3862         bdev = ext4_blkdev_get(j_dev, sb);
3863         if (bdev == NULL)
3864                 return NULL;
3865
3866         blocksize = sb->s_blocksize;
3867         hblock = bdev_logical_block_size(bdev);
3868         if (blocksize < hblock) {
3869                 ext4_msg(sb, KERN_ERR,
3870                         "blocksize too small for journal device");
3871                 goto out_bdev;
3872         }
3873
3874         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3875         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3876         set_blocksize(bdev, blocksize);
3877         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3878                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3879                        "external journal");
3880                 goto out_bdev;
3881         }
3882
3883         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3884         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3885             !(le32_to_cpu(es->s_feature_incompat) &
3886               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3887                 ext4_msg(sb, KERN_ERR, "external journal has "
3888                                         "bad superblock");
3889                 brelse(bh);
3890                 goto out_bdev;
3891         }
3892
3893         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3894                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3895                 brelse(bh);
3896                 goto out_bdev;
3897         }
3898
3899         len = ext4_blocks_count(es);
3900         start = sb_block + 1;
3901         brelse(bh);     /* we're done with the superblock */
3902
3903         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3904                                         start, len, blocksize);
3905         if (!journal) {
3906                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3907                 goto out_bdev;
3908         }
3909         journal->j_private = sb;
3910         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3911         wait_on_buffer(journal->j_sb_buffer);
3912         if (!buffer_uptodate(journal->j_sb_buffer)) {
3913                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3914                 goto out_journal;
3915         }
3916         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3917                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3918                                         "user (unsupported) - %d",
3919                         be32_to_cpu(journal->j_superblock->s_nr_users));
3920                 goto out_journal;
3921         }
3922         EXT4_SB(sb)->journal_bdev = bdev;
3923         ext4_init_journal_params(sb, journal);
3924         return journal;
3925
3926 out_journal:
3927         jbd2_journal_destroy(journal);
3928 out_bdev:
3929         ext4_blkdev_put(bdev);
3930         return NULL;
3931 }
3932
3933 static int ext4_load_journal(struct super_block *sb,
3934                              struct ext4_super_block *es,
3935                              unsigned long journal_devnum)
3936 {
3937         journal_t *journal;
3938         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3939         dev_t journal_dev;
3940         int err = 0;
3941         int really_read_only;
3942
3943         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3944
3945         if (journal_devnum &&
3946             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3947                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3948                         "numbers have changed");
3949                 journal_dev = new_decode_dev(journal_devnum);
3950         } else
3951                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3952
3953         really_read_only = bdev_read_only(sb->s_bdev);
3954
3955         /*
3956          * Are we loading a blank journal or performing recovery after a
3957          * crash?  For recovery, we need to check in advance whether we
3958          * can get read-write access to the device.
3959          */
3960         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3961                 if (sb->s_flags & MS_RDONLY) {
3962                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3963                                         "required on readonly filesystem");
3964                         if (really_read_only) {
3965                                 ext4_msg(sb, KERN_ERR, "write access "
3966                                         "unavailable, cannot proceed");
3967                                 return -EROFS;
3968                         }
3969                         ext4_msg(sb, KERN_INFO, "write access will "
3970                                "be enabled during recovery");
3971                 }
3972         }
3973
3974         if (journal_inum && journal_dev) {
3975                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3976                        "and inode journals!");
3977                 return -EINVAL;
3978         }
3979
3980         if (journal_inum) {
3981                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3982                         return -EINVAL;
3983         } else {
3984                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3985                         return -EINVAL;
3986         }
3987
3988         if (!(journal->j_flags & JBD2_BARRIER))
3989                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3990
3991         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3992                 err = jbd2_journal_update_format(journal);
3993                 if (err)  {
3994                         ext4_msg(sb, KERN_ERR, "error updating journal");
3995                         jbd2_journal_destroy(journal);
3996                         return err;
3997                 }
3998         }
3999
4000         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4001                 err = jbd2_journal_wipe(journal, !really_read_only);
4002         if (!err) {
4003                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4004                 if (save)
4005                         memcpy(save, ((char *) es) +
4006                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4007                 err = jbd2_journal_load(journal);
4008                 if (save)
4009                         memcpy(((char *) es) + EXT4_S_ERR_START,
4010                                save, EXT4_S_ERR_LEN);
4011                 kfree(save);
4012         }
4013
4014         if (err) {
4015                 ext4_msg(sb, KERN_ERR, "error loading journal");
4016                 jbd2_journal_destroy(journal);
4017                 return err;
4018         }
4019
4020         EXT4_SB(sb)->s_journal = journal;
4021         ext4_clear_journal_err(sb, es);
4022
4023         if (!really_read_only && journal_devnum &&
4024             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4025                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4026
4027                 /* Make sure we flush the recovery flag to disk. */
4028                 ext4_commit_super(sb, 1);
4029         }
4030
4031         return 0;
4032 }
4033
4034 static int ext4_commit_super(struct super_block *sb, int sync)
4035 {
4036         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4037         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4038         int error = 0;
4039
4040         if (!sbh)
4041                 return error;
4042         if (buffer_write_io_error(sbh)) {
4043                 /*
4044                  * Oh, dear.  A previous attempt to write the
4045                  * superblock failed.  This could happen because the
4046                  * USB device was yanked out.  Or it could happen to
4047                  * be a transient write error and maybe the block will
4048                  * be remapped.  Nothing we can do but to retry the
4049                  * write and hope for the best.
4050                  */
4051                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4052                        "superblock detected");
4053                 clear_buffer_write_io_error(sbh);
4054                 set_buffer_uptodate(sbh);
4055         }
4056         /*
4057          * If the file system is mounted read-only, don't update the
4058          * superblock write time.  This avoids updating the superblock
4059          * write time when we are mounting the root file system
4060          * read/only but we need to replay the journal; at that point,
4061          * for people who are east of GMT and who make their clock
4062          * tick in localtime for Windows bug-for-bug compatibility,
4063          * the clock is set in the future, and this will cause e2fsck
4064          * to complain and force a full file system check.
4065          */
4066         if (!(sb->s_flags & MS_RDONLY))
4067                 es->s_wtime = cpu_to_le32(get_seconds());
4068         if (sb->s_bdev->bd_part)
4069                 es->s_kbytes_written =
4070                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4071                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4072                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4073         else
4074                 es->s_kbytes_written =
4075                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4076         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4077                                            &EXT4_SB(sb)->s_freeblocks_counter));
4078         es->s_free_inodes_count =
4079                 cpu_to_le32(percpu_counter_sum_positive(
4080                                 &EXT4_SB(sb)->s_freeinodes_counter));
4081         sb->s_dirt = 0;
4082         BUFFER_TRACE(sbh, "marking dirty");
4083         mark_buffer_dirty(sbh);
4084         if (sync) {
4085                 error = sync_dirty_buffer(sbh);
4086                 if (error)
4087                         return error;
4088
4089                 error = buffer_write_io_error(sbh);
4090                 if (error) {
4091                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4092                                "superblock");
4093                         clear_buffer_write_io_error(sbh);
4094                         set_buffer_uptodate(sbh);
4095                 }
4096         }
4097         return error;
4098 }
4099
4100 /*
4101  * Have we just finished recovery?  If so, and if we are mounting (or
4102  * remounting) the filesystem readonly, then we will end up with a
4103  * consistent fs on disk.  Record that fact.
4104  */
4105 static void ext4_mark_recovery_complete(struct super_block *sb,
4106                                         struct ext4_super_block *es)
4107 {
4108         journal_t *journal = EXT4_SB(sb)->s_journal;
4109
4110         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4111                 BUG_ON(journal != NULL);
4112                 return;
4113         }
4114         jbd2_journal_lock_updates(journal);
4115         if (jbd2_journal_flush(journal) < 0)
4116                 goto out;
4117
4118         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4119             sb->s_flags & MS_RDONLY) {
4120                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4121                 ext4_commit_super(sb, 1);
4122         }
4123
4124 out:
4125         jbd2_journal_unlock_updates(journal);
4126 }
4127
4128 /*
4129  * If we are mounting (or read-write remounting) a filesystem whose journal
4130  * has recorded an error from a previous lifetime, move that error to the
4131  * main filesystem now.
4132  */
4133 static void ext4_clear_journal_err(struct super_block *sb,
4134                                    struct ext4_super_block *es)
4135 {
4136         journal_t *journal;
4137         int j_errno;
4138         const char *errstr;
4139
4140         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4141
4142         journal = EXT4_SB(sb)->s_journal;
4143
4144         /*
4145          * Now check for any error status which may have been recorded in the
4146          * journal by a prior ext4_error() or ext4_abort()
4147          */
4148
4149         j_errno = jbd2_journal_errno(journal);
4150         if (j_errno) {
4151                 char nbuf[16];
4152
4153                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4154                 ext4_warning(sb, "Filesystem error recorded "
4155                              "from previous mount: %s", errstr);
4156                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4157
4158                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4159                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4160                 ext4_commit_super(sb, 1);
4161
4162                 jbd2_journal_clear_err(journal);
4163         }
4164 }
4165
4166 /*
4167  * Force the running and committing transactions to commit,
4168  * and wait on the commit.
4169  */
4170 int ext4_force_commit(struct super_block *sb)
4171 {
4172         journal_t *journal;
4173         int ret = 0;
4174
4175         if (sb->s_flags & MS_RDONLY)
4176                 return 0;
4177
4178         journal = EXT4_SB(sb)->s_journal;
4179         if (journal) {
4180                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4181                 ret = ext4_journal_force_commit(journal);
4182         }
4183
4184         return ret;
4185 }
4186
4187 static void ext4_write_super(struct super_block *sb)
4188 {
4189         lock_super(sb);
4190         ext4_commit_super(sb, 1);
4191         unlock_super(sb);
4192 }
4193
4194 static int ext4_sync_fs(struct super_block *sb, int wait)
4195 {
4196         int ret = 0;
4197         tid_t target;
4198         struct ext4_sb_info *sbi = EXT4_SB(sb);
4199
4200         trace_ext4_sync_fs(sb, wait);
4201         flush_workqueue(sbi->dio_unwritten_wq);
4202         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4203                 if (wait)
4204                         jbd2_log_wait_commit(sbi->s_journal, target);
4205         }
4206         return ret;
4207 }
4208
4209 /*
4210  * LVM calls this function before a (read-only) snapshot is created.  This
4211  * gives us a chance to flush the journal completely and mark the fs clean.
4212  *
4213  * Note that only this function cannot bring a filesystem to be in a clean
4214  * state independently, because ext4 prevents a new handle from being started
4215  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4216  * the upper layer.
4217  */
4218 static int ext4_freeze(struct super_block *sb)
4219 {
4220         int error = 0;
4221         journal_t *journal;
4222
4223         if (sb->s_flags & MS_RDONLY)
4224                 return 0;
4225
4226         journal = EXT4_SB(sb)->s_journal;
4227
4228         /* Now we set up the journal barrier. */
4229         jbd2_journal_lock_updates(journal);
4230
4231         /*
4232          * Don't clear the needs_recovery flag if we failed to flush
4233          * the journal.
4234          */
4235         error = jbd2_journal_flush(journal);
4236         if (error < 0)
4237                 goto out;
4238
4239         /* Journal blocked and flushed, clear needs_recovery flag. */
4240         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4241         error = ext4_commit_super(sb, 1);
4242 out:
4243         /* we rely on s_frozen to stop further updates */
4244         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4245         return error;
4246 }
4247
4248 /*
4249  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4250  * flag here, even though the filesystem is not technically dirty yet.
4251  */
4252 static int ext4_unfreeze(struct super_block *sb)
4253 {
4254         if (sb->s_flags & MS_RDONLY)
4255                 return 0;
4256
4257         lock_super(sb);
4258         /* Reset the needs_recovery flag before the fs is unlocked. */
4259         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4260         ext4_commit_super(sb, 1);
4261         unlock_super(sb);
4262         return 0;
4263 }
4264
4265 /*
4266  * Structure to save mount options for ext4_remount's benefit
4267  */
4268 struct ext4_mount_options {
4269         unsigned long s_mount_opt;
4270         unsigned long s_mount_opt2;
4271         uid_t s_resuid;
4272         gid_t s_resgid;
4273         unsigned long s_commit_interval;
4274         u32 s_min_batch_time, s_max_batch_time;
4275 #ifdef CONFIG_QUOTA
4276         int s_jquota_fmt;
4277         char *s_qf_names[MAXQUOTAS];
4278 #endif
4279 };
4280
4281 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4282 {
4283         struct ext4_super_block *es;
4284         struct ext4_sb_info *sbi = EXT4_SB(sb);
4285         ext4_fsblk_t n_blocks_count = 0;
4286         unsigned long old_sb_flags;
4287         struct ext4_mount_options old_opts;
4288         int enable_quota = 0;
4289         ext4_group_t g;
4290         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4291         int err = 0;
4292 #ifdef CONFIG_QUOTA
4293         int i;
4294 #endif
4295         char *orig_data = kstrdup(data, GFP_KERNEL);
4296
4297         /* Store the original options */
4298         lock_super(sb);
4299         old_sb_flags = sb->s_flags;
4300         old_opts.s_mount_opt = sbi->s_mount_opt;
4301         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4302         old_opts.s_resuid = sbi->s_resuid;
4303         old_opts.s_resgid = sbi->s_resgid;
4304         old_opts.s_commit_interval = sbi->s_commit_interval;
4305         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4306         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4307 #ifdef CONFIG_QUOTA
4308         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4309         for (i = 0; i < MAXQUOTAS; i++)
4310                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4311 #endif
4312         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4313                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4314
4315         /*
4316          * Allow the "check" option to be passed as a remount option.
4317          */
4318         if (!parse_options(data, sb, NULL, &journal_ioprio,
4319                            &n_blocks_count, 1)) {
4320                 err = -EINVAL;
4321                 goto restore_opts;
4322         }
4323
4324         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4325                 ext4_abort(sb, "Abort forced by user");
4326
4327         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4328                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4329
4330         es = sbi->s_es;
4331
4332         if (sbi->s_journal) {
4333                 ext4_init_journal_params(sb, sbi->s_journal);
4334                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4335         }
4336
4337         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4338                 n_blocks_count > ext4_blocks_count(es)) {
4339                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4340                         err = -EROFS;
4341                         goto restore_opts;
4342                 }
4343
4344                 if (*flags & MS_RDONLY) {
4345                         err = dquot_suspend(sb, -1);
4346                         if (err < 0)
4347                                 goto restore_opts;
4348
4349                         /*
4350                          * First of all, the unconditional stuff we have to do
4351                          * to disable replay of the journal when we next remount
4352                          */
4353                         sb->s_flags |= MS_RDONLY;
4354
4355                         /*
4356                          * OK, test if we are remounting a valid rw partition
4357                          * readonly, and if so set the rdonly flag and then
4358                          * mark the partition as valid again.
4359                          */
4360                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4361                             (sbi->s_mount_state & EXT4_VALID_FS))
4362                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4363
4364                         if (sbi->s_journal)
4365                                 ext4_mark_recovery_complete(sb, es);
4366                 } else {
4367                         /* Make sure we can mount this feature set readwrite */
4368                         if (!ext4_feature_set_ok(sb, 0)) {
4369                                 err = -EROFS;
4370                                 goto restore_opts;
4371                         }
4372                         /*
4373                          * Make sure the group descriptor checksums
4374                          * are sane.  If they aren't, refuse to remount r/w.
4375                          */
4376                         for (g = 0; g < sbi->s_groups_count; g++) {
4377                                 struct ext4_group_desc *gdp =
4378                                         ext4_get_group_desc(sb, g, NULL);
4379
4380                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4381                                         ext4_msg(sb, KERN_ERR,
4382                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4383                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4384                                                le16_to_cpu(gdp->bg_checksum));
4385                                         err = -EINVAL;
4386                                         goto restore_opts;
4387                                 }
4388                         }
4389
4390                         /*
4391                          * If we have an unprocessed orphan list hanging
4392                          * around from a previously readonly bdev mount,
4393                          * require a full umount/remount for now.
4394                          */
4395                         if (es->s_last_orphan) {
4396                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4397                                        "remount RDWR because of unprocessed "
4398                                        "orphan inode list.  Please "
4399                                        "umount/remount instead");
4400                                 err = -EINVAL;
4401                                 goto restore_opts;
4402                         }
4403
4404                         /*
4405                          * Mounting a RDONLY partition read-write, so reread
4406                          * and store the current valid flag.  (It may have
4407                          * been changed by e2fsck since we originally mounted
4408                          * the partition.)
4409                          */
4410                         if (sbi->s_journal)
4411                                 ext4_clear_journal_err(sb, es);
4412                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4413                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4414                                 goto restore_opts;
4415                         if (!ext4_setup_super(sb, es, 0))
4416                                 sb->s_flags &= ~MS_RDONLY;
4417                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4418                                                      EXT4_FEATURE_INCOMPAT_MMP))
4419                                 if (ext4_multi_mount_protect(sb,
4420                                                 le64_to_cpu(es->s_mmp_block))) {
4421                                         err = -EROFS;
4422                                         goto restore_opts;
4423                                 }
4424                         enable_quota = 1;
4425                 }
4426         }
4427
4428         /*
4429          * Reinitialize lazy itable initialization thread based on
4430          * current settings
4431          */
4432         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4433                 ext4_unregister_li_request(sb);
4434         else {
4435                 ext4_group_t first_not_zeroed;
4436                 first_not_zeroed = ext4_has_uninit_itable(sb);
4437                 ext4_register_li_request(sb, first_not_zeroed);
4438         }
4439
4440         ext4_setup_system_zone(sb);
4441         if (sbi->s_journal == NULL)
4442                 ext4_commit_super(sb, 1);
4443
4444 #ifdef CONFIG_QUOTA
4445         /* Release old quota file names */
4446         for (i = 0; i < MAXQUOTAS; i++)
4447                 if (old_opts.s_qf_names[i] &&
4448                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4449                         kfree(old_opts.s_qf_names[i]);
4450 #endif
4451         unlock_super(sb);
4452         if (enable_quota)
4453                 dquot_resume(sb, -1);
4454
4455         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4456         kfree(orig_data);
4457         return 0;
4458
4459 restore_opts:
4460         sb->s_flags = old_sb_flags;
4461         sbi->s_mount_opt = old_opts.s_mount_opt;
4462         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4463         sbi->s_resuid = old_opts.s_resuid;
4464         sbi->s_resgid = old_opts.s_resgid;
4465         sbi->s_commit_interval = old_opts.s_commit_interval;
4466         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4467         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4468 #ifdef CONFIG_QUOTA
4469         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4470         for (i = 0; i < MAXQUOTAS; i++) {
4471                 if (sbi->s_qf_names[i] &&
4472                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4473                         kfree(sbi->s_qf_names[i]);
4474                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4475         }
4476 #endif
4477         unlock_super(sb);
4478         kfree(orig_data);
4479         return err;
4480 }
4481
4482 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4483 {
4484         struct super_block *sb = dentry->d_sb;
4485         struct ext4_sb_info *sbi = EXT4_SB(sb);
4486         struct ext4_super_block *es = sbi->s_es;
4487         u64 fsid;
4488         s64 bfree;
4489
4490         if (test_opt(sb, MINIX_DF)) {
4491                 sbi->s_overhead_last = 0;
4492         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4493                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4494                 ext4_fsblk_t overhead = 0;
4495
4496                 /*
4497                  * Compute the overhead (FS structures).  This is constant
4498                  * for a given filesystem unless the number of block groups
4499                  * changes so we cache the previous value until it does.
4500                  */
4501
4502                 /*
4503                  * All of the blocks before first_data_block are
4504                  * overhead
4505                  */
4506                 overhead = le32_to_cpu(es->s_first_data_block);
4507
4508                 /*
4509                  * Add the overhead attributed to the superblock and
4510                  * block group descriptors.  If the sparse superblocks
4511                  * feature is turned on, then not all groups have this.
4512                  */
4513                 for (i = 0; i < ngroups; i++) {
4514                         overhead += ext4_bg_has_super(sb, i) +
4515                                 ext4_bg_num_gdb(sb, i);
4516                         cond_resched();
4517                 }
4518
4519                 /*
4520                  * Every block group has an inode bitmap, a block
4521                  * bitmap, and an inode table.
4522                  */
4523                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4524                 sbi->s_overhead_last = overhead;
4525                 smp_wmb();
4526                 sbi->s_blocks_last = ext4_blocks_count(es);
4527         }
4528
4529         buf->f_type = EXT4_SUPER_MAGIC;
4530         buf->f_bsize = sb->s_blocksize;
4531         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4532         bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4533                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4534         /* prevent underflow in case that few free space is available */
4535         buf->f_bfree = max_t(s64, bfree, 0);
4536         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4537         if (buf->f_bfree < ext4_r_blocks_count(es))
4538                 buf->f_bavail = 0;
4539         buf->f_files = le32_to_cpu(es->s_inodes_count);
4540         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4541         buf->f_namelen = EXT4_NAME_LEN;
4542         fsid = le64_to_cpup((void *)es->s_uuid) ^
4543                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4544         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4545         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4546
4547         return 0;
4548 }
4549
4550 /* Helper function for writing quotas on sync - we need to start transaction
4551  * before quota file is locked for write. Otherwise the are possible deadlocks:
4552  * Process 1                         Process 2
4553  * ext4_create()                     quota_sync()
4554  *   jbd2_journal_start()                  write_dquot()
4555  *   dquot_initialize()                         down(dqio_mutex)
4556  *     down(dqio_mutex)                    jbd2_journal_start()
4557  *
4558  */
4559
4560 #ifdef CONFIG_QUOTA
4561
4562 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4563 {
4564         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4565 }
4566
4567 static int ext4_write_dquot(struct dquot *dquot)
4568 {
4569         int ret, err;
4570         handle_t *handle;
4571         struct inode *inode;
4572
4573         inode = dquot_to_inode(dquot);
4574         handle = ext4_journal_start(inode,
4575                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4576         if (IS_ERR(handle))
4577                 return PTR_ERR(handle);
4578         ret = dquot_commit(dquot);
4579         err = ext4_journal_stop(handle);
4580         if (!ret)
4581                 ret = err;
4582         return ret;
4583 }
4584
4585 static int ext4_acquire_dquot(struct dquot *dquot)
4586 {
4587         int ret, err;
4588         handle_t *handle;
4589
4590         handle = ext4_journal_start(dquot_to_inode(dquot),
4591                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4592         if (IS_ERR(handle))
4593                 return PTR_ERR(handle);
4594         ret = dquot_acquire(dquot);
4595         err = ext4_journal_stop(handle);
4596         if (!ret)
4597                 ret = err;
4598         return ret;
4599 }
4600
4601 static int ext4_release_dquot(struct dquot *dquot)
4602 {
4603         int ret, err;
4604         handle_t *handle;
4605
4606         handle = ext4_journal_start(dquot_to_inode(dquot),
4607                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4608         if (IS_ERR(handle)) {
4609                 /* Release dquot anyway to avoid endless cycle in dqput() */
4610                 dquot_release(dquot);
4611                 return PTR_ERR(handle);
4612         }
4613         ret = dquot_release(dquot);
4614         err = ext4_journal_stop(handle);
4615         if (!ret)
4616                 ret = err;
4617         return ret;
4618 }
4619
4620 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4621 {
4622         /* Are we journaling quotas? */
4623         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4624             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4625                 dquot_mark_dquot_dirty(dquot);
4626                 return ext4_write_dquot(dquot);
4627         } else {
4628                 return dquot_mark_dquot_dirty(dquot);
4629         }
4630 }
4631
4632 static int ext4_write_info(struct super_block *sb, int type)
4633 {
4634         int ret, err;
4635         handle_t *handle;
4636
4637         /* Data block + inode block */
4638         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4639         if (IS_ERR(handle))
4640                 return PTR_ERR(handle);
4641         ret = dquot_commit_info(sb, type);
4642         err = ext4_journal_stop(handle);
4643         if (!ret)
4644                 ret = err;
4645         return ret;
4646 }
4647
4648 /*
4649  * Turn on quotas during mount time - we need to find
4650  * the quota file and such...
4651  */
4652 static int ext4_quota_on_mount(struct super_block *sb, int type)
4653 {
4654         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4655                                         EXT4_SB(sb)->s_jquota_fmt, type);
4656 }
4657
4658 /*
4659  * Standard function to be called on quota_on
4660  */
4661 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4662                          struct path *path)
4663 {
4664         int err;
4665
4666         if (!test_opt(sb, QUOTA))
4667                 return -EINVAL;
4668
4669         /* Quotafile not on the same filesystem? */
4670         if (path->mnt->mnt_sb != sb)
4671                 return -EXDEV;
4672         /* Journaling quota? */
4673         if (EXT4_SB(sb)->s_qf_names[type]) {
4674                 /* Quotafile not in fs root? */
4675                 if (path->dentry->d_parent != sb->s_root)
4676                         ext4_msg(sb, KERN_WARNING,
4677                                 "Quota file not on filesystem root. "
4678                                 "Journaled quota will not work");
4679         }
4680
4681         /*
4682          * When we journal data on quota file, we have to flush journal to see
4683          * all updates to the file when we bypass pagecache...
4684          */
4685         if (EXT4_SB(sb)->s_journal &&
4686             ext4_should_journal_data(path->dentry->d_inode)) {
4687                 /*
4688                  * We don't need to lock updates but journal_flush() could
4689                  * otherwise be livelocked...
4690                  */
4691                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4692                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4693                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4694                 if (err)
4695                         return err;
4696         }
4697
4698         return dquot_quota_on(sb, type, format_id, path);
4699 }
4700
4701 static int ext4_quota_off(struct super_block *sb, int type)
4702 {
4703         struct inode *inode = sb_dqopt(sb)->files[type];
4704         handle_t *handle;
4705
4706         /* Force all delayed allocation blocks to be allocated.
4707          * Caller already holds s_umount sem */
4708         if (test_opt(sb, DELALLOC))
4709                 sync_filesystem(sb);
4710
4711         if (!inode)
4712                 goto out;
4713
4714         /* Update modification times of quota files when userspace can
4715          * start looking at them */
4716         handle = ext4_journal_start(inode, 1);
4717         if (IS_ERR(handle))
4718                 goto out;
4719         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4720         ext4_mark_inode_dirty(handle, inode);
4721         ext4_journal_stop(handle);
4722
4723 out:
4724         return dquot_quota_off(sb, type);
4725 }
4726
4727 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4728  * acquiring the locks... As quota files are never truncated and quota code
4729  * itself serializes the operations (and no one else should touch the files)
4730  * we don't have to be afraid of races */
4731 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4732                                size_t len, loff_t off)
4733 {
4734         struct inode *inode = sb_dqopt(sb)->files[type];
4735         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4736         int err = 0;
4737         int offset = off & (sb->s_blocksize - 1);
4738         int tocopy;
4739         size_t toread;
4740         struct buffer_head *bh;
4741         loff_t i_size = i_size_read(inode);
4742
4743         if (off > i_size)
4744                 return 0;
4745         if (off+len > i_size)
4746                 len = i_size-off;
4747         toread = len;
4748         while (toread > 0) {
4749                 tocopy = sb->s_blocksize - offset < toread ?
4750                                 sb->s_blocksize - offset : toread;
4751                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4752                 if (err)
4753                         return err;
4754                 if (!bh)        /* A hole? */
4755                         memset(data, 0, tocopy);
4756                 else
4757                         memcpy(data, bh->b_data+offset, tocopy);
4758                 brelse(bh);
4759                 offset = 0;
4760                 toread -= tocopy;
4761                 data += tocopy;
4762                 blk++;
4763         }
4764         return len;
4765 }
4766
4767 /* Write to quotafile (we know the transaction is already started and has
4768  * enough credits) */
4769 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4770                                 const char *data, size_t len, loff_t off)
4771 {
4772         struct inode *inode = sb_dqopt(sb)->files[type];
4773         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4774         int err = 0;
4775         int offset = off & (sb->s_blocksize - 1);
4776         struct buffer_head *bh;
4777         handle_t *handle = journal_current_handle();
4778
4779         if (EXT4_SB(sb)->s_journal && !handle) {
4780                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4781                         " cancelled because transaction is not started",
4782                         (unsigned long long)off, (unsigned long long)len);
4783                 return -EIO;
4784         }
4785         /*
4786          * Since we account only one data block in transaction credits,
4787          * then it is impossible to cross a block boundary.
4788          */
4789         if (sb->s_blocksize - offset < len) {
4790                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4791                         " cancelled because not block aligned",
4792                         (unsigned long long)off, (unsigned long long)len);
4793                 return -EIO;
4794         }
4795
4796         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4797         bh = ext4_bread(handle, inode, blk, 1, &err);
4798         if (!bh)
4799                 goto out;
4800         err = ext4_journal_get_write_access(handle, bh);
4801         if (err) {
4802                 brelse(bh);
4803                 goto out;
4804         }
4805         lock_buffer(bh);
4806         memcpy(bh->b_data+offset, data, len);
4807         flush_dcache_page(bh->b_page);
4808         unlock_buffer(bh);
4809         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4810         brelse(bh);
4811 out:
4812         if (err) {
4813                 mutex_unlock(&inode->i_mutex);
4814                 return err;
4815         }
4816         if (inode->i_size < off + len) {
4817                 i_size_write(inode, off + len);
4818                 EXT4_I(inode)->i_disksize = inode->i_size;
4819                 ext4_mark_inode_dirty(handle, inode);
4820         }
4821         mutex_unlock(&inode->i_mutex);
4822         return len;
4823 }
4824
4825 #endif
4826
4827 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4828                        const char *dev_name, void *data)
4829 {
4830         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4831 }
4832
4833 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4834 static inline void register_as_ext2(void)
4835 {
4836         int err = register_filesystem(&ext2_fs_type);
4837         if (err)
4838                 printk(KERN_WARNING
4839                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4840 }
4841
4842 static inline void unregister_as_ext2(void)
4843 {
4844         unregister_filesystem(&ext2_fs_type);
4845 }
4846
4847 static inline int ext2_feature_set_ok(struct super_block *sb)
4848 {
4849         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4850                 return 0;
4851         if (sb->s_flags & MS_RDONLY)
4852                 return 1;
4853         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4854                 return 0;
4855         return 1;
4856 }
4857 MODULE_ALIAS("ext2");
4858 #else
4859 static inline void register_as_ext2(void) { }
4860 static inline void unregister_as_ext2(void) { }
4861 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4862 #endif
4863
4864 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4865 static inline void register_as_ext3(void)
4866 {
4867         int err = register_filesystem(&ext3_fs_type);
4868         if (err)
4869                 printk(KERN_WARNING
4870                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4871 }
4872
4873 static inline void unregister_as_ext3(void)
4874 {
4875         unregister_filesystem(&ext3_fs_type);
4876 }
4877
4878 static inline int ext3_feature_set_ok(struct super_block *sb)
4879 {
4880         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4881                 return 0;
4882         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4883                 return 0;
4884         if (sb->s_flags & MS_RDONLY)
4885                 return 1;
4886         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4887                 return 0;
4888         return 1;
4889 }
4890 MODULE_ALIAS("ext3");
4891 #else
4892 static inline void register_as_ext3(void) { }
4893 static inline void unregister_as_ext3(void) { }
4894 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4895 #endif
4896
4897 static struct file_system_type ext4_fs_type = {
4898         .owner          = THIS_MODULE,
4899         .name           = "ext4",
4900         .mount          = ext4_mount,
4901         .kill_sb        = kill_block_super,
4902         .fs_flags       = FS_REQUIRES_DEV,
4903 };
4904
4905 static int __init ext4_init_feat_adverts(void)
4906 {
4907         struct ext4_features *ef;
4908         int ret = -ENOMEM;
4909
4910         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4911         if (!ef)
4912                 goto out;
4913
4914         ef->f_kobj.kset = ext4_kset;
4915         init_completion(&ef->f_kobj_unregister);
4916         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4917                                    "features");
4918         if (ret) {
4919                 kfree(ef);
4920                 goto out;
4921         }
4922
4923         ext4_feat = ef;
4924         ret = 0;
4925 out:
4926         return ret;
4927 }
4928
4929 static void ext4_exit_feat_adverts(void)
4930 {
4931         kobject_put(&ext4_feat->f_kobj);
4932         wait_for_completion(&ext4_feat->f_kobj_unregister);
4933         kfree(ext4_feat);
4934 }
4935
4936 /* Shared across all ext4 file systems */
4937 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4938 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4939
4940 static int __init ext4_init_fs(void)
4941 {
4942         int i, err;
4943
4944         ext4_check_flag_values();
4945
4946         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4947                 mutex_init(&ext4__aio_mutex[i]);
4948                 init_waitqueue_head(&ext4__ioend_wq[i]);
4949         }
4950
4951         err = ext4_init_pageio();
4952         if (err)
4953                 return err;
4954         err = ext4_init_system_zone();
4955         if (err)
4956                 goto out7;
4957         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4958         if (!ext4_kset)
4959                 goto out6;
4960         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4961         if (!ext4_proc_root)
4962                 goto out5;
4963
4964         err = ext4_init_feat_adverts();
4965         if (err)
4966                 goto out4;
4967
4968         err = ext4_init_mballoc();
4969         if (err)
4970                 goto out3;
4971
4972         err = ext4_init_xattr();
4973         if (err)
4974                 goto out2;
4975         err = init_inodecache();
4976         if (err)
4977                 goto out1;
4978         register_as_ext3();
4979         register_as_ext2();
4980         err = register_filesystem(&ext4_fs_type);
4981         if (err)
4982                 goto out;
4983
4984         ext4_li_info = NULL;
4985         mutex_init(&ext4_li_mtx);
4986         return 0;
4987 out:
4988         unregister_as_ext2();
4989         unregister_as_ext3();
4990         destroy_inodecache();
4991 out1:
4992         ext4_exit_xattr();
4993 out2:
4994         ext4_exit_mballoc();
4995 out3:
4996         ext4_exit_feat_adverts();
4997 out4:
4998         remove_proc_entry("fs/ext4", NULL);
4999 out5:
5000         kset_unregister(ext4_kset);
5001 out6:
5002         ext4_exit_system_zone();
5003 out7:
5004         ext4_exit_pageio();
5005         return err;
5006 }
5007
5008 static void __exit ext4_exit_fs(void)
5009 {
5010         ext4_destroy_lazyinit_thread();
5011         unregister_as_ext2();
5012         unregister_as_ext3();
5013         unregister_filesystem(&ext4_fs_type);
5014         destroy_inodecache();
5015         ext4_exit_xattr();
5016         ext4_exit_mballoc();
5017         ext4_exit_feat_adverts();
5018         remove_proc_entry("fs/ext4", NULL);
5019         kset_unregister(ext4_kset);
5020         ext4_exit_system_zone();
5021         ext4_exit_pageio();
5022 }
5023
5024 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5025 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5026 MODULE_LICENSE("GPL");
5027 module_init(ext4_init_fs)
5028 module_exit(ext4_exit_fs)