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 <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_jbd2.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "mballoc.h"
  51
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/ext4.h>
  54
  55 static struct proc_dir_entry *ext4_proc_root;
  56 static struct kset *ext4_kset;
  57 static struct ext4_lazy_init *ext4_li_info;
  58 static struct mutex ext4_li_mtx;
  59 static struct ext4_features *ext4_feat;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static const char *ext4_decode_error(struct super_block *sb, int errno,
  70                                      char nbuf[16]);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static void ext4_write_super(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static inline int ext2_feature_set_ok(struct super_block *sb);
  79 static inline int ext3_feature_set_ok(struct super_block *sb);
  80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  81 static void ext4_destroy_lazyinit_thread(void);
  82 static void ext4_unregister_li_request(struct super_block *sb);
  83 static void ext4_clear_request_list(void);
  84
  85 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  86 static struct file_system_type ext2_fs_type = {
  87         .owner          = THIS_MODULE,
  88         .name           = "ext2",
  89         .mount          = ext4_mount,
  90         .kill_sb        = kill_block_super,
  91         .fs_flags       = FS_REQUIRES_DEV,
  92 };
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 100 static struct file_system_type ext3_fs_type = {
 101         .owner          = THIS_MODULE,
 102         .name           = "ext3",
 103         .mount          = ext4_mount,
 104         .kill_sb        = kill_block_super,
 105         .fs_flags       = FS_REQUIRES_DEV,
 106 };
 107 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 108 #else
 109 #define IS_EXT3_SB(sb) (0)
 110 #endif
 111
 112 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 113                                struct ext4_group_desc *bg)
 114 {
 115         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 116                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 117                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 118 }
 119
 120 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 121                                struct ext4_group_desc *bg)
 122 {
 123         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 124                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 125                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 126 }
 127
 128 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 129                               struct ext4_group_desc *bg)
 130 {
 131         return le32_to_cpu(bg->bg_inode_table_lo) |
 132                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 133                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 134 }
 135
 136 __u32 ext4_free_blks_count(struct super_block *sb,
 137                               struct ext4_group_desc *bg)
 138 {
 139         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 140                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 141                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 142 }
 143
 144 __u32 ext4_free_inodes_count(struct super_block *sb,
 145                               struct ext4_group_desc *bg)
 146 {
 147         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 150 }
 151
 152 __u32 ext4_used_dirs_count(struct super_block *sb,
 153                               struct ext4_group_desc *bg)
 154 {
 155         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 156                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 157                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 158 }
 159
 160 __u32 ext4_itable_unused_count(struct super_block *sb,
 161                               struct ext4_group_desc *bg)
 162 {
 163         return le16_to_cpu(bg->bg_itable_unused_lo) |
 164                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 165                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 166 }
 167
 168 void ext4_block_bitmap_set(struct super_block *sb,
 169                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 170 {
 171         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 172         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 173                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 174 }
 175
 176 void ext4_inode_bitmap_set(struct super_block *sb,
 177                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 178 {
 179         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 180         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 181                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 182 }
 183
 184 void ext4_inode_table_set(struct super_block *sb,
 185                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 186 {
 187         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 188         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 189                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 190 }
 191
 192 void ext4_free_blks_set(struct super_block *sb,
 193                           struct ext4_group_desc *bg, __u32 count)
 194 {
 195         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 196         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 197                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 198 }
 199
 200 void ext4_free_inodes_set(struct super_block *sb,
 201                           struct ext4_group_desc *bg, __u32 count)
 202 {
 203         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 206 }
 207
 208 void ext4_used_dirs_set(struct super_block *sb,
 209                           struct ext4_group_desc *bg, __u32 count)
 210 {
 211         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 212         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 213                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 214 }
 215
 216 void ext4_itable_unused_set(struct super_block *sb,
 217                           struct ext4_group_desc *bg, __u32 count)
 218 {
 219         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 220         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 221                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 222 }
 223
 224
 225 /* Just increment the non-pointer handle value */
 226 static handle_t *ext4_get_nojournal(void)
 227 {
 228         handle_t *handle = current->journal_info;
 229         unsigned long ref_cnt = (unsigned long)handle;
 230
 231         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 232
 233         ref_cnt++;
 234         handle = (handle_t *)ref_cnt;
 235
 236         current->journal_info = handle;
 237         return handle;
 238 }
 239
 240
 241 /* Decrement the non-pointer handle value */
 242 static void ext4_put_nojournal(handle_t *handle)
 243 {
 244         unsigned long ref_cnt = (unsigned long)handle;
 245
 246         BUG_ON(ref_cnt == 0);
 247
 248         ref_cnt--;
 249         handle = (handle_t *)ref_cnt;
 250
 251         current->journal_info = handle;
 252 }
 253
 254 /*
 255  * Wrappers for jbd2_journal_start/end.
 256  *
 257  * The only special thing we need to do here is to make sure that all
 258  * journal_end calls result in the superblock being marked dirty, so
 259  * that sync() will call the filesystem's write_super callback if
 260  * appropriate.
 261  *
 262  * To avoid j_barrier hold in userspace when a user calls freeze(),
 263  * ext4 prevents a new handle from being started by s_frozen, which
 264  * is in an upper layer.
 265  */
 266 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 267 {
 268         journal_t *journal;
 269         handle_t  *handle;
 270
 271         if (sb->s_flags & MS_RDONLY)
 272                 return ERR_PTR(-EROFS);
 273
 274         journal = EXT4_SB(sb)->s_journal;
 275         handle = ext4_journal_current_handle();
 276
 277         /*
 278          * If a handle has been started, it should be allowed to
 279          * finish, otherwise deadlock could happen between freeze
 280          * and others(e.g. truncate) due to the restart of the
 281          * journal handle if the filesystem is forzen and active
 282          * handles are not stopped.
 283          */
 284         if (!handle)
 285                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 286
 287         if (!journal)
 288                 return ext4_get_nojournal();
 289         /*
 290          * Special case here: if the journal has aborted behind our
 291          * backs (eg. EIO in the commit thread), then we still need to
 292          * take the FS itself readonly cleanly.
 293          */
 294         if (is_journal_aborted(journal)) {
 295                 ext4_abort(sb, "Detected aborted journal");
 296                 return ERR_PTR(-EROFS);
 297         }
 298         return jbd2_journal_start(journal, nblocks);
 299 }
 300
 301 /*
 302  * The only special thing we need to do here is to make sure that all
 303  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 304  * that sync() will call the filesystem's write_super callback if
 305  * appropriate.
 306  */
 307 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 308 {
 309         struct super_block *sb;
 310         int err;
 311         int rc;
 312
 313         if (!ext4_handle_valid(handle)) {
 314                 ext4_put_nojournal(handle);
 315                 return 0;
 316         }
 317         sb = handle->h_transaction->t_journal->j_private;
 318         err = handle->h_err;
 319         rc = jbd2_journal_stop(handle);
 320
 321         if (!err)
 322                 err = rc;
 323         if (err)
 324                 __ext4_std_error(sb, where, line, err);
 325         return err;
 326 }
 327
 328 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 329                                const char *err_fn, struct buffer_head *bh,
 330                                handle_t *handle, int err)
 331 {
 332         char nbuf[16];
 333         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 334
 335         BUG_ON(!ext4_handle_valid(handle));
 336
 337         if (bh)
 338                 BUFFER_TRACE(bh, "abort");
 339
 340         if (!handle->h_err)
 341                 handle->h_err = err;
 342
 343         if (is_handle_aborted(handle))
 344                 return;
 345
 346         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 347                caller, line, errstr, err_fn);
 348
 349         jbd2_journal_abort_handle(handle);
 350 }
 351
 352 static void __save_error_info(struct super_block *sb, const char *func,
 353                             unsigned int line)
 354 {
 355         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 356
 357         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 358         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 359         es->s_last_error_time = cpu_to_le32(get_seconds());
 360         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 361         es->s_last_error_line = cpu_to_le32(line);
 362         if (!es->s_first_error_time) {
 363                 es->s_first_error_time = es->s_last_error_time;
 364                 strncpy(es->s_first_error_func, func,
 365                         sizeof(es->s_first_error_func));
 366                 es->s_first_error_line = cpu_to_le32(line);
 367                 es->s_first_error_ino = es->s_last_error_ino;
 368                 es->s_first_error_block = es->s_last_error_block;
 369         }
 370         /*
 371          * Start the daily error reporting function if it hasn't been
 372          * started already
 373          */
 374         if (!es->s_error_count)
 375                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 376         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 377 }
 378
 379 static void save_error_info(struct super_block *sb, const char *func,
 380                             unsigned int line)
 381 {
 382         __save_error_info(sb, func, line);
 383         ext4_commit_super(sb, 1);
 384 }
 385
 386
 387 /* Deal with the reporting of failure conditions on a filesystem such as
 388  * inconsistencies detected or read IO failures.
 389  *
 390  * On ext2, we can store the error state of the filesystem in the
 391  * superblock.  That is not possible on ext4, because we may have other
 392  * write ordering constraints on the superblock which prevent us from
 393  * writing it out straight away; and given that the journal is about to
 394  * be aborted, we can't rely on the current, or future, transactions to
 395  * write out the superblock safely.
 396  *
 397  * We'll just use the jbd2_journal_abort() error code to record an error in
 398  * the journal instead.  On recovery, the journal will complain about
 399  * that error until we've noted it down and cleared it.
 400  */
 401
 402 static void ext4_handle_error(struct super_block *sb)
 403 {
 404         if (sb->s_flags & MS_RDONLY)
 405                 return;
 406
 407         if (!test_opt(sb, ERRORS_CONT)) {
 408                 journal_t *journal = EXT4_SB(sb)->s_journal;
 409
 410                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 411                 if (journal)
 412                         jbd2_journal_abort(journal, -EIO);
 413         }
 414         if (test_opt(sb, ERRORS_RO)) {
 415                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 416                 sb->s_flags |= MS_RDONLY;
 417         }
 418         if (test_opt(sb, ERRORS_PANIC))
 419                 panic("EXT4-fs (device %s): panic forced after error\n",
 420                         sb->s_id);
 421 }
 422
 423 void __ext4_error(struct super_block *sb, const char *function,
 424                   unsigned int line, const char *fmt, ...)
 425 {
 426         struct va_format vaf;
 427         va_list args;
 428
 429         va_start(args, fmt);
 430         vaf.fmt = fmt;
 431         vaf.va = &args;
 432         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 433                sb->s_id, function, line, current->comm, &vaf);
 434         va_end(args);
 435
 436         ext4_handle_error(sb);
 437 }
 438
 439 void ext4_error_inode(struct inode *inode, const char *function,
 440                       unsigned int line, ext4_fsblk_t block,
 441                       const char *fmt, ...)
 442 {
 443         va_list args;
 444         struct va_format vaf;
 445         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 446
 447         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 448         es->s_last_error_block = cpu_to_le64(block);
 449         save_error_info(inode->i_sb, function, line);
 450         va_start(args, fmt);
 451         vaf.fmt = fmt;
 452         vaf.va = &args;
 453         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 454                inode->i_sb->s_id, function, line, inode->i_ino);
 455         if (block)
 456                 printk(KERN_CONT "block %llu: ", block);
 457         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 458         va_end(args);
 459
 460         ext4_handle_error(inode->i_sb);
 461 }
 462
 463 void ext4_error_file(struct file *file, const char *function,
 464                      unsigned int line, ext4_fsblk_t block,
 465                      const char *fmt, ...)
 466 {
 467         va_list args;
 468         struct va_format vaf;
 469         struct ext4_super_block *es;
 470         struct inode *inode = file->f_dentry->d_inode;
 471         char pathname[80], *path;
 472
 473         es = EXT4_SB(inode->i_sb)->s_es;
 474         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 475         save_error_info(inode->i_sb, function, line);
 476         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 477         if (IS_ERR(path))
 478                 path = "(unknown)";
 479         printk(KERN_CRIT
 480                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 481                inode->i_sb->s_id, function, line, inode->i_ino);
 482         if (block)
 483                 printk(KERN_CONT "block %llu: ", block);
 484         va_start(args, fmt);
 485         vaf.fmt = fmt;
 486         vaf.va = &args;
 487         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 488         va_end(args);
 489
 490         ext4_handle_error(inode->i_sb);
 491 }
 492
 493 static const char *ext4_decode_error(struct super_block *sb, int errno,
 494                                      char nbuf[16])
 495 {
 496         char *errstr = NULL;
 497
 498         switch (errno) {
 499         case -EIO:
 500                 errstr = "IO failure";
 501                 break;
 502         case -ENOMEM:
 503                 errstr = "Out of memory";
 504                 break;
 505         case -EROFS:
 506                 if (!sb || (EXT4_SB(sb)->s_journal &&
 507                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 508                         errstr = "Journal has aborted";
 509                 else
 510                         errstr = "Readonly filesystem";
 511                 break;
 512         default:
 513                 /* If the caller passed in an extra buffer for unknown
 514                  * errors, textualise them now.  Else we just return
 515                  * NULL. */
 516                 if (nbuf) {
 517                         /* Check for truncated error codes... */
 518                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 519                                 errstr = nbuf;
 520                 }
 521                 break;
 522         }
 523
 524         return errstr;
 525 }
 526
 527 /* __ext4_std_error decodes expected errors from journaling functions
 528  * automatically and invokes the appropriate error response.  */
 529
 530 void __ext4_std_error(struct super_block *sb, const char *function,
 531                       unsigned int line, int errno)
 532 {
 533         char nbuf[16];
 534         const char *errstr;
 535
 536         /* Special case: if the error is EROFS, and we're not already
 537          * inside a transaction, then there's really no point in logging
 538          * an error. */
 539         if (errno == -EROFS && journal_current_handle() == NULL &&
 540             (sb->s_flags & MS_RDONLY))
 541                 return;
 542
 543         errstr = ext4_decode_error(sb, errno, nbuf);
 544         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 545                sb->s_id, function, line, errstr);
 546         save_error_info(sb, function, line);
 547
 548         ext4_handle_error(sb);
 549 }
 550
 551 /*
 552  * ext4_abort is a much stronger failure handler than ext4_error.  The
 553  * abort function may be used to deal with unrecoverable failures such
 554  * as journal IO errors or ENOMEM at a critical moment in log management.
 555  *
 556  * We unconditionally force the filesystem into an ABORT|READONLY state,
 557  * unless the error response on the fs has been set to panic in which
 558  * case we take the easy way out and panic immediately.
 559  */
 560
 561 void __ext4_abort(struct super_block *sb, const char *function,
 562                 unsigned int line, const char *fmt, ...)
 563 {
 564         va_list args;
 565
 566         save_error_info(sb, function, line);
 567         va_start(args, fmt);
 568         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 569                function, line);
 570         vprintk(fmt, args);
 571         printk("\n");
 572         va_end(args);
 573
 574         if ((sb->s_flags & MS_RDONLY) == 0) {
 575                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 576                 sb->s_flags |= MS_RDONLY;
 577                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 578                 if (EXT4_SB(sb)->s_journal)
 579                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 580                 save_error_info(sb, function, line);
 581         }
 582         if (test_opt(sb, ERRORS_PANIC))
 583                 panic("EXT4-fs panic from previous error\n");
 584 }
 585
 586 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 587 {
 588         struct va_format vaf;
 589         va_list args;
 590
 591         va_start(args, fmt);
 592         vaf.fmt = fmt;
 593         vaf.va = &args;
 594         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 595         va_end(args);
 596 }
 597
 598 void __ext4_warning(struct super_block *sb, const char *function,
 599                     unsigned int line, const char *fmt, ...)
 600 {
 601         struct va_format vaf;
 602         va_list args;
 603
 604         va_start(args, fmt);
 605         vaf.fmt = fmt;
 606         vaf.va = &args;
 607         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 608                sb->s_id, function, line, &vaf);
 609         va_end(args);
 610 }
 611
 612 void __ext4_grp_locked_error(const char *function, unsigned int line,
 613                              struct super_block *sb, ext4_group_t grp,
 614                              unsigned long ino, ext4_fsblk_t block,
 615                              const char *fmt, ...)
 616 __releases(bitlock)
 617 __acquires(bitlock)
 618 {
 619         struct va_format vaf;
 620         va_list args;
 621         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 622
 623         es->s_last_error_ino = cpu_to_le32(ino);
 624         es->s_last_error_block = cpu_to_le64(block);
 625         __save_error_info(sb, function, line);
 626
 627         va_start(args, fmt);
 628
 629         vaf.fmt = fmt;
 630         vaf.va = &args;
 631         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 632                sb->s_id, function, line, grp);
 633         if (ino)
 634                 printk(KERN_CONT "inode %lu: ", ino);
 635         if (block)
 636                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 637         printk(KERN_CONT "%pV\n", &vaf);
 638         va_end(args);
 639
 640         if (test_opt(sb, ERRORS_CONT)) {
 641                 ext4_commit_super(sb, 0);
 642                 return;
 643         }
 644
 645         ext4_unlock_group(sb, grp);
 646         ext4_handle_error(sb);
 647         /*
 648          * We only get here in the ERRORS_RO case; relocking the group
 649          * may be dangerous, but nothing bad will happen since the
 650          * filesystem will have already been marked read/only and the
 651          * journal has been aborted.  We return 1 as a hint to callers
 652          * who might what to use the return value from
 653          * ext4_grp_locked_error() to distinguish between the
 654          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 655          * aggressively from the ext4 function in question, with a
 656          * more appropriate error code.
 657          */
 658         ext4_lock_group(sb, grp);
 659         return;
 660 }
 661
 662 void ext4_update_dynamic_rev(struct super_block *sb)
 663 {
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 667                 return;
 668
 669         ext4_warning(sb,
 670                      "updating to rev %d because of new feature flag, "
 671                      "running e2fsck is recommended",
 672                      EXT4_DYNAMIC_REV);
 673
 674         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 675         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 676         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 677         /* leave es->s_feature_*compat flags alone */
 678         /* es->s_uuid will be set by e2fsck if empty */
 679
 680         /*
 681          * The rest of the superblock fields should be zero, and if not it
 682          * means they are likely already in use, so leave them alone.  We
 683          * can leave it up to e2fsck to clean up any inconsistencies there.
 684          */
 685 }
 686
 687 /*
 688  * Open the external journal device
 689  */
 690 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 691 {
 692         struct block_device *bdev;
 693         char b[BDEVNAME_SIZE];
 694
 695         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 696         if (IS_ERR(bdev))
 697                 goto fail;
 698         return bdev;
 699
 700 fail:
 701         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 702                         __bdevname(dev, b), PTR_ERR(bdev));
 703         return NULL;
 704 }
 705
 706 /*
 707  * Release the journal device
 708  */
 709 static int ext4_blkdev_put(struct block_device *bdev)
 710 {
 711         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 712 }
 713
 714 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 715 {
 716         struct block_device *bdev;
 717         int ret = -ENODEV;
 718
 719         bdev = sbi->journal_bdev;
 720         if (bdev) {
 721                 ret = ext4_blkdev_put(bdev);
 722                 sbi->journal_bdev = NULL;
 723         }
 724         return ret;
 725 }
 726
 727 static inline struct inode *orphan_list_entry(struct list_head *l)
 728 {
 729         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 730 }
 731
 732 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 733 {
 734         struct list_head *l;
 735
 736         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 737                  le32_to_cpu(sbi->s_es->s_last_orphan));
 738
 739         printk(KERN_ERR "sb_info orphan list:\n");
 740         list_for_each(l, &sbi->s_orphan) {
 741                 struct inode *inode = orphan_list_entry(l);
 742                 printk(KERN_ERR "  "
 743                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 744                        inode->i_sb->s_id, inode->i_ino, inode,
 745                        inode->i_mode, inode->i_nlink,
 746                        NEXT_ORPHAN(inode));
 747         }
 748 }
 749
 750 static void ext4_put_super(struct super_block *sb)
 751 {
 752         struct ext4_sb_info *sbi = EXT4_SB(sb);
 753         struct ext4_super_block *es = sbi->s_es;
 754         int i, err;
 755
 756         ext4_unregister_li_request(sb);
 757         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 758
 759         flush_workqueue(sbi->dio_unwritten_wq);
 760         destroy_workqueue(sbi->dio_unwritten_wq);
 761
 762         lock_super(sb);
 763         if (sb->s_dirt)
 764                 ext4_commit_super(sb, 1);
 765
 766         if (sbi->s_journal) {
 767                 err = jbd2_journal_destroy(sbi->s_journal);
 768                 sbi->s_journal = NULL;
 769                 if (err < 0)
 770                         ext4_abort(sb, "Couldn't clean up the journal");
 771         }
 772
 773         del_timer(&sbi->s_err_report);
 774         ext4_release_system_zone(sb);
 775         ext4_mb_release(sb);
 776         ext4_ext_release(sb);
 777         ext4_xattr_put_super(sb);
 778
 779         if (!(sb->s_flags & MS_RDONLY)) {
 780                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 781                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 782                 ext4_commit_super(sb, 1);
 783         }
 784         if (sbi->s_proc) {
 785                 remove_proc_entry(sb->s_id, ext4_proc_root);
 786         }
 787         kobject_del(&sbi->s_kobj);
 788
 789         for (i = 0; i < sbi->s_gdb_count; i++)
 790                 brelse(sbi->s_group_desc[i]);
 791         kfree(sbi->s_group_desc);
 792         if (is_vmalloc_addr(sbi->s_flex_groups))
 793                 vfree(sbi->s_flex_groups);
 794         else
 795                 kfree(sbi->s_flex_groups);
 796         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 797         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 798         percpu_counter_destroy(&sbi->s_dirs_counter);
 799         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 800         brelse(sbi->s_sbh);
 801 #ifdef CONFIG_QUOTA
 802         for (i = 0; i < MAXQUOTAS; i++)
 803                 kfree(sbi->s_qf_names[i]);
 804 #endif
 805
 806         /* Debugging code just in case the in-memory inode orphan list
 807          * isn't empty.  The on-disk one can be non-empty if we've
 808          * detected an error and taken the fs readonly, but the
 809          * in-memory list had better be clean by this point. */
 810         if (!list_empty(&sbi->s_orphan))
 811                 dump_orphan_list(sb, sbi);
 812         J_ASSERT(list_empty(&sbi->s_orphan));
 813
 814         invalidate_bdev(sb->s_bdev);
 815         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 816                 /*
 817                  * Invalidate the journal device's buffers.  We don't want them
 818                  * floating about in memory - the physical journal device may
 819                  * hotswapped, and it breaks the `ro-after' testing code.
 820                  */
 821                 sync_blockdev(sbi->journal_bdev);
 822                 invalidate_bdev(sbi->journal_bdev);
 823                 ext4_blkdev_remove(sbi);
 824         }
 825         if (sbi->s_mmp_tsk)
 826                 kthread_stop(sbi->s_mmp_tsk);
 827         sb->s_fs_info = NULL;
 828         /*
 829          * Now that we are completely done shutting down the
 830          * superblock, we need to actually destroy the kobject.
 831          */
 832         unlock_super(sb);
 833         kobject_put(&sbi->s_kobj);
 834         wait_for_completion(&sbi->s_kobj_unregister);
 835         kfree(sbi->s_blockgroup_lock);
 836         kfree(sbi);
 837 }
 838
 839 static struct kmem_cache *ext4_inode_cachep;
 840
 841 /*
 842  * Called inside transaction, so use GFP_NOFS
 843  */
 844 static struct inode *ext4_alloc_inode(struct super_block *sb)
 845 {
 846         struct ext4_inode_info *ei;
 847
 848         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 849         if (!ei)
 850                 return NULL;
 851
 852         ei->vfs_inode.i_version = 1;
 853         ei->vfs_inode.i_data.writeback_index = 0;
 854         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 855         INIT_LIST_HEAD(&ei->i_prealloc_list);
 856         spin_lock_init(&ei->i_prealloc_lock);
 857         ei->i_reserved_data_blocks = 0;
 858         ei->i_reserved_meta_blocks = 0;
 859         ei->i_allocated_meta_blocks = 0;
 860         ei->i_da_metadata_calc_len = 0;
 861         spin_lock_init(&(ei->i_block_reservation_lock));
 862 #ifdef CONFIG_QUOTA
 863         ei->i_reserved_quota = 0;
 864 #endif
 865         ei->jinode = NULL;
 866         INIT_LIST_HEAD(&ei->i_completed_io_list);
 867         spin_lock_init(&ei->i_completed_io_lock);
 868         ei->cur_aio_dio = NULL;
 869         ei->i_sync_tid = 0;
 870         ei->i_datasync_tid = 0;
 871         atomic_set(&ei->i_ioend_count, 0);
 872         atomic_set(&ei->i_aiodio_unwritten, 0);
 873
 874         return &ei->vfs_inode;
 875 }
 876
 877 static int ext4_drop_inode(struct inode *inode)
 878 {
 879         int drop = generic_drop_inode(inode);
 880
 881         trace_ext4_drop_inode(inode, drop);
 882         return drop;
 883 }
 884
 885 static void ext4_i_callback(struct rcu_head *head)
 886 {
 887         struct inode *inode = container_of(head, struct inode, i_rcu);
 888         INIT_LIST_HEAD(&inode->i_dentry);
 889         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 890 }
 891
 892 static void ext4_destroy_inode(struct inode *inode)
 893 {
 894         ext4_ioend_wait(inode);
 895         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 896                 ext4_msg(inode->i_sb, KERN_ERR,
 897                          "Inode %lu (%p): orphan list check failed!",
 898                          inode->i_ino, EXT4_I(inode));
 899                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 900                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 901                                 true);
 902                 dump_stack();
 903         }
 904         call_rcu(&inode->i_rcu, ext4_i_callback);
 905 }
 906
 907 static void init_once(void *foo)
 908 {
 909         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 910
 911         INIT_LIST_HEAD(&ei->i_orphan);
 912 #ifdef CONFIG_EXT4_FS_XATTR
 913         init_rwsem(&ei->xattr_sem);
 914 #endif
 915         init_rwsem(&ei->i_data_sem);
 916         inode_init_once(&ei->vfs_inode);
 917 }
 918
 919 static int init_inodecache(void)
 920 {
 921         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 922                                              sizeof(struct ext4_inode_info),
 923                                              0, (SLAB_RECLAIM_ACCOUNT|
 924                                                 SLAB_MEM_SPREAD),
 925                                              init_once);
 926         if (ext4_inode_cachep == NULL)
 927                 return -ENOMEM;
 928         return 0;
 929 }
 930
 931 static void destroy_inodecache(void)
 932 {
 933         kmem_cache_destroy(ext4_inode_cachep);
 934 }
 935
 936 void ext4_clear_inode(struct inode *inode)
 937 {
 938         invalidate_inode_buffers(inode);
 939         end_writeback(inode);
 940         dquot_drop(inode);
 941         ext4_discard_preallocations(inode);
 942         if (EXT4_I(inode)->jinode) {
 943                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 944                                                EXT4_I(inode)->jinode);
 945                 jbd2_free_inode(EXT4_I(inode)->jinode);
 946                 EXT4_I(inode)->jinode = NULL;
 947         }
 948 }
 949
 950 static inline void ext4_show_quota_options(struct seq_file *seq,
 951                                            struct super_block *sb)
 952 {
 953 #if defined(CONFIG_QUOTA)
 954         struct ext4_sb_info *sbi = EXT4_SB(sb);
 955
 956         if (sbi->s_jquota_fmt) {
 957                 char *fmtname = "";
 958
 959                 switch (sbi->s_jquota_fmt) {
 960                 case QFMT_VFS_OLD:
 961                         fmtname = "vfsold";
 962                         break;
 963                 case QFMT_VFS_V0:
 964                         fmtname = "vfsv0";
 965                         break;
 966                 case QFMT_VFS_V1:
 967                         fmtname = "vfsv1";
 968                         break;
 969                 }
 970                 seq_printf(seq, ",jqfmt=%s", fmtname);
 971         }
 972
 973         if (sbi->s_qf_names[USRQUOTA])
 974                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 975
 976         if (sbi->s_qf_names[GRPQUOTA])
 977                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 978
 979         if (test_opt(sb, USRQUOTA))
 980                 seq_puts(seq, ",usrquota");
 981
 982         if (test_opt(sb, GRPQUOTA))
 983                 seq_puts(seq, ",grpquota");
 984 #endif
 985 }
 986
 987 /*
 988  * Show an option if
 989  *  - it's set to a non-default value OR
 990  *  - if the per-sb default is different from the global default
 991  */
 992 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 993 {
 994         int def_errors;
 995         unsigned long def_mount_opts;
 996         struct super_block *sb = vfs->mnt_sb;
 997         struct ext4_sb_info *sbi = EXT4_SB(sb);
 998         struct ext4_super_block *es = sbi->s_es;
 999
1000         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1001         def_errors     = le16_to_cpu(es->s_errors);
1002
1003         if (sbi->s_sb_block != 1)
1004                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1005         if (test_opt(sb, MINIX_DF))
1006                 seq_puts(seq, ",minixdf");
1007         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1008                 seq_puts(seq, ",grpid");
1009         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1010                 seq_puts(seq, ",nogrpid");
1011         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1012             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1013                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1014         }
1015         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1016             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1017                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1018         }
1019         if (test_opt(sb, ERRORS_RO)) {
1020                 if (def_errors == EXT4_ERRORS_PANIC ||
1021                     def_errors == EXT4_ERRORS_CONTINUE) {
1022                         seq_puts(seq, ",errors=remount-ro");
1023                 }
1024         }
1025         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1026                 seq_puts(seq, ",errors=continue");
1027         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1028                 seq_puts(seq, ",errors=panic");
1029         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1030                 seq_puts(seq, ",nouid32");
1031         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1032                 seq_puts(seq, ",debug");
1033         if (test_opt(sb, OLDALLOC))
1034                 seq_puts(seq, ",oldalloc");
1035 #ifdef CONFIG_EXT4_FS_XATTR
1036         if (test_opt(sb, XATTR_USER))
1037                 seq_puts(seq, ",user_xattr");
1038         if (!test_opt(sb, XATTR_USER))
1039                 seq_puts(seq, ",nouser_xattr");
1040 #endif
1041 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1042         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1043                 seq_puts(seq, ",acl");
1044         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1045                 seq_puts(seq, ",noacl");
1046 #endif
1047         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1048                 seq_printf(seq, ",commit=%u",
1049                            (unsigned) (sbi->s_commit_interval / HZ));
1050         }
1051         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1052                 seq_printf(seq, ",min_batch_time=%u",
1053                            (unsigned) sbi->s_min_batch_time);
1054         }
1055         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1056                 seq_printf(seq, ",max_batch_time=%u",
1057                            (unsigned) sbi->s_min_batch_time);
1058         }
1059
1060         /*
1061          * We're changing the default of barrier mount option, so
1062          * let's always display its mount state so it's clear what its
1063          * status is.
1064          */
1065         seq_puts(seq, ",barrier=");
1066         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1067         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1068                 seq_puts(seq, ",journal_async_commit");
1069         else if (test_opt(sb, JOURNAL_CHECKSUM))
1070                 seq_puts(seq, ",journal_checksum");
1071         if (test_opt(sb, I_VERSION))
1072                 seq_puts(seq, ",i_version");
1073         if (!test_opt(sb, DELALLOC) &&
1074             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1075                 seq_puts(seq, ",nodelalloc");
1076
1077         if (!test_opt(sb, MBLK_IO_SUBMIT))
1078                 seq_puts(seq, ",nomblk_io_submit");
1079         if (sbi->s_stripe)
1080                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1081         /*
1082          * journal mode get enabled in different ways
1083          * So just print the value even if we didn't specify it
1084          */
1085         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1086                 seq_puts(seq, ",data=journal");
1087         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1088                 seq_puts(seq, ",data=ordered");
1089         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1090                 seq_puts(seq, ",data=writeback");
1091
1092         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1093                 seq_printf(seq, ",inode_readahead_blks=%u",
1094                            sbi->s_inode_readahead_blks);
1095
1096         if (test_opt(sb, DATA_ERR_ABORT))
1097                 seq_puts(seq, ",data_err=abort");
1098
1099         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1100                 seq_puts(seq, ",noauto_da_alloc");
1101
1102         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1103                 seq_puts(seq, ",discard");
1104
1105         if (test_opt(sb, NOLOAD))
1106                 seq_puts(seq, ",norecovery");
1107
1108         if (test_opt(sb, DIOREAD_NOLOCK))
1109                 seq_puts(seq, ",dioread_nolock");
1110
1111         if (test_opt(sb, BLOCK_VALIDITY) &&
1112             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1113                 seq_puts(seq, ",block_validity");
1114
1115         if (!test_opt(sb, INIT_INODE_TABLE))
1116                 seq_puts(seq, ",noinit_inode_table");
1117         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1118                 seq_printf(seq, ",init_inode_table=%u",
1119                            (unsigned) sbi->s_li_wait_mult);
1120
1121         ext4_show_quota_options(seq, sb);
1122
1123         return 0;
1124 }
1125
1126 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1127                                         u64 ino, u32 generation)
1128 {
1129         struct inode *inode;
1130
1131         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1132                 return ERR_PTR(-ESTALE);
1133         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1134                 return ERR_PTR(-ESTALE);
1135
1136         /* iget isn't really right if the inode is currently unallocated!!
1137          *
1138          * ext4_read_inode will return a bad_inode if the inode had been
1139          * deleted, so we should be safe.
1140          *
1141          * Currently we don't know the generation for parent directory, so
1142          * a generation of 0 means "accept any"
1143          */
1144         inode = ext4_iget(sb, ino);
1145         if (IS_ERR(inode))
1146                 return ERR_CAST(inode);
1147         if (generation && inode->i_generation != generation) {
1148                 iput(inode);
1149                 return ERR_PTR(-ESTALE);
1150         }
1151
1152         return inode;
1153 }
1154
1155 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1156                                         int fh_len, int fh_type)
1157 {
1158         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1159                                     ext4_nfs_get_inode);
1160 }
1161
1162 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1163                                         int fh_len, int fh_type)
1164 {
1165         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1166                                     ext4_nfs_get_inode);
1167 }
1168
1169 /*
1170  * Try to release metadata pages (indirect blocks, directories) which are
1171  * mapped via the block device.  Since these pages could have journal heads
1172  * which would prevent try_to_free_buffers() from freeing them, we must use
1173  * jbd2 layer's try_to_free_buffers() function to release them.
1174  */
1175 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1176                                  gfp_t wait)
1177 {
1178         journal_t *journal = EXT4_SB(sb)->s_journal;
1179
1180         WARN_ON(PageChecked(page));
1181         if (!page_has_buffers(page))
1182                 return 0;
1183         if (journal)
1184                 return jbd2_journal_try_to_free_buffers(journal, page,
1185                                                         wait & ~__GFP_WAIT);
1186         return try_to_free_buffers(page);
1187 }
1188
1189 #ifdef CONFIG_QUOTA
1190 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1191 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1192
1193 static int ext4_write_dquot(struct dquot *dquot);
1194 static int ext4_acquire_dquot(struct dquot *dquot);
1195 static int ext4_release_dquot(struct dquot *dquot);
1196 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1197 static int ext4_write_info(struct super_block *sb, int type);
1198 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1199                          struct path *path);
1200 static int ext4_quota_off(struct super_block *sb, int type);
1201 static int ext4_quota_on_mount(struct super_block *sb, int type);
1202 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1203                                size_t len, loff_t off);
1204 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1205                                 const char *data, size_t len, loff_t off);
1206
1207 static const struct dquot_operations ext4_quota_operations = {
1208         .get_reserved_space = ext4_get_reserved_space,
1209         .write_dquot    = ext4_write_dquot,
1210         .acquire_dquot  = ext4_acquire_dquot,
1211         .release_dquot  = ext4_release_dquot,
1212         .mark_dirty     = ext4_mark_dquot_dirty,
1213         .write_info     = ext4_write_info,
1214         .alloc_dquot    = dquot_alloc,
1215         .destroy_dquot  = dquot_destroy,
1216 };
1217
1218 static const struct quotactl_ops ext4_qctl_operations = {
1219         .quota_on       = ext4_quota_on,
1220         .quota_off      = ext4_quota_off,
1221         .quota_sync     = dquot_quota_sync,
1222         .get_info       = dquot_get_dqinfo,
1223         .set_info       = dquot_set_dqinfo,
1224         .get_dqblk      = dquot_get_dqblk,
1225         .set_dqblk      = dquot_set_dqblk
1226 };
1227 #endif
1228
1229 static const struct super_operations ext4_sops = {
1230         .alloc_inode    = ext4_alloc_inode,
1231         .destroy_inode  = ext4_destroy_inode,
1232         .write_inode    = ext4_write_inode,
1233         .dirty_inode    = ext4_dirty_inode,
1234         .drop_inode     = ext4_drop_inode,
1235         .evict_inode    = ext4_evict_inode,
1236         .put_super      = ext4_put_super,
1237         .sync_fs        = ext4_sync_fs,
1238         .freeze_fs      = ext4_freeze,
1239         .unfreeze_fs    = ext4_unfreeze,
1240         .statfs         = ext4_statfs,
1241         .remount_fs     = ext4_remount,
1242         .show_options   = ext4_show_options,
1243 #ifdef CONFIG_QUOTA
1244         .quota_read     = ext4_quota_read,
1245         .quota_write    = ext4_quota_write,
1246 #endif
1247         .bdev_try_to_free_page = bdev_try_to_free_page,
1248 };
1249
1250 static const struct super_operations ext4_nojournal_sops = {
1251         .alloc_inode    = ext4_alloc_inode,
1252         .destroy_inode  = ext4_destroy_inode,
1253         .write_inode    = ext4_write_inode,
1254         .dirty_inode    = ext4_dirty_inode,
1255         .drop_inode     = ext4_drop_inode,
1256         .evict_inode    = ext4_evict_inode,
1257         .write_super    = ext4_write_super,
1258         .put_super      = ext4_put_super,
1259         .statfs         = ext4_statfs,
1260         .remount_fs     = ext4_remount,
1261         .show_options   = ext4_show_options,
1262 #ifdef CONFIG_QUOTA
1263         .quota_read     = ext4_quota_read,
1264         .quota_write    = ext4_quota_write,
1265 #endif
1266         .bdev_try_to_free_page = bdev_try_to_free_page,
1267 };
1268
1269 static const struct export_operations ext4_export_ops = {
1270         .fh_to_dentry = ext4_fh_to_dentry,
1271         .fh_to_parent = ext4_fh_to_parent,
1272         .get_parent = ext4_get_parent,
1273 };
1274
1275 enum {
1276         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1277         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1278         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1279         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1280         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1281         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1282         Opt_journal_update, Opt_journal_dev,
1283         Opt_journal_checksum, Opt_journal_async_commit,
1284         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1285         Opt_data_err_abort, Opt_data_err_ignore,
1286         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1287         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1288         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1289         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1290         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1291         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1292         Opt_inode_readahead_blks, Opt_journal_ioprio,
1293         Opt_dioread_nolock, Opt_dioread_lock,
1294         Opt_discard, Opt_nodiscard,
1295         Opt_init_inode_table, Opt_noinit_inode_table,
1296 };
1297
1298 static const match_table_t tokens = {
1299         {Opt_bsd_df, "bsddf"},
1300         {Opt_minix_df, "minixdf"},
1301         {Opt_grpid, "grpid"},
1302         {Opt_grpid, "bsdgroups"},
1303         {Opt_nogrpid, "nogrpid"},
1304         {Opt_nogrpid, "sysvgroups"},
1305         {Opt_resgid, "resgid=%u"},
1306         {Opt_resuid, "resuid=%u"},
1307         {Opt_sb, "sb=%u"},
1308         {Opt_err_cont, "errors=continue"},
1309         {Opt_err_panic, "errors=panic"},
1310         {Opt_err_ro, "errors=remount-ro"},
1311         {Opt_nouid32, "nouid32"},
1312         {Opt_debug, "debug"},
1313         {Opt_oldalloc, "oldalloc"},
1314         {Opt_orlov, "orlov"},
1315         {Opt_user_xattr, "user_xattr"},
1316         {Opt_nouser_xattr, "nouser_xattr"},
1317         {Opt_acl, "acl"},
1318         {Opt_noacl, "noacl"},
1319         {Opt_noload, "noload"},
1320         {Opt_noload, "norecovery"},
1321         {Opt_nobh, "nobh"},
1322         {Opt_bh, "bh"},
1323         {Opt_commit, "commit=%u"},
1324         {Opt_min_batch_time, "min_batch_time=%u"},
1325         {Opt_max_batch_time, "max_batch_time=%u"},
1326         {Opt_journal_update, "journal=update"},
1327         {Opt_journal_dev, "journal_dev=%u"},
1328         {Opt_journal_checksum, "journal_checksum"},
1329         {Opt_journal_async_commit, "journal_async_commit"},
1330         {Opt_abort, "abort"},
1331         {Opt_data_journal, "data=journal"},
1332         {Opt_data_ordered, "data=ordered"},
1333         {Opt_data_writeback, "data=writeback"},
1334         {Opt_data_err_abort, "data_err=abort"},
1335         {Opt_data_err_ignore, "data_err=ignore"},
1336         {Opt_offusrjquota, "usrjquota="},
1337         {Opt_usrjquota, "usrjquota=%s"},
1338         {Opt_offgrpjquota, "grpjquota="},
1339         {Opt_grpjquota, "grpjquota=%s"},
1340         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1341         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1342         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1343         {Opt_grpquota, "grpquota"},
1344         {Opt_noquota, "noquota"},
1345         {Opt_quota, "quota"},
1346         {Opt_usrquota, "usrquota"},
1347         {Opt_barrier, "barrier=%u"},
1348         {Opt_barrier, "barrier"},
1349         {Opt_nobarrier, "nobarrier"},
1350         {Opt_i_version, "i_version"},
1351         {Opt_stripe, "stripe=%u"},
1352         {Opt_resize, "resize"},
1353         {Opt_delalloc, "delalloc"},
1354         {Opt_nodelalloc, "nodelalloc"},
1355         {Opt_mblk_io_submit, "mblk_io_submit"},
1356         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1357         {Opt_block_validity, "block_validity"},
1358         {Opt_noblock_validity, "noblock_validity"},
1359         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1360         {Opt_journal_ioprio, "journal_ioprio=%u"},
1361         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1362         {Opt_auto_da_alloc, "auto_da_alloc"},
1363         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1364         {Opt_dioread_nolock, "dioread_nolock"},
1365         {Opt_dioread_lock, "dioread_lock"},
1366         {Opt_discard, "discard"},
1367         {Opt_nodiscard, "nodiscard"},
1368         {Opt_init_inode_table, "init_itable=%u"},
1369         {Opt_init_inode_table, "init_itable"},
1370         {Opt_noinit_inode_table, "noinit_itable"},
1371         {Opt_err, NULL},
1372 };
1373
1374 static ext4_fsblk_t get_sb_block(void **data)
1375 {
1376         ext4_fsblk_t    sb_block;
1377         char            *options = (char *) *data;
1378
1379         if (!options || strncmp(options, "sb=", 3) != 0)
1380                 return 1;       /* Default location */
1381
1382         options += 3;
1383         /* TODO: use simple_strtoll with >32bit ext4 */
1384         sb_block = simple_strtoul(options, &options, 0);
1385         if (*options && *options != ',') {
1386                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1387                        (char *) *data);
1388                 return 1;
1389         }
1390         if (*options == ',')
1391                 options++;
1392         *data = (void *) options;
1393
1394         return sb_block;
1395 }
1396
1397 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1398 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1399         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1400
1401 #ifdef CONFIG_QUOTA
1402 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1403 {
1404         struct ext4_sb_info *sbi = EXT4_SB(sb);
1405         char *qname;
1406
1407         if (sb_any_quota_loaded(sb) &&
1408                 !sbi->s_qf_names[qtype]) {
1409                 ext4_msg(sb, KERN_ERR,
1410                         "Cannot change journaled "
1411                         "quota options when quota turned on");
1412                 return 0;
1413         }
1414         qname = match_strdup(args);
1415         if (!qname) {
1416                 ext4_msg(sb, KERN_ERR,
1417                         "Not enough memory for storing quotafile name");
1418                 return 0;
1419         }
1420         if (sbi->s_qf_names[qtype] &&
1421                 strcmp(sbi->s_qf_names[qtype], qname)) {
1422                 ext4_msg(sb, KERN_ERR,
1423                         "%s quota file already specified", QTYPE2NAME(qtype));
1424                 kfree(qname);
1425                 return 0;
1426         }
1427         sbi->s_qf_names[qtype] = qname;
1428         if (strchr(sbi->s_qf_names[qtype], '/')) {
1429                 ext4_msg(sb, KERN_ERR,
1430                         "quotafile must be on filesystem root");
1431                 kfree(sbi->s_qf_names[qtype]);
1432                 sbi->s_qf_names[qtype] = NULL;
1433                 return 0;
1434         }
1435         set_opt(sb, QUOTA);
1436         return 1;
1437 }
1438
1439 static int clear_qf_name(struct super_block *sb, int qtype)
1440 {
1441
1442         struct ext4_sb_info *sbi = EXT4_SB(sb);
1443
1444         if (sb_any_quota_loaded(sb) &&
1445                 sbi->s_qf_names[qtype]) {
1446                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1447                         " when quota turned on");
1448                 return 0;
1449         }
1450         /*
1451          * The space will be released later when all options are confirmed
1452          * to be correct
1453          */
1454         sbi->s_qf_names[qtype] = NULL;
1455         return 1;
1456 }
1457 #endif
1458
1459 static int parse_options(char *options, struct super_block *sb,
1460                          unsigned long *journal_devnum,
1461                          unsigned int *journal_ioprio,
1462                          ext4_fsblk_t *n_blocks_count, int is_remount)
1463 {
1464         struct ext4_sb_info *sbi = EXT4_SB(sb);
1465         char *p;
1466         substring_t args[MAX_OPT_ARGS];
1467         int data_opt = 0;
1468         int option;
1469 #ifdef CONFIG_QUOTA
1470         int qfmt;
1471 #endif
1472
1473         if (!options)
1474                 return 1;
1475
1476         while ((p = strsep(&options, ",")) != NULL) {
1477                 int token;
1478                 if (!*p)
1479                         continue;
1480
1481                 /*
1482                  * Initialize args struct so we know whether arg was
1483                  * found; some options take optional arguments.
1484                  */
1485                 args[0].to = args[0].from = NULL;
1486                 token = match_token(p, tokens, args);
1487                 switch (token) {
1488                 case Opt_bsd_df:
1489                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1490                         clear_opt(sb, MINIX_DF);
1491                         break;
1492                 case Opt_minix_df:
1493                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1494                         set_opt(sb, MINIX_DF);
1495
1496                         break;
1497                 case Opt_grpid:
1498                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1499                         set_opt(sb, GRPID);
1500
1501                         break;
1502                 case Opt_nogrpid:
1503                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1504                         clear_opt(sb, GRPID);
1505
1506                         break;
1507                 case Opt_resuid:
1508                         if (match_int(&args[0], &option))
1509                                 return 0;
1510                         sbi->s_resuid = option;
1511                         break;
1512                 case Opt_resgid:
1513                         if (match_int(&args[0], &option))
1514                                 return 0;
1515                         sbi->s_resgid = option;
1516                         break;
1517                 case Opt_sb:
1518                         /* handled by get_sb_block() instead of here */
1519                         /* *sb_block = match_int(&args[0]); */
1520                         break;
1521                 case Opt_err_panic:
1522                         clear_opt(sb, ERRORS_CONT);
1523                         clear_opt(sb, ERRORS_RO);
1524                         set_opt(sb, ERRORS_PANIC);
1525                         break;
1526                 case Opt_err_ro:
1527                         clear_opt(sb, ERRORS_CONT);
1528                         clear_opt(sb, ERRORS_PANIC);
1529                         set_opt(sb, ERRORS_RO);
1530                         break;
1531                 case Opt_err_cont:
1532                         clear_opt(sb, ERRORS_RO);
1533                         clear_opt(sb, ERRORS_PANIC);
1534                         set_opt(sb, ERRORS_CONT);
1535                         break;
1536                 case Opt_nouid32:
1537                         set_opt(sb, NO_UID32);
1538                         break;
1539                 case Opt_debug:
1540                         set_opt(sb, DEBUG);
1541                         break;
1542                 case Opt_oldalloc:
1543                         set_opt(sb, OLDALLOC);
1544                         break;
1545                 case Opt_orlov:
1546                         clear_opt(sb, OLDALLOC);
1547                         break;
1548 #ifdef CONFIG_EXT4_FS_XATTR
1549                 case Opt_user_xattr:
1550                         set_opt(sb, XATTR_USER);
1551                         break;
1552                 case Opt_nouser_xattr:
1553                         clear_opt(sb, XATTR_USER);
1554                         break;
1555 #else
1556                 case Opt_user_xattr:
1557                 case Opt_nouser_xattr:
1558                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1559                         break;
1560 #endif
1561 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1562                 case Opt_acl:
1563                         set_opt(sb, POSIX_ACL);
1564                         break;
1565                 case Opt_noacl:
1566                         clear_opt(sb, POSIX_ACL);
1567                         break;
1568 #else
1569                 case Opt_acl:
1570                 case Opt_noacl:
1571                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1572                         break;
1573 #endif
1574                 case Opt_journal_update:
1575                         /* @@@ FIXME */
1576                         /* Eventually we will want to be able to create
1577                            a journal file here.  For now, only allow the
1578                            user to specify an existing inode to be the
1579                            journal file. */
1580                         if (is_remount) {
1581                                 ext4_msg(sb, KERN_ERR,
1582                                          "Cannot specify journal on remount");
1583                                 return 0;
1584                         }
1585                         set_opt(sb, UPDATE_JOURNAL);
1586                         break;
1587                 case Opt_journal_dev:
1588                         if (is_remount) {
1589                                 ext4_msg(sb, KERN_ERR,
1590                                         "Cannot specify journal on remount");
1591                                 return 0;
1592                         }
1593                         if (match_int(&args[0], &option))
1594                                 return 0;
1595                         *journal_devnum = option;
1596                         break;
1597                 case Opt_journal_checksum:
1598                         set_opt(sb, JOURNAL_CHECKSUM);
1599                         break;
1600                 case Opt_journal_async_commit:
1601                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1602                         set_opt(sb, JOURNAL_CHECKSUM);
1603                         break;
1604                 case Opt_noload:
1605                         set_opt(sb, NOLOAD);
1606                         break;
1607                 case Opt_commit:
1608                         if (match_int(&args[0], &option))
1609                                 return 0;
1610                         if (option < 0)
1611                                 return 0;
1612                         if (option == 0)
1613                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1614                         sbi->s_commit_interval = HZ * option;
1615                         break;
1616                 case Opt_max_batch_time:
1617                         if (match_int(&args[0], &option))
1618                                 return 0;
1619                         if (option < 0)
1620                                 return 0;
1621                         if (option == 0)
1622                                 option = EXT4_DEF_MAX_BATCH_TIME;
1623                         sbi->s_max_batch_time = option;
1624                         break;
1625                 case Opt_min_batch_time:
1626                         if (match_int(&args[0], &option))
1627                                 return 0;
1628                         if (option < 0)
1629                                 return 0;
1630                         sbi->s_min_batch_time = option;
1631                         break;
1632                 case Opt_data_journal:
1633                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1634                         goto datacheck;
1635                 case Opt_data_ordered:
1636                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1637                         goto datacheck;
1638                 case Opt_data_writeback:
1639                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1640                 datacheck:
1641                         if (is_remount) {
1642                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1643                                         ext4_msg(sb, KERN_ERR,
1644                                                 "Cannot change data mode on remount");
1645                                         return 0;
1646                                 }
1647                         } else {
1648                                 clear_opt(sb, DATA_FLAGS);
1649                                 sbi->s_mount_opt |= data_opt;
1650                         }
1651                         break;
1652                 case Opt_data_err_abort:
1653                         set_opt(sb, DATA_ERR_ABORT);
1654                         break;
1655                 case Opt_data_err_ignore:
1656                         clear_opt(sb, DATA_ERR_ABORT);
1657                         break;
1658 #ifdef CONFIG_QUOTA
1659                 case Opt_usrjquota:
1660                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1661                                 return 0;
1662                         break;
1663                 case Opt_grpjquota:
1664                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1665                                 return 0;
1666                         break;
1667                 case Opt_offusrjquota:
1668                         if (!clear_qf_name(sb, USRQUOTA))
1669                                 return 0;
1670                         break;
1671                 case Opt_offgrpjquota:
1672                         if (!clear_qf_name(sb, GRPQUOTA))
1673                                 return 0;
1674                         break;
1675
1676                 case Opt_jqfmt_vfsold:
1677                         qfmt = QFMT_VFS_OLD;
1678                         goto set_qf_format;
1679                 case Opt_jqfmt_vfsv0:
1680                         qfmt = QFMT_VFS_V0;
1681                         goto set_qf_format;
1682                 case Opt_jqfmt_vfsv1:
1683                         qfmt = QFMT_VFS_V1;
1684 set_qf_format:
1685                         if (sb_any_quota_loaded(sb) &&
1686                             sbi->s_jquota_fmt != qfmt) {
1687                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1688                                         "journaled quota options when "
1689                                         "quota turned on");
1690                                 return 0;
1691                         }
1692                         sbi->s_jquota_fmt = qfmt;
1693                         break;
1694                 case Opt_quota:
1695                 case Opt_usrquota:
1696                         set_opt(sb, QUOTA);
1697                         set_opt(sb, USRQUOTA);
1698                         break;
1699                 case Opt_grpquota:
1700                         set_opt(sb, QUOTA);
1701                         set_opt(sb, GRPQUOTA);
1702                         break;
1703                 case Opt_noquota:
1704                         if (sb_any_quota_loaded(sb)) {
1705                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1706                                         "options when quota turned on");
1707                                 return 0;
1708                         }
1709                         clear_opt(sb, QUOTA);
1710                         clear_opt(sb, USRQUOTA);
1711                         clear_opt(sb, GRPQUOTA);
1712                         break;
1713 #else
1714                 case Opt_quota:
1715                 case Opt_usrquota:
1716                 case Opt_grpquota:
1717                         ext4_msg(sb, KERN_ERR,
1718                                 "quota options not supported");
1719                         break;
1720                 case Opt_usrjquota:
1721                 case Opt_grpjquota:
1722                 case Opt_offusrjquota:
1723                 case Opt_offgrpjquota:
1724                 case Opt_jqfmt_vfsold:
1725                 case Opt_jqfmt_vfsv0:
1726                 case Opt_jqfmt_vfsv1:
1727                         ext4_msg(sb, KERN_ERR,
1728                                 "journaled quota options not supported");
1729                         break;
1730                 case Opt_noquota:
1731                         break;
1732 #endif
1733                 case Opt_abort:
1734                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1735                         break;
1736                 case Opt_nobarrier:
1737                         clear_opt(sb, BARRIER);
1738                         break;
1739                 case Opt_barrier:
1740                         if (args[0].from) {
1741                                 if (match_int(&args[0], &option))
1742                                         return 0;
1743                         } else
1744                                 option = 1;     /* No argument, default to 1 */
1745                         if (option)
1746                                 set_opt(sb, BARRIER);
1747                         else
1748                                 clear_opt(sb, BARRIER);
1749                         break;
1750                 case Opt_ignore:
1751                         break;
1752                 case Opt_resize:
1753                         if (!is_remount) {
1754                                 ext4_msg(sb, KERN_ERR,
1755                                         "resize option only available "
1756                                         "for remount");
1757                                 return 0;
1758                         }
1759                         if (match_int(&args[0], &option) != 0)
1760                                 return 0;
1761                         *n_blocks_count = option;
1762                         break;
1763                 case Opt_nobh:
1764                         ext4_msg(sb, KERN_WARNING,
1765                                  "Ignoring deprecated nobh option");
1766                         break;
1767                 case Opt_bh:
1768                         ext4_msg(sb, KERN_WARNING,
1769                                  "Ignoring deprecated bh option");
1770                         break;
1771                 case Opt_i_version:
1772                         set_opt(sb, I_VERSION);
1773                         sb->s_flags |= MS_I_VERSION;
1774                         break;
1775                 case Opt_nodelalloc:
1776                         clear_opt(sb, DELALLOC);
1777                         break;
1778                 case Opt_mblk_io_submit:
1779                         set_opt(sb, MBLK_IO_SUBMIT);
1780                         break;
1781                 case Opt_nomblk_io_submit:
1782                         clear_opt(sb, MBLK_IO_SUBMIT);
1783                         break;
1784                 case Opt_stripe:
1785                         if (match_int(&args[0], &option))
1786                                 return 0;
1787                         if (option < 0)
1788                                 return 0;
1789                         sbi->s_stripe = option;
1790                         break;
1791                 case Opt_delalloc:
1792                         set_opt(sb, DELALLOC);
1793                         break;
1794                 case Opt_block_validity:
1795                         set_opt(sb, BLOCK_VALIDITY);
1796                         break;
1797                 case Opt_noblock_validity:
1798                         clear_opt(sb, BLOCK_VALIDITY);
1799                         break;
1800                 case Opt_inode_readahead_blks:
1801                         if (match_int(&args[0], &option))
1802                                 return 0;
1803                         if (option < 0 || option > (1 << 30))
1804                                 return 0;
1805                         if (option && !is_power_of_2(option)) {
1806                                 ext4_msg(sb, KERN_ERR,
1807                                          "EXT4-fs: inode_readahead_blks"
1808                                          " must be a power of 2");
1809                                 return 0;
1810                         }
1811                         sbi->s_inode_readahead_blks = option;
1812                         break;
1813                 case Opt_journal_ioprio:
1814                         if (match_int(&args[0], &option))
1815                                 return 0;
1816                         if (option < 0 || option > 7)
1817                                 break;
1818                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1819                                                             option);
1820                         break;
1821                 case Opt_noauto_da_alloc:
1822                         set_opt(sb, NO_AUTO_DA_ALLOC);
1823                         break;
1824                 case Opt_auto_da_alloc:
1825                         if (args[0].from) {
1826                                 if (match_int(&args[0], &option))
1827                                         return 0;
1828                         } else
1829                                 option = 1;     /* No argument, default to 1 */
1830                         if (option)
1831                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1832                         else
1833                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1834                         break;
1835                 case Opt_discard:
1836                         set_opt(sb, DISCARD);
1837                         break;
1838                 case Opt_nodiscard:
1839                         clear_opt(sb, DISCARD);
1840                         break;
1841                 case Opt_dioread_nolock:
1842                         set_opt(sb, DIOREAD_NOLOCK);
1843                         break;
1844                 case Opt_dioread_lock:
1845                         clear_opt(sb, DIOREAD_NOLOCK);
1846                         break;
1847                 case Opt_init_inode_table:
1848                         set_opt(sb, INIT_INODE_TABLE);
1849                         if (args[0].from) {
1850                                 if (match_int(&args[0], &option))
1851                                         return 0;
1852                         } else
1853                                 option = EXT4_DEF_LI_WAIT_MULT;
1854                         if (option < 0)
1855                                 return 0;
1856                         sbi->s_li_wait_mult = option;
1857                         break;
1858                 case Opt_noinit_inode_table:
1859                         clear_opt(sb, INIT_INODE_TABLE);
1860                         break;
1861                 default:
1862                         ext4_msg(sb, KERN_ERR,
1863                                "Unrecognized mount option \"%s\" "
1864                                "or missing value", p);
1865                         return 0;
1866                 }
1867         }
1868 #ifdef CONFIG_QUOTA
1869         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1870                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1871                         clear_opt(sb, USRQUOTA);
1872
1873                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1874                         clear_opt(sb, GRPQUOTA);
1875
1876                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1877                         ext4_msg(sb, KERN_ERR, "old and new quota "
1878                                         "format mixing");
1879                         return 0;
1880                 }
1881
1882                 if (!sbi->s_jquota_fmt) {
1883                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1884                                         "not specified");
1885                         return 0;
1886                 }
1887         } else {
1888                 if (sbi->s_jquota_fmt) {
1889                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1890                                         "specified with no journaling "
1891                                         "enabled");
1892                         return 0;
1893                 }
1894         }
1895 #endif
1896         return 1;
1897 }
1898
1899 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1900                             int read_only)
1901 {
1902         struct ext4_sb_info *sbi = EXT4_SB(sb);
1903         int res = 0;
1904
1905         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1906                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1907                          "forcing read-only mode");
1908                 res = MS_RDONLY;
1909         }
1910         if (read_only)
1911                 return res;
1912         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1913                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1914                          "running e2fsck is recommended");
1915         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1916                 ext4_msg(sb, KERN_WARNING,
1917                          "warning: mounting fs with errors, "
1918                          "running e2fsck is recommended");
1919         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1920                  le16_to_cpu(es->s_mnt_count) >=
1921                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1922                 ext4_msg(sb, KERN_WARNING,
1923                          "warning: maximal mount count reached, "
1924                          "running e2fsck is recommended");
1925         else if (le32_to_cpu(es->s_checkinterval) &&
1926                 (le32_to_cpu(es->s_lastcheck) +
1927                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1928                 ext4_msg(sb, KERN_WARNING,
1929                          "warning: checktime reached, "
1930                          "running e2fsck is recommended");
1931         if (!sbi->s_journal)
1932                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1933         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1934                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1935         le16_add_cpu(&es->s_mnt_count, 1);
1936         es->s_mtime = cpu_to_le32(get_seconds());
1937         ext4_update_dynamic_rev(sb);
1938         if (sbi->s_journal)
1939                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1940
1941         ext4_commit_super(sb, 1);
1942         if (test_opt(sb, DEBUG))
1943                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1944                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1945                         sb->s_blocksize,
1946                         sbi->s_groups_count,
1947                         EXT4_BLOCKS_PER_GROUP(sb),
1948                         EXT4_INODES_PER_GROUP(sb),
1949                         sbi->s_mount_opt, sbi->s_mount_opt2);
1950
1951         return res;
1952 }
1953
1954 static int ext4_fill_flex_info(struct super_block *sb)
1955 {
1956         struct ext4_sb_info *sbi = EXT4_SB(sb);
1957         struct ext4_group_desc *gdp = NULL;
1958         ext4_group_t flex_group_count;
1959         ext4_group_t flex_group;
1960         int groups_per_flex = 0;
1961         size_t size;
1962         int i;
1963
1964         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1965         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1966
1967         if (groups_per_flex < 2) {
1968                 sbi->s_log_groups_per_flex = 0;
1969                 return 1;
1970         }
1971
1972         /* We allocate both existing and potentially added groups */
1973         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1974                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1975                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1976         size = flex_group_count * sizeof(struct flex_groups);
1977         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1978         if (sbi->s_flex_groups == NULL) {
1979                 sbi->s_flex_groups = vzalloc(size);
1980                 if (sbi->s_flex_groups == NULL) {
1981                         ext4_msg(sb, KERN_ERR,
1982                                  "not enough memory for %u flex groups",
1983                                  flex_group_count);
1984                         goto failed;
1985                 }
1986         }
1987
1988         for (i = 0; i < sbi->s_groups_count; i++) {
1989                 gdp = ext4_get_group_desc(sb, i, NULL);
1990
1991                 flex_group = ext4_flex_group(sbi, i);
1992                 atomic_add(ext4_free_inodes_count(sb, gdp),
1993                            &sbi->s_flex_groups[flex_group].free_inodes);
1994                 atomic_add(ext4_free_blks_count(sb, gdp),
1995                            &sbi->s_flex_groups[flex_group].free_blocks);
1996                 atomic_add(ext4_used_dirs_count(sb, gdp),
1997                            &sbi->s_flex_groups[flex_group].used_dirs);
1998         }
1999
2000         return 1;
2001 failed:
2002         return 0;
2003 }
2004
2005 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2006                             struct ext4_group_desc *gdp)
2007 {
2008         __u16 crc = 0;
2009
2010         if (sbi->s_es->s_feature_ro_compat &
2011             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2012                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2013                 __le32 le_group = cpu_to_le32(block_group);
2014
2015                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2016                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2017                 crc = crc16(crc, (__u8 *)gdp, offset);
2018                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2019                 /* for checksum of struct ext4_group_desc do the rest...*/
2020                 if ((sbi->s_es->s_feature_incompat &
2021                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2022                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2023                         crc = crc16(crc, (__u8 *)gdp + offset,
2024                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2025                                         offset);
2026         }
2027
2028         return cpu_to_le16(crc);
2029 }
2030
2031 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2032                                 struct ext4_group_desc *gdp)
2033 {
2034         if ((sbi->s_es->s_feature_ro_compat &
2035              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2036             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2037                 return 0;
2038
2039         return 1;
2040 }
2041
2042 /* Called at mount-time, super-block is locked */
2043 static int ext4_check_descriptors(struct super_block *sb,
2044                                   ext4_group_t *first_not_zeroed)
2045 {
2046         struct ext4_sb_info *sbi = EXT4_SB(sb);
2047         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2048         ext4_fsblk_t last_block;
2049         ext4_fsblk_t block_bitmap;
2050         ext4_fsblk_t inode_bitmap;
2051         ext4_fsblk_t inode_table;
2052         int flexbg_flag = 0;
2053         ext4_group_t i, grp = sbi->s_groups_count;
2054
2055         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2056                 flexbg_flag = 1;
2057
2058         ext4_debug("Checking group descriptors");
2059
2060         for (i = 0; i < sbi->s_groups_count; i++) {
2061                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2062
2063                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2064                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2065                 else
2066                         last_block = first_block +
2067                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2068
2069                 if ((grp == sbi->s_groups_count) &&
2070                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2071                         grp = i;
2072
2073                 block_bitmap = ext4_block_bitmap(sb, gdp);
2074                 if (block_bitmap < first_block || block_bitmap > last_block) {
2075                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2076                                "Block bitmap for group %u not in group "
2077                                "(block %llu)!", i, block_bitmap);
2078                         return 0;
2079                 }
2080                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2081                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2082                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2083                                "Inode bitmap for group %u not in group "
2084                                "(block %llu)!", i, inode_bitmap);
2085                         return 0;
2086                 }
2087                 inode_table = ext4_inode_table(sb, gdp);
2088                 if (inode_table < first_block ||
2089                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2090                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2091                                "Inode table for group %u not in group "
2092                                "(block %llu)!", i, inode_table);
2093                         return 0;
2094                 }
2095                 ext4_lock_group(sb, i);
2096                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2097                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2098                                  "Checksum for group %u failed (%u!=%u)",
2099                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2100                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2101                         if (!(sb->s_flags & MS_RDONLY)) {
2102                                 ext4_unlock_group(sb, i);
2103                                 return 0;
2104                         }
2105                 }
2106                 ext4_unlock_group(sb, i);
2107                 if (!flexbg_flag)
2108                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2109         }
2110         if (NULL != first_not_zeroed)
2111                 *first_not_zeroed = grp;
2112
2113         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2114         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2115         return 1;
2116 }
2117
2118 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2119  * the superblock) which were deleted from all directories, but held open by
2120  * a process at the time of a crash.  We walk the list and try to delete these
2121  * inodes at recovery time (only with a read-write filesystem).
2122  *
2123  * In order to keep the orphan inode chain consistent during traversal (in
2124  * case of crash during recovery), we link each inode into the superblock
2125  * orphan list_head and handle it the same way as an inode deletion during
2126  * normal operation (which journals the operations for us).
2127  *
2128  * We only do an iget() and an iput() on each inode, which is very safe if we
2129  * accidentally point at an in-use or already deleted inode.  The worst that
2130  * can happen in this case is that we get a "bit already cleared" message from
2131  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2132  * e2fsck was run on this filesystem, and it must have already done the orphan
2133  * inode cleanup for us, so we can safely abort without any further action.
2134  */
2135 static void ext4_orphan_cleanup(struct super_block *sb,
2136                                 struct ext4_super_block *es)
2137 {
2138         unsigned int s_flags = sb->s_flags;
2139         int nr_orphans = 0, nr_truncates = 0;
2140 #ifdef CONFIG_QUOTA
2141         int i;
2142 #endif
2143         if (!es->s_last_orphan) {
2144                 jbd_debug(4, "no orphan inodes to clean up\n");
2145                 return;
2146         }
2147
2148         if (bdev_read_only(sb->s_bdev)) {
2149                 ext4_msg(sb, KERN_ERR, "write access "
2150                         "unavailable, skipping orphan cleanup");
2151                 return;
2152         }
2153
2154         /* Check if feature set would not allow a r/w mount */
2155         if (!ext4_feature_set_ok(sb, 0)) {
2156                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2157                          "unknown ROCOMPAT features");
2158                 return;
2159         }
2160
2161         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2162                 if (es->s_last_orphan)
2163                         jbd_debug(1, "Errors on filesystem, "
2164                                   "clearing orphan list.\n");
2165                 es->s_last_orphan = 0;
2166                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2167                 return;
2168         }
2169
2170         if (s_flags & MS_RDONLY) {
2171                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2172                 sb->s_flags &= ~MS_RDONLY;
2173         }
2174 #ifdef CONFIG_QUOTA
2175         /* Needed for iput() to work correctly and not trash data */
2176         sb->s_flags |= MS_ACTIVE;
2177         /* Turn on quotas so that they are updated correctly */
2178         for (i = 0; i < MAXQUOTAS; i++) {
2179                 if (EXT4_SB(sb)->s_qf_names[i]) {
2180                         int ret = ext4_quota_on_mount(sb, i);
2181                         if (ret < 0)
2182                                 ext4_msg(sb, KERN_ERR,
2183                                         "Cannot turn on journaled "
2184                                         "quota: error %d", ret);
2185                 }
2186         }
2187 #endif
2188
2189         while (es->s_last_orphan) {
2190                 struct inode *inode;
2191
2192                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2193                 if (IS_ERR(inode)) {
2194                         es->s_last_orphan = 0;
2195                         break;
2196                 }
2197
2198                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2199                 dquot_initialize(inode);
2200                 if (inode->i_nlink) {
2201                         ext4_msg(sb, KERN_DEBUG,
2202                                 "%s: truncating inode %lu to %lld bytes",
2203                                 __func__, inode->i_ino, inode->i_size);
2204                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2205                                   inode->i_ino, inode->i_size);
2206                         ext4_truncate(inode);
2207                         nr_truncates++;
2208                 } else {
2209                         ext4_msg(sb, KERN_DEBUG,
2210                                 "%s: deleting unreferenced inode %lu",
2211                                 __func__, inode->i_ino);
2212                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2213                                   inode->i_ino);
2214                         nr_orphans++;
2215                 }
2216                 iput(inode);  /* The delete magic happens here! */
2217         }
2218
2219 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2220
2221         if (nr_orphans)
2222                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2223                        PLURAL(nr_orphans));
2224         if (nr_truncates)
2225                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2226                        PLURAL(nr_truncates));
2227 #ifdef CONFIG_QUOTA
2228         /* Turn quotas off */
2229         for (i = 0; i < MAXQUOTAS; i++) {
2230                 if (sb_dqopt(sb)->files[i])
2231                         dquot_quota_off(sb, i);
2232         }
2233 #endif
2234         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2235 }
2236
2237 /*
2238  * Maximal extent format file size.
2239  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2240  * extent format containers, within a sector_t, and within i_blocks
2241  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2242  * so that won't be a limiting factor.
2243  *
2244  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2245  */
2246 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2247 {
2248         loff_t res;
2249         loff_t upper_limit = MAX_LFS_FILESIZE;
2250
2251         /* small i_blocks in vfs inode? */
2252         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2253                 /*
2254                  * CONFIG_LBDAF is not enabled implies the inode
2255                  * i_block represent total blocks in 512 bytes
2256                  * 32 == size of vfs inode i_blocks * 8
2257                  */
2258                 upper_limit = (1LL << 32) - 1;
2259
2260                 /* total blocks in file system block size */
2261                 upper_limit >>= (blkbits - 9);
2262                 upper_limit <<= blkbits;
2263         }
2264
2265         /* 32-bit extent-start container, ee_block */
2266         res = 1LL << 32;
2267         res <<= blkbits;
2268         res -= 1;
2269
2270         /* Sanity check against vm- & vfs- imposed limits */
2271         if (res > upper_limit)
2272                 res = upper_limit;
2273
2274         return res;
2275 }
2276
2277 /*
2278  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2279  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2280  * We need to be 1 filesystem block less than the 2^48 sector limit.
2281  */
2282 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2283 {
2284         loff_t res = EXT4_NDIR_BLOCKS;
2285         int meta_blocks;
2286         loff_t upper_limit;
2287         /* This is calculated to be the largest file size for a dense, block
2288          * mapped file such that the file's total number of 512-byte sectors,
2289          * including data and all indirect blocks, does not exceed (2^48 - 1).
2290          *
2291          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2292          * number of 512-byte sectors of the file.
2293          */
2294
2295         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2296                 /*
2297                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2298                  * the inode i_block field represents total file blocks in
2299                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2300                  */
2301                 upper_limit = (1LL << 32) - 1;
2302
2303                 /* total blocks in file system block size */
2304                 upper_limit >>= (bits - 9);
2305
2306         } else {
2307                 /*
2308                  * We use 48 bit ext4_inode i_blocks
2309                  * With EXT4_HUGE_FILE_FL set the i_blocks
2310                  * represent total number of blocks in
2311                  * file system block size
2312                  */
2313                 upper_limit = (1LL << 48) - 1;
2314
2315         }
2316
2317         /* indirect blocks */
2318         meta_blocks = 1;
2319         /* double indirect blocks */
2320         meta_blocks += 1 + (1LL << (bits-2));
2321         /* tripple indirect blocks */
2322         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2323
2324         upper_limit -= meta_blocks;
2325         upper_limit <<= bits;
2326
2327         res += 1LL << (bits-2);
2328         res += 1LL << (2*(bits-2));
2329         res += 1LL << (3*(bits-2));
2330         res <<= bits;
2331         if (res > upper_limit)
2332                 res = upper_limit;
2333
2334         if (res > MAX_LFS_FILESIZE)
2335                 res = MAX_LFS_FILESIZE;
2336
2337         return res;
2338 }
2339
2340 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2341                                    ext4_fsblk_t logical_sb_block, int nr)
2342 {
2343         struct ext4_sb_info *sbi = EXT4_SB(sb);
2344         ext4_group_t bg, first_meta_bg;
2345         int has_super = 0;
2346
2347         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2348
2349         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2350             nr < first_meta_bg)
2351                 return logical_sb_block + nr + 1;
2352         bg = sbi->s_desc_per_block * nr;
2353         if (ext4_bg_has_super(sb, bg))
2354                 has_super = 1;
2355
2356         return (has_super + ext4_group_first_block_no(sb, bg));
2357 }
2358
2359 /**
2360  * ext4_get_stripe_size: Get the stripe size.
2361  * @sbi: In memory super block info
2362  *
2363  * If we have specified it via mount option, then
2364  * use the mount option value. If the value specified at mount time is
2365  * greater than the blocks per group use the super block value.
2366  * If the super block value is greater than blocks per group return 0.
2367  * Allocator needs it be less than blocks per group.
2368  *
2369  */
2370 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2371 {
2372         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2373         unsigned long stripe_width =
2374                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2375
2376         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2377                 return sbi->s_stripe;
2378
2379         if (stripe_width <= sbi->s_blocks_per_group)
2380                 return stripe_width;
2381
2382         if (stride <= sbi->s_blocks_per_group)
2383                 return stride;
2384
2385         return 0;
2386 }
2387
2388 /* sysfs supprt */
2389
2390 struct ext4_attr {
2391         struct attribute attr;
2392         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2393         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2394                          const char *, size_t);
2395         int offset;
2396 };
2397
2398 static int parse_strtoul(const char *buf,
2399                 unsigned long max, unsigned long *value)
2400 {
2401         char *endp;
2402
2403         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2404         endp = skip_spaces(endp);
2405         if (*endp || *value > max)
2406                 return -EINVAL;
2407
2408         return 0;
2409 }
2410
2411 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2412                                               struct ext4_sb_info *sbi,
2413                                               char *buf)
2414 {
2415         return snprintf(buf, PAGE_SIZE, "%llu\n",
2416                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2417 }
2418
2419 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2420                                          struct ext4_sb_info *sbi, char *buf)
2421 {
2422         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2423
2424         if (!sb->s_bdev->bd_part)
2425                 return snprintf(buf, PAGE_SIZE, "0\n");
2426         return snprintf(buf, PAGE_SIZE, "%lu\n",
2427                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2428                          sbi->s_sectors_written_start) >> 1);
2429 }
2430
2431 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2432                                           struct ext4_sb_info *sbi, char *buf)
2433 {
2434         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2435
2436         if (!sb->s_bdev->bd_part)
2437                 return snprintf(buf, PAGE_SIZE, "0\n");
2438         return snprintf(buf, PAGE_SIZE, "%llu\n",
2439                         (unsigned long long)(sbi->s_kbytes_written +
2440                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2441                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2442 }
2443
2444 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2445                                       struct ext4_sb_info *sbi, char *buf)
2446 {
2447         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2448 }
2449
2450 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2451                                         struct ext4_sb_info *sbi, char *buf)
2452 {
2453         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2454 }
2455
2456 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2457                                           struct ext4_sb_info *sbi,
2458                                           const char *buf, size_t count)
2459 {
2460         unsigned long t;
2461
2462         if (parse_strtoul(buf, 0x40000000, &t))
2463                 return -EINVAL;
2464
2465         if (t && !is_power_of_2(t))
2466                 return -EINVAL;
2467
2468         sbi->s_inode_readahead_blks = t;
2469         return count;
2470 }
2471
2472 static ssize_t sbi_ui_show(struct ext4_attr *a,
2473                            struct ext4_sb_info *sbi, char *buf)
2474 {
2475         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2476
2477         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2478 }
2479
2480 static ssize_t sbi_ui_store(struct ext4_attr *a,
2481                             struct ext4_sb_info *sbi,
2482                             const char *buf, size_t count)
2483 {
2484         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2485         unsigned long t;
2486
2487         if (parse_strtoul(buf, 0xffffffff, &t))
2488                 return -EINVAL;
2489         *ui = t;
2490         return count;
2491 }
2492
2493 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2494 static struct ext4_attr ext4_attr_##_name = {                   \
2495         .attr = {.name = __stringify(_name), .mode = _mode },   \
2496         .show   = _show,                                        \
2497         .store  = _store,                                       \
2498         .offset = offsetof(struct ext4_sb_info, _elname),       \
2499 }
2500 #define EXT4_ATTR(name, mode, show, store) \
2501 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2502
2503 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2504 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2505 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2506 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2507         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2508 #define ATTR_LIST(name) &ext4_attr_##name.attr
2509
2510 EXT4_RO_ATTR(delayed_allocation_blocks);
2511 EXT4_RO_ATTR(session_write_kbytes);
2512 EXT4_RO_ATTR(lifetime_write_kbytes);
2513 EXT4_RO_ATTR(extent_cache_hits);
2514 EXT4_RO_ATTR(extent_cache_misses);
2515 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2516                  inode_readahead_blks_store, s_inode_readahead_blks);
2517 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2518 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2519 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2520 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2521 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2522 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2523 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2524 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2525
2526 static struct attribute *ext4_attrs[] = {
2527         ATTR_LIST(delayed_allocation_blocks),
2528         ATTR_LIST(session_write_kbytes),
2529         ATTR_LIST(lifetime_write_kbytes),
2530         ATTR_LIST(extent_cache_hits),
2531         ATTR_LIST(extent_cache_misses),
2532         ATTR_LIST(inode_readahead_blks),
2533         ATTR_LIST(inode_goal),
2534         ATTR_LIST(mb_stats),
2535         ATTR_LIST(mb_max_to_scan),
2536         ATTR_LIST(mb_min_to_scan),
2537         ATTR_LIST(mb_order2_req),
2538         ATTR_LIST(mb_stream_req),
2539         ATTR_LIST(mb_group_prealloc),
2540         ATTR_LIST(max_writeback_mb_bump),
2541         NULL,
2542 };
2543
2544 /* Features this copy of ext4 supports */
2545 EXT4_INFO_ATTR(lazy_itable_init);
2546 EXT4_INFO_ATTR(batched_discard);
2547
2548 static struct attribute *ext4_feat_attrs[] = {
2549         ATTR_LIST(lazy_itable_init),
2550         ATTR_LIST(batched_discard),
2551         NULL,
2552 };
2553
2554 static ssize_t ext4_attr_show(struct kobject *kobj,
2555                               struct attribute *attr, char *buf)
2556 {
2557         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2558                                                 s_kobj);
2559         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2560
2561         return a->show ? a->show(a, sbi, buf) : 0;
2562 }
2563
2564 static ssize_t ext4_attr_store(struct kobject *kobj,
2565                                struct attribute *attr,
2566                                const char *buf, size_t len)
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->store ? a->store(a, sbi, buf, len) : 0;
2573 }
2574
2575 static void ext4_sb_release(struct kobject *kobj)
2576 {
2577         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2578                                                 s_kobj);
2579         complete(&sbi->s_kobj_unregister);
2580 }
2581
2582 static const struct sysfs_ops ext4_attr_ops = {
2583         .show   = ext4_attr_show,
2584         .store  = ext4_attr_store,
2585 };
2586
2587 static struct kobj_type ext4_ktype = {
2588         .default_attrs  = ext4_attrs,
2589         .sysfs_ops      = &ext4_attr_ops,
2590         .release        = ext4_sb_release,
2591 };
2592
2593 static void ext4_feat_release(struct kobject *kobj)
2594 {
2595         complete(&ext4_feat->f_kobj_unregister);
2596 }
2597
2598 static struct kobj_type ext4_feat_ktype = {
2599         .default_attrs  = ext4_feat_attrs,
2600         .sysfs_ops      = &ext4_attr_ops,
2601         .release        = ext4_feat_release,
2602 };
2603
2604 /*
2605  * Check whether this filesystem can be mounted based on
2606  * the features present and the RDONLY/RDWR mount requested.
2607  * Returns 1 if this filesystem can be mounted as requested,
2608  * 0 if it cannot be.
2609  */
2610 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2611 {
2612         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2613                 ext4_msg(sb, KERN_ERR,
2614                         "Couldn't mount because of "
2615                         "unsupported optional features (%x)",
2616                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2617                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2618                 return 0;
2619         }
2620
2621         if (readonly)
2622                 return 1;
2623
2624         /* Check that feature set is OK for a read-write mount */
2625         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2626                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2627                          "unsupported optional features (%x)",
2628                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2629                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2630                 return 0;
2631         }
2632         /*
2633          * Large file size enabled file system can only be mounted
2634          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2635          */
2636         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2637                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2638                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2639                                  "cannot be mounted RDWR without "
2640                                  "CONFIG_LBDAF");
2641                         return 0;
2642                 }
2643         }
2644         return 1;
2645 }
2646
2647 /*
2648  * This function is called once a day if we have errors logged
2649  * on the file system
2650  */
2651 static void print_daily_error_info(unsigned long arg)
2652 {
2653         struct super_block *sb = (struct super_block *) arg;
2654         struct ext4_sb_info *sbi;
2655         struct ext4_super_block *es;
2656
2657         sbi = EXT4_SB(sb);
2658         es = sbi->s_es;
2659
2660         if (es->s_error_count)
2661                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2662                          le32_to_cpu(es->s_error_count));
2663         if (es->s_first_error_time) {
2664                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2665                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2666                        (int) sizeof(es->s_first_error_func),
2667                        es->s_first_error_func,
2668                        le32_to_cpu(es->s_first_error_line));
2669                 if (es->s_first_error_ino)
2670                         printk(": inode %u",
2671                                le32_to_cpu(es->s_first_error_ino));
2672                 if (es->s_first_error_block)
2673                         printk(": block %llu", (unsigned long long)
2674                                le64_to_cpu(es->s_first_error_block));
2675                 printk("\n");
2676         }
2677         if (es->s_last_error_time) {
2678                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2679                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2680                        (int) sizeof(es->s_last_error_func),
2681                        es->s_last_error_func,
2682                        le32_to_cpu(es->s_last_error_line));
2683                 if (es->s_last_error_ino)
2684                         printk(": inode %u",
2685                                le32_to_cpu(es->s_last_error_ino));
2686                 if (es->s_last_error_block)
2687                         printk(": block %llu", (unsigned long long)
2688                                le64_to_cpu(es->s_last_error_block));
2689                 printk("\n");
2690         }
2691         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2692 }
2693
2694 /* Find next suitable group and run ext4_init_inode_table */
2695 static int ext4_run_li_request(struct ext4_li_request *elr)
2696 {
2697         struct ext4_group_desc *gdp = NULL;
2698         ext4_group_t group, ngroups;
2699         struct super_block *sb;
2700         unsigned long timeout = 0;
2701         int ret = 0;
2702
2703         sb = elr->lr_super;
2704         ngroups = EXT4_SB(sb)->s_groups_count;
2705
2706         for (group = elr->lr_next_group; group < ngroups; group++) {
2707                 gdp = ext4_get_group_desc(sb, group, NULL);
2708                 if (!gdp) {
2709                         ret = 1;
2710                         break;
2711                 }
2712
2713                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2714                         break;
2715         }
2716
2717         if (group == ngroups)
2718                 ret = 1;
2719
2720         if (!ret) {
2721                 timeout = jiffies;
2722                 ret = ext4_init_inode_table(sb, group,
2723                                             elr->lr_timeout ? 0 : 1);
2724                 if (elr->lr_timeout == 0) {
2725                         timeout = (jiffies - timeout) *
2726                                   elr->lr_sbi->s_li_wait_mult;
2727                         elr->lr_timeout = timeout;
2728                 }
2729                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2730                 elr->lr_next_group = group + 1;
2731         }
2732
2733         return ret;
2734 }
2735
2736 /*
2737  * Remove lr_request from the list_request and free the
2738  * request structure. Should be called with li_list_mtx held
2739  */
2740 static void ext4_remove_li_request(struct ext4_li_request *elr)
2741 {
2742         struct ext4_sb_info *sbi;
2743
2744         if (!elr)
2745                 return;
2746
2747         sbi = elr->lr_sbi;
2748
2749         list_del(&elr->lr_request);
2750         sbi->s_li_request = NULL;
2751         kfree(elr);
2752 }
2753
2754 static void ext4_unregister_li_request(struct super_block *sb)
2755 {
2756         mutex_lock(&ext4_li_mtx);
2757         if (!ext4_li_info) {
2758                 mutex_unlock(&ext4_li_mtx);
2759                 return;
2760         }
2761
2762         mutex_lock(&ext4_li_info->li_list_mtx);
2763         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2764         mutex_unlock(&ext4_li_info->li_list_mtx);
2765         mutex_unlock(&ext4_li_mtx);
2766 }
2767
2768 static struct task_struct *ext4_lazyinit_task;
2769
2770 /*
2771  * This is the function where ext4lazyinit thread lives. It walks
2772  * through the request list searching for next scheduled filesystem.
2773  * When such a fs is found, run the lazy initialization request
2774  * (ext4_rn_li_request) and keep track of the time spend in this
2775  * function. Based on that time we compute next schedule time of
2776  * the request. When walking through the list is complete, compute
2777  * next waking time and put itself into sleep.
2778  */
2779 static int ext4_lazyinit_thread(void *arg)
2780 {
2781         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2782         struct list_head *pos, *n;
2783         struct ext4_li_request *elr;
2784         unsigned long next_wakeup, cur;
2785
2786         BUG_ON(NULL == eli);
2787
2788 cont_thread:
2789         while (true) {
2790                 next_wakeup = MAX_JIFFY_OFFSET;
2791
2792                 mutex_lock(&eli->li_list_mtx);
2793                 if (list_empty(&eli->li_request_list)) {
2794                         mutex_unlock(&eli->li_list_mtx);
2795                         goto exit_thread;
2796                 }
2797
2798                 list_for_each_safe(pos, n, &eli->li_request_list) {
2799                         elr = list_entry(pos, struct ext4_li_request,
2800                                          lr_request);
2801
2802                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2803                                 if (ext4_run_li_request(elr) != 0) {
2804                                         /* error, remove the lazy_init job */
2805                                         ext4_remove_li_request(elr);
2806                                         continue;
2807                                 }
2808                         }
2809
2810                         if (time_before(elr->lr_next_sched, next_wakeup))
2811                                 next_wakeup = elr->lr_next_sched;
2812                 }
2813                 mutex_unlock(&eli->li_list_mtx);
2814
2815                 if (freezing(current))
2816                         refrigerator();
2817
2818                 cur = jiffies;
2819                 if ((time_after_eq(cur, next_wakeup)) ||
2820                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2821                         cond_resched();
2822                         continue;
2823                 }
2824
2825                 schedule_timeout_interruptible(next_wakeup - cur);
2826
2827                 if (kthread_should_stop()) {
2828                         ext4_clear_request_list();
2829                         goto exit_thread;
2830                 }
2831         }
2832
2833 exit_thread:
2834         /*
2835          * It looks like the request list is empty, but we need
2836          * to check it under the li_list_mtx lock, to prevent any
2837          * additions into it, and of course we should lock ext4_li_mtx
2838          * to atomically free the list and ext4_li_info, because at
2839          * this point another ext4 filesystem could be registering
2840          * new one.
2841          */
2842         mutex_lock(&ext4_li_mtx);
2843         mutex_lock(&eli->li_list_mtx);
2844         if (!list_empty(&eli->li_request_list)) {
2845                 mutex_unlock(&eli->li_list_mtx);
2846                 mutex_unlock(&ext4_li_mtx);
2847                 goto cont_thread;
2848         }
2849         mutex_unlock(&eli->li_list_mtx);
2850         kfree(ext4_li_info);
2851         ext4_li_info = NULL;
2852         mutex_unlock(&ext4_li_mtx);
2853
2854         return 0;
2855 }
2856
2857 static void ext4_clear_request_list(void)
2858 {
2859         struct list_head *pos, *n;
2860         struct ext4_li_request *elr;
2861
2862         mutex_lock(&ext4_li_info->li_list_mtx);
2863         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2864                 elr = list_entry(pos, struct ext4_li_request,
2865                                  lr_request);
2866                 ext4_remove_li_request(elr);
2867         }
2868         mutex_unlock(&ext4_li_info->li_list_mtx);
2869 }
2870
2871 static int ext4_run_lazyinit_thread(void)
2872 {
2873         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2874                                          ext4_li_info, "ext4lazyinit");
2875         if (IS_ERR(ext4_lazyinit_task)) {
2876                 int err = PTR_ERR(ext4_lazyinit_task);
2877                 ext4_clear_request_list();
2878                 kfree(ext4_li_info);
2879                 ext4_li_info = NULL;
2880                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2881                                  "initialization thread\n",
2882                                  err);
2883                 return err;
2884         }
2885         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2886         return 0;
2887 }
2888
2889 /*
2890  * Check whether it make sense to run itable init. thread or not.
2891  * If there is at least one uninitialized inode table, return
2892  * corresponding group number, else the loop goes through all
2893  * groups and return total number of groups.
2894  */
2895 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2896 {
2897         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2898         struct ext4_group_desc *gdp = NULL;
2899
2900         for (group = 0; group < ngroups; group++) {
2901                 gdp = ext4_get_group_desc(sb, group, NULL);
2902                 if (!gdp)
2903                         continue;
2904
2905                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2906                         break;
2907         }
2908
2909         return group;
2910 }
2911
2912 static int ext4_li_info_new(void)
2913 {
2914         struct ext4_lazy_init *eli = NULL;
2915
2916         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2917         if (!eli)
2918                 return -ENOMEM;
2919
2920         INIT_LIST_HEAD(&eli->li_request_list);
2921         mutex_init(&eli->li_list_mtx);
2922
2923         eli->li_state |= EXT4_LAZYINIT_QUIT;
2924
2925         ext4_li_info = eli;
2926
2927         return 0;
2928 }
2929
2930 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2931                                             ext4_group_t start)
2932 {
2933         struct ext4_sb_info *sbi = EXT4_SB(sb);
2934         struct ext4_li_request *elr;
2935         unsigned long rnd;
2936
2937         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2938         if (!elr)
2939                 return NULL;
2940
2941         elr->lr_super = sb;
2942         elr->lr_sbi = sbi;
2943         elr->lr_next_group = start;
2944
2945         /*
2946          * Randomize first schedule time of the request to
2947          * spread the inode table initialization requests
2948          * better.
2949          */
2950         get_random_bytes(&rnd, sizeof(rnd));
2951         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2952                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2953
2954         return elr;
2955 }
2956
2957 static int ext4_register_li_request(struct super_block *sb,
2958                                     ext4_group_t first_not_zeroed)
2959 {
2960         struct ext4_sb_info *sbi = EXT4_SB(sb);
2961         struct ext4_li_request *elr;
2962         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2963         int ret = 0;
2964
2965         if (sbi->s_li_request != NULL) {
2966                 /*
2967                  * Reset timeout so it can be computed again, because
2968                  * s_li_wait_mult might have changed.
2969                  */
2970                 sbi->s_li_request->lr_timeout = 0;
2971                 return 0;
2972         }
2973
2974         if (first_not_zeroed == ngroups ||
2975             (sb->s_flags & MS_RDONLY) ||
2976             !test_opt(sb, INIT_INODE_TABLE))
2977                 return 0;
2978
2979         elr = ext4_li_request_new(sb, first_not_zeroed);
2980         if (!elr)
2981                 return -ENOMEM;
2982
2983         mutex_lock(&ext4_li_mtx);
2984
2985         if (NULL == ext4_li_info) {
2986                 ret = ext4_li_info_new();
2987                 if (ret)
2988                         goto out;
2989         }
2990
2991         mutex_lock(&ext4_li_info->li_list_mtx);
2992         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2993         mutex_unlock(&ext4_li_info->li_list_mtx);
2994
2995         sbi->s_li_request = elr;
2996         /*
2997          * set elr to NULL here since it has been inserted to
2998          * the request_list and the removal and free of it is
2999          * handled by ext4_clear_request_list from now on.
3000          */
3001         elr = NULL;
3002
3003         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3004                 ret = ext4_run_lazyinit_thread();
3005                 if (ret)
3006                         goto out;
3007         }
3008 out:
3009         mutex_unlock(&ext4_li_mtx);
3010         if (ret)
3011                 kfree(elr);
3012         return ret;
3013 }
3014
3015 /*
3016  * We do not need to lock anything since this is called on
3017  * module unload.
3018  */
3019 static void ext4_destroy_lazyinit_thread(void)
3020 {
3021         /*
3022          * If thread exited earlier
3023          * there's nothing to be done.
3024          */
3025         if (!ext4_li_info || !ext4_lazyinit_task)
3026                 return;
3027
3028         kthread_stop(ext4_lazyinit_task);
3029 }
3030
3031 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3032                                 __releases(kernel_lock)
3033                                 __acquires(kernel_lock)
3034 {
3035         char *orig_data = kstrdup(data, GFP_KERNEL);
3036         struct buffer_head *bh;
3037         struct ext4_super_block *es = NULL;
3038         struct ext4_sb_info *sbi;
3039         ext4_fsblk_t block;
3040         ext4_fsblk_t sb_block = get_sb_block(&data);
3041         ext4_fsblk_t logical_sb_block;
3042         unsigned long offset = 0;
3043         unsigned long journal_devnum = 0;
3044         unsigned long def_mount_opts;
3045         struct inode *root;
3046         char *cp;
3047         const char *descr;
3048         int ret = -ENOMEM;
3049         int blocksize;
3050         unsigned int db_count;
3051         unsigned int i;
3052         int needs_recovery, has_huge_files;
3053         __u64 blocks_count;
3054         int err;
3055         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3056         ext4_group_t first_not_zeroed;
3057
3058         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3059         if (!sbi)
3060                 goto out_free_orig;
3061
3062         sbi->s_blockgroup_lock =
3063                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3064         if (!sbi->s_blockgroup_lock) {
3065                 kfree(sbi);
3066                 goto out_free_orig;
3067         }
3068         sb->s_fs_info = sbi;
3069         sbi->s_mount_opt = 0;
3070         sbi->s_resuid = EXT4_DEF_RESUID;
3071         sbi->s_resgid = EXT4_DEF_RESGID;
3072         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3073         sbi->s_sb_block = sb_block;
3074         if (sb->s_bdev->bd_part)
3075                 sbi->s_sectors_written_start =
3076                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3077
3078         /* Cleanup superblock name */
3079         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3080                 *cp = '!';
3081
3082         ret = -EINVAL;
3083         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3084         if (!blocksize) {
3085                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3086                 goto out_fail;
3087         }
3088
3089         /*
3090          * The ext4 superblock will not be buffer aligned for other than 1kB
3091          * block sizes.  We need to calculate the offset from buffer start.
3092          */
3093         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3094                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3095                 offset = do_div(logical_sb_block, blocksize);
3096         } else {
3097                 logical_sb_block = sb_block;
3098         }
3099
3100         if (!(bh = sb_bread(sb, logical_sb_block))) {
3101                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3102                 goto out_fail;
3103         }
3104         /*
3105          * Note: s_es must be initialized as soon as possible because
3106          *       some ext4 macro-instructions depend on its value
3107          */
3108         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3109         sbi->s_es = es;
3110         sb->s_magic = le16_to_cpu(es->s_magic);
3111         if (sb->s_magic != EXT4_SUPER_MAGIC)
3112                 goto cantfind_ext4;
3113         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3114
3115         /* Set defaults before we parse the mount options */
3116         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3117         set_opt(sb, INIT_INODE_TABLE);
3118         if (def_mount_opts & EXT4_DEFM_DEBUG)
3119                 set_opt(sb, DEBUG);
3120         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3121                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3122                         "2.6.38");
3123                 set_opt(sb, GRPID);
3124         }
3125         if (def_mount_opts & EXT4_DEFM_UID16)
3126                 set_opt(sb, NO_UID32);
3127         /* xattr user namespace & acls are now defaulted on */
3128 #ifdef CONFIG_EXT4_FS_XATTR
3129         set_opt(sb, XATTR_USER);
3130 #endif
3131 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3132         set_opt(sb, POSIX_ACL);
3133 #endif
3134         set_opt(sb, MBLK_IO_SUBMIT);
3135         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3136                 set_opt(sb, JOURNAL_DATA);
3137         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3138                 set_opt(sb, ORDERED_DATA);
3139         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3140                 set_opt(sb, WRITEBACK_DATA);
3141
3142         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3143                 set_opt(sb, ERRORS_PANIC);
3144         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3145                 set_opt(sb, ERRORS_CONT);
3146         else
3147                 set_opt(sb, ERRORS_RO);
3148         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3149                 set_opt(sb, BLOCK_VALIDITY);
3150         if (def_mount_opts & EXT4_DEFM_DISCARD)
3151                 set_opt(sb, DISCARD);
3152
3153         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3154         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3155         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3156         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3157         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3158
3159         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3160                 set_opt(sb, BARRIER);
3161
3162         /*
3163          * enable delayed allocation by default
3164          * Use -o nodelalloc to turn it off
3165          */
3166         if (!IS_EXT3_SB(sb) &&
3167             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3168                 set_opt(sb, DELALLOC);
3169
3170         /*
3171          * set default s_li_wait_mult for lazyinit, for the case there is
3172          * no mount option specified.
3173          */
3174         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3175
3176         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3177                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3178                 ext4_msg(sb, KERN_WARNING,
3179                          "failed to parse options in superblock: %s",
3180                          sbi->s_es->s_mount_opts);
3181         }
3182         if (!parse_options((char *) data, sb, &journal_devnum,
3183                            &journal_ioprio, NULL, 0))
3184                 goto failed_mount;
3185
3186         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3187                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3188
3189         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3190             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3191              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3192              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3193                 ext4_msg(sb, KERN_WARNING,
3194                        "feature flags set on rev 0 fs, "
3195                        "running e2fsck is recommended");
3196
3197         if (IS_EXT2_SB(sb)) {
3198                 if (ext2_feature_set_ok(sb))
3199                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3200                                  "using the ext4 subsystem");
3201                 else {
3202                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3203                                  "to feature incompatibilities");
3204                         goto failed_mount;
3205                 }
3206         }
3207
3208         if (IS_EXT3_SB(sb)) {
3209                 if (ext3_feature_set_ok(sb))
3210                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3211                                  "using the ext4 subsystem");
3212                 else {
3213                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3214                                  "to feature incompatibilities");
3215                         goto failed_mount;
3216                 }
3217         }
3218
3219         /*
3220          * Check feature flags regardless of the revision level, since we
3221          * previously didn't change the revision level when setting the flags,
3222          * so there is a chance incompat flags are set on a rev 0 filesystem.
3223          */
3224         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3225                 goto failed_mount;
3226
3227         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3228
3229         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3230             blocksize > EXT4_MAX_BLOCK_SIZE) {
3231                 ext4_msg(sb, KERN_ERR,
3232                        "Unsupported filesystem blocksize %d", blocksize);
3233                 goto failed_mount;
3234         }
3235
3236         if (sb->s_blocksize != blocksize) {
3237                 /* Validate the filesystem blocksize */
3238                 if (!sb_set_blocksize(sb, blocksize)) {
3239                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3240                                         blocksize);
3241                         goto failed_mount;
3242                 }
3243
3244                 brelse(bh);
3245                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3246                 offset = do_div(logical_sb_block, blocksize);
3247                 bh = sb_bread(sb, logical_sb_block);
3248                 if (!bh) {
3249                         ext4_msg(sb, KERN_ERR,
3250                                "Can't read superblock on 2nd try");
3251                         goto failed_mount;
3252                 }
3253                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3254                 sbi->s_es = es;
3255                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3256                         ext4_msg(sb, KERN_ERR,
3257                                "Magic mismatch, very weird!");
3258                         goto failed_mount;
3259                 }
3260         }
3261
3262         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3263                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3264         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3265                                                       has_huge_files);
3266         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3267
3268         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3269                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3270                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3271         } else {
3272                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3273                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3274                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3275                     (!is_power_of_2(sbi->s_inode_size)) ||
3276                     (sbi->s_inode_size > blocksize)) {
3277                         ext4_msg(sb, KERN_ERR,
3278                                "unsupported inode size: %d",
3279                                sbi->s_inode_size);
3280                         goto failed_mount;
3281                 }
3282                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3283                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3284         }
3285
3286         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3287         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3288                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3289                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3290                     !is_power_of_2(sbi->s_desc_size)) {
3291                         ext4_msg(sb, KERN_ERR,
3292                                "unsupported descriptor size %lu",
3293                                sbi->s_desc_size);
3294                         goto failed_mount;
3295                 }
3296         } else
3297                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3298
3299         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3300         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3301         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3302                 goto cantfind_ext4;
3303
3304         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3305         if (sbi->s_inodes_per_block == 0)
3306                 goto cantfind_ext4;
3307         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3308                                         sbi->s_inodes_per_block;
3309         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3310         sbi->s_sbh = bh;
3311         sbi->s_mount_state = le16_to_cpu(es->s_state);
3312         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3313         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3314
3315         for (i = 0; i < 4; i++)
3316                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3317         sbi->s_def_hash_version = es->s_def_hash_version;
3318         i = le32_to_cpu(es->s_flags);
3319         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3320                 sbi->s_hash_unsigned = 3;
3321         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3322 #ifdef __CHAR_UNSIGNED__
3323                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3324                 sbi->s_hash_unsigned = 3;
3325 #else
3326                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3327 #endif
3328                 sb->s_dirt = 1;
3329         }
3330
3331         if (sbi->s_blocks_per_group > blocksize * 8) {
3332                 ext4_msg(sb, KERN_ERR,
3333                        "#blocks per group too big: %lu",
3334                        sbi->s_blocks_per_group);
3335                 goto failed_mount;
3336         }
3337         if (sbi->s_inodes_per_group > blocksize * 8) {
3338                 ext4_msg(sb, KERN_ERR,
3339                        "#inodes per group too big: %lu",
3340                        sbi->s_inodes_per_group);
3341                 goto failed_mount;
3342         }
3343
3344         /*
3345          * Test whether we have more sectors than will fit in sector_t,
3346          * and whether the max offset is addressable by the page cache.
3347          */
3348         err = generic_check_addressable(sb->s_blocksize_bits,
3349                                         ext4_blocks_count(es));
3350         if (err) {
3351                 ext4_msg(sb, KERN_ERR, "filesystem"
3352                          " too large to mount safely on this system");
3353                 if (sizeof(sector_t) < 8)
3354                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3355                 ret = err;
3356                 goto failed_mount;
3357         }
3358
3359         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3360                 goto cantfind_ext4;
3361
3362         /* check blocks count against device size */
3363         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3364         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3365                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3366                        "exceeds size of device (%llu blocks)",
3367                        ext4_blocks_count(es), blocks_count);
3368                 goto failed_mount;
3369         }
3370
3371         /*
3372          * It makes no sense for the first data block to be beyond the end
3373          * of the filesystem.
3374          */
3375         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3376                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3377                          "block %u is beyond end of filesystem (%llu)",
3378                          le32_to_cpu(es->s_first_data_block),
3379                          ext4_blocks_count(es));
3380                 goto failed_mount;
3381         }
3382         blocks_count = (ext4_blocks_count(es) -
3383                         le32_to_cpu(es->s_first_data_block) +
3384                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3385         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3386         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3387                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3388                        "(block count %llu, first data block %u, "
3389                        "blocks per group %lu)", sbi->s_groups_count,
3390                        ext4_blocks_count(es),
3391                        le32_to_cpu(es->s_first_data_block),
3392                        EXT4_BLOCKS_PER_GROUP(sb));
3393                 goto failed_mount;
3394         }
3395         sbi->s_groups_count = blocks_count;
3396         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3397                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3398         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3399                    EXT4_DESC_PER_BLOCK(sb);
3400         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3401                                     GFP_KERNEL);
3402         if (sbi->s_group_desc == NULL) {
3403                 ext4_msg(sb, KERN_ERR, "not enough memory");
3404                 goto failed_mount;
3405         }
3406
3407 #ifdef CONFIG_PROC_FS
3408         if (ext4_proc_root)
3409                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3410 #endif
3411
3412         bgl_lock_init(sbi->s_blockgroup_lock);
3413
3414         for (i = 0; i < db_count; i++) {
3415                 block = descriptor_loc(sb, logical_sb_block, i);
3416                 sbi->s_group_desc[i] = sb_bread(sb, block);
3417                 if (!sbi->s_group_desc[i]) {
3418                         ext4_msg(sb, KERN_ERR,
3419                                "can't read group descriptor %d", i);
3420                         db_count = i;
3421                         goto failed_mount2;
3422                 }
3423         }
3424         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3425                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3426                 goto failed_mount2;
3427         }
3428         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3429                 if (!ext4_fill_flex_info(sb)) {
3430                         ext4_msg(sb, KERN_ERR,
3431                                "unable to initialize "
3432                                "flex_bg meta info!");
3433                         goto failed_mount2;
3434                 }
3435
3436         sbi->s_gdb_count = db_count;
3437         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3438         spin_lock_init(&sbi->s_next_gen_lock);
3439
3440         init_timer(&sbi->s_err_report);
3441         sbi->s_err_report.function = print_daily_error_info;
3442         sbi->s_err_report.data = (unsigned long) sb;
3443
3444         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3445                         ext4_count_free_blocks(sb));
3446         if (!err) {
3447                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3448                                 ext4_count_free_inodes(sb));
3449         }
3450         if (!err) {
3451                 err = percpu_counter_init(&sbi->s_dirs_counter,
3452                                 ext4_count_dirs(sb));
3453         }
3454         if (!err) {
3455                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3456         }
3457         if (err) {
3458                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3459                 goto failed_mount3;
3460         }
3461
3462         sbi->s_stripe = ext4_get_stripe_size(sbi);
3463         sbi->s_max_writeback_mb_bump = 128;
3464
3465         /*
3466          * set up enough so that it can read an inode
3467          */
3468         if (!test_opt(sb, NOLOAD) &&
3469             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3470                 sb->s_op = &ext4_sops;
3471         else
3472                 sb->s_op = &ext4_nojournal_sops;
3473         sb->s_export_op = &ext4_export_ops;
3474         sb->s_xattr = ext4_xattr_handlers;
3475 #ifdef CONFIG_QUOTA
3476         sb->s_qcop = &ext4_qctl_operations;
3477         sb->dq_op = &ext4_quota_operations;
3478 #endif
3479         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3480
3481         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3482         mutex_init(&sbi->s_orphan_lock);
3483         mutex_init(&sbi->s_resize_lock);
3484
3485         sb->s_root = NULL;
3486
3487         needs_recovery = (es->s_last_orphan != 0 ||
3488                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3489                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3490
3491         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3492             !(sb->s_flags & MS_RDONLY))
3493                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3494                         goto failed_mount3;
3495
3496         /*
3497          * The first inode we look at is the journal inode.  Don't try
3498          * root first: it may be modified in the journal!
3499          */
3500         if (!test_opt(sb, NOLOAD) &&
3501             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3502                 if (ext4_load_journal(sb, es, journal_devnum))
3503                         goto failed_mount3;
3504         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3505               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3506                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3507                        "suppressed and not mounted read-only");
3508                 goto failed_mount_wq;
3509         } else {
3510                 clear_opt(sb, DATA_FLAGS);
3511                 sbi->s_journal = NULL;
3512                 needs_recovery = 0;
3513                 goto no_journal;
3514         }
3515
3516         if (ext4_blocks_count(es) > 0xffffffffULL &&
3517             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3518                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3519                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3520                 goto failed_mount_wq;
3521         }
3522
3523         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3524                 jbd2_journal_set_features(sbi->s_journal,
3525                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3526                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3527         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3528                 jbd2_journal_set_features(sbi->s_journal,
3529                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3530                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3531                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3532         } else {
3533                 jbd2_journal_clear_features(sbi->s_journal,
3534                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3535                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3536         }
3537
3538         /* We have now updated the journal if required, so we can
3539          * validate the data journaling mode. */
3540         switch (test_opt(sb, DATA_FLAGS)) {
3541         case 0:
3542                 /* No mode set, assume a default based on the journal
3543                  * capabilities: ORDERED_DATA if the journal can
3544                  * cope, else JOURNAL_DATA
3545                  */
3546                 if (jbd2_journal_check_available_features
3547                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3548                         set_opt(sb, ORDERED_DATA);
3549                 else
3550                         set_opt(sb, JOURNAL_DATA);
3551                 break;
3552
3553         case EXT4_MOUNT_ORDERED_DATA:
3554         case EXT4_MOUNT_WRITEBACK_DATA:
3555                 if (!jbd2_journal_check_available_features
3556                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3557                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3558                                "requested data journaling mode");
3559                         goto failed_mount_wq;
3560                 }
3561         default:
3562                 break;
3563         }
3564         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3565
3566         /*
3567          * The journal may have updated the bg summary counts, so we
3568          * need to update the global counters.
3569          */
3570         percpu_counter_set(&sbi->s_freeblocks_counter,
3571                            ext4_count_free_blocks(sb));
3572         percpu_counter_set(&sbi->s_freeinodes_counter,
3573                            ext4_count_free_inodes(sb));
3574         percpu_counter_set(&sbi->s_dirs_counter,
3575                            ext4_count_dirs(sb));
3576         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3577
3578 no_journal:
3579         /*
3580          * The maximum number of concurrent works can be high and
3581          * concurrency isn't really necessary.  Limit it to 1.
3582          */
3583         EXT4_SB(sb)->dio_unwritten_wq =
3584                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3585         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3586                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3587                 goto failed_mount_wq;
3588         }
3589
3590         /*
3591          * The jbd2_journal_load will have done any necessary log recovery,
3592          * so we can safely mount the rest of the filesystem now.
3593          */
3594
3595         root = ext4_iget(sb, EXT4_ROOT_INO);
3596         if (IS_ERR(root)) {
3597                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3598                 ret = PTR_ERR(root);
3599                 root = NULL;
3600                 goto failed_mount4;
3601         }
3602         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3603                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3604                 goto failed_mount4;
3605         }
3606         sb->s_root = d_alloc_root(root);
3607         if (!sb->s_root) {
3608                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3609                 ret = -ENOMEM;
3610                 goto failed_mount4;
3611         }
3612
3613         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3614
3615         /* determine the minimum size of new large inodes, if present */
3616         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3617                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3618                                                      EXT4_GOOD_OLD_INODE_SIZE;
3619                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3620                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3621                         if (sbi->s_want_extra_isize <
3622                             le16_to_cpu(es->s_want_extra_isize))
3623                                 sbi->s_want_extra_isize =
3624                                         le16_to_cpu(es->s_want_extra_isize);
3625                         if (sbi->s_want_extra_isize <
3626                             le16_to_cpu(es->s_min_extra_isize))
3627                                 sbi->s_want_extra_isize =
3628                                         le16_to_cpu(es->s_min_extra_isize);
3629                 }
3630         }
3631         /* Check if enough inode space is available */
3632         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3633                                                         sbi->s_inode_size) {
3634                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3635                                                        EXT4_GOOD_OLD_INODE_SIZE;
3636                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3637                          "available");
3638         }
3639
3640         if (test_opt(sb, DELALLOC) &&
3641             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3642                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3643                          "requested data journaling mode");
3644                 clear_opt(sb, DELALLOC);
3645         }
3646         if (test_opt(sb, DIOREAD_NOLOCK)) {
3647                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3648                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3649                                 "option - requested data journaling mode");
3650                         clear_opt(sb, DIOREAD_NOLOCK);
3651                 }
3652                 if (sb->s_blocksize < PAGE_SIZE) {
3653                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3654                                 "option - block size is too small");
3655                         clear_opt(sb, DIOREAD_NOLOCK);
3656                 }
3657         }
3658
3659         err = ext4_setup_system_zone(sb);
3660         if (err) {
3661                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3662                          "zone (%d)", err);
3663                 goto failed_mount4;
3664         }
3665
3666         ext4_ext_init(sb);
3667         err = ext4_mb_init(sb, needs_recovery);
3668         if (err) {
3669                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3670                          err);
3671                 goto failed_mount4;
3672         }
3673
3674         err = ext4_register_li_request(sb, first_not_zeroed);
3675         if (err)
3676                 goto failed_mount4;
3677
3678         sbi->s_kobj.kset = ext4_kset;
3679         init_completion(&sbi->s_kobj_unregister);
3680         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3681                                    "%s", sb->s_id);
3682         if (err) {
3683                 ext4_mb_release(sb);
3684                 ext4_ext_release(sb);
3685                 goto failed_mount4;
3686         };
3687
3688         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3689         ext4_orphan_cleanup(sb, es);
3690         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3691         if (needs_recovery) {
3692                 ext4_msg(sb, KERN_INFO, "recovery complete");
3693                 ext4_mark_recovery_complete(sb, es);
3694         }
3695         if (EXT4_SB(sb)->s_journal) {
3696                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3697                         descr = " journalled data mode";
3698                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3699                         descr = " ordered data mode";
3700                 else
3701                         descr = " writeback data mode";
3702         } else
3703                 descr = "out journal";
3704
3705         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3706                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3707                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3708
3709         if (es->s_error_count)
3710                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3711
3712         kfree(orig_data);
3713         return 0;
3714
3715 cantfind_ext4:
3716         if (!silent)
3717                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3718         goto failed_mount;
3719
3720 failed_mount4:
3721         iput(root);
3722         sb->s_root = NULL;
3723         ext4_msg(sb, KERN_ERR, "mount failed");
3724         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3725 failed_mount_wq:
3726         ext4_release_system_zone(sb);
3727         if (sbi->s_journal) {
3728                 jbd2_journal_destroy(sbi->s_journal);
3729                 sbi->s_journal = NULL;
3730         }
3731 failed_mount3:
3732         del_timer(&sbi->s_err_report);
3733         if (sbi->s_flex_groups) {
3734                 if (is_vmalloc_addr(sbi->s_flex_groups))
3735                         vfree(sbi->s_flex_groups);
3736                 else
3737                         kfree(sbi->s_flex_groups);
3738         }
3739         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3740         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3741         percpu_counter_destroy(&sbi->s_dirs_counter);
3742         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3743         if (sbi->s_mmp_tsk)
3744                 kthread_stop(sbi->s_mmp_tsk);
3745 failed_mount2:
3746         for (i = 0; i < db_count; i++)
3747                 brelse(sbi->s_group_desc[i]);
3748         kfree(sbi->s_group_desc);
3749 failed_mount:
3750         if (sbi->s_proc) {
3751                 remove_proc_entry(sb->s_id, ext4_proc_root);
3752         }
3753 #ifdef CONFIG_QUOTA
3754         for (i = 0; i < MAXQUOTAS; i++)
3755                 kfree(sbi->s_qf_names[i]);
3756 #endif
3757         ext4_blkdev_remove(sbi);
3758         brelse(bh);
3759 out_fail:
3760         sb->s_fs_info = NULL;
3761         kfree(sbi->s_blockgroup_lock);
3762         kfree(sbi);
3763 out_free_orig:
3764         kfree(orig_data);
3765         return ret;
3766 }
3767
3768 /*
3769  * Setup any per-fs journal parameters now.  We'll do this both on
3770  * initial mount, once the journal has been initialised but before we've
3771  * done any recovery; and again on any subsequent remount.
3772  */
3773 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3774 {
3775         struct ext4_sb_info *sbi = EXT4_SB(sb);
3776
3777         journal->j_commit_interval = sbi->s_commit_interval;
3778         journal->j_min_batch_time = sbi->s_min_batch_time;
3779         journal->j_max_batch_time = sbi->s_max_batch_time;
3780
3781         write_lock(&journal->j_state_lock);
3782         if (test_opt(sb, BARRIER))
3783                 journal->j_flags |= JBD2_BARRIER;
3784         else
3785                 journal->j_flags &= ~JBD2_BARRIER;
3786         if (test_opt(sb, DATA_ERR_ABORT))
3787                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3788         else
3789                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3790         write_unlock(&journal->j_state_lock);
3791 }
3792
3793 static journal_t *ext4_get_journal(struct super_block *sb,
3794                                    unsigned int journal_inum)
3795 {
3796         struct inode *journal_inode;
3797         journal_t *journal;
3798
3799         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3800
3801         /* First, test for the existence of a valid inode on disk.  Bad
3802          * things happen if we iget() an unused inode, as the subsequent
3803          * iput() will try to delete it. */
3804
3805         journal_inode = ext4_iget(sb, journal_inum);
3806         if (IS_ERR(journal_inode)) {
3807                 ext4_msg(sb, KERN_ERR, "no journal found");
3808                 return NULL;
3809         }
3810         if (!journal_inode->i_nlink) {
3811                 make_bad_inode(journal_inode);
3812                 iput(journal_inode);
3813                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3814                 return NULL;
3815         }
3816
3817         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3818                   journal_inode, journal_inode->i_size);
3819         if (!S_ISREG(journal_inode->i_mode)) {
3820                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3821                 iput(journal_inode);
3822                 return NULL;
3823         }
3824
3825         journal = jbd2_journal_init_inode(journal_inode);
3826         if (!journal) {
3827                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3828                 iput(journal_inode);
3829                 return NULL;
3830         }
3831         journal->j_private = sb;
3832         ext4_init_journal_params(sb, journal);
3833         return journal;
3834 }
3835
3836 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3837                                        dev_t j_dev)
3838 {
3839         struct buffer_head *bh;
3840         journal_t *journal;
3841         ext4_fsblk_t start;
3842         ext4_fsblk_t len;
3843         int hblock, blocksize;
3844         ext4_fsblk_t sb_block;
3845         unsigned long offset;
3846         struct ext4_super_block *es;
3847         struct block_device *bdev;
3848
3849         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3850
3851         bdev = ext4_blkdev_get(j_dev, sb);
3852         if (bdev == NULL)
3853                 return NULL;
3854
3855         blocksize = sb->s_blocksize;
3856         hblock = bdev_logical_block_size(bdev);
3857         if (blocksize < hblock) {
3858                 ext4_msg(sb, KERN_ERR,
3859                         "blocksize too small for journal device");
3860                 goto out_bdev;
3861         }
3862
3863         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3864         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3865         set_blocksize(bdev, blocksize);
3866         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3867                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3868                        "external journal");
3869                 goto out_bdev;
3870         }
3871
3872         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3873         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3874             !(le32_to_cpu(es->s_feature_incompat) &
3875               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3876                 ext4_msg(sb, KERN_ERR, "external journal has "
3877                                         "bad superblock");
3878                 brelse(bh);
3879                 goto out_bdev;
3880         }
3881
3882         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3883                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3884                 brelse(bh);
3885                 goto out_bdev;
3886         }
3887
3888         len = ext4_blocks_count(es);
3889         start = sb_block + 1;
3890         brelse(bh);     /* we're done with the superblock */
3891
3892         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3893                                         start, len, blocksize);
3894         if (!journal) {
3895                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3896                 goto out_bdev;
3897         }
3898         journal->j_private = sb;
3899         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3900         wait_on_buffer(journal->j_sb_buffer);
3901         if (!buffer_uptodate(journal->j_sb_buffer)) {
3902                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3903                 goto out_journal;
3904         }
3905         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3906                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3907                                         "user (unsupported) - %d",
3908                         be32_to_cpu(journal->j_superblock->s_nr_users));
3909                 goto out_journal;
3910         }
3911         EXT4_SB(sb)->journal_bdev = bdev;
3912         ext4_init_journal_params(sb, journal);
3913         return journal;
3914
3915 out_journal:
3916         jbd2_journal_destroy(journal);
3917 out_bdev:
3918         ext4_blkdev_put(bdev);
3919         return NULL;
3920 }
3921
3922 static int ext4_load_journal(struct super_block *sb,
3923                              struct ext4_super_block *es,
3924                              unsigned long journal_devnum)
3925 {
3926         journal_t *journal;
3927         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3928         dev_t journal_dev;
3929         int err = 0;
3930         int really_read_only;
3931
3932         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3933
3934         if (journal_devnum &&
3935             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3936                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3937                         "numbers have changed");
3938                 journal_dev = new_decode_dev(journal_devnum);
3939         } else
3940                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3941
3942         really_read_only = bdev_read_only(sb->s_bdev);
3943
3944         /*
3945          * Are we loading a blank journal or performing recovery after a
3946          * crash?  For recovery, we need to check in advance whether we
3947          * can get read-write access to the device.
3948          */
3949         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3950                 if (sb->s_flags & MS_RDONLY) {
3951                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3952                                         "required on readonly filesystem");
3953                         if (really_read_only) {
3954                                 ext4_msg(sb, KERN_ERR, "write access "
3955                                         "unavailable, cannot proceed");
3956                                 return -EROFS;
3957                         }
3958                         ext4_msg(sb, KERN_INFO, "write access will "
3959                                "be enabled during recovery");
3960                 }
3961         }
3962
3963         if (journal_inum && journal_dev) {
3964                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3965                        "and inode journals!");
3966                 return -EINVAL;
3967         }
3968
3969         if (journal_inum) {
3970                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3971                         return -EINVAL;
3972         } else {
3973                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3974                         return -EINVAL;
3975         }
3976
3977         if (!(journal->j_flags & JBD2_BARRIER))
3978                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3979
3980         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3981                 err = jbd2_journal_update_format(journal);
3982                 if (err)  {
3983                         ext4_msg(sb, KERN_ERR, "error updating journal");
3984                         jbd2_journal_destroy(journal);
3985                         return err;
3986                 }
3987         }
3988
3989         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3990                 err = jbd2_journal_wipe(journal, !really_read_only);
3991         if (!err) {
3992                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3993                 if (save)
3994                         memcpy(save, ((char *) es) +
3995                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3996                 err = jbd2_journal_load(journal);
3997                 if (save)
3998                         memcpy(((char *) es) + EXT4_S_ERR_START,
3999                                save, EXT4_S_ERR_LEN);
4000                 kfree(save);
4001         }
4002
4003         if (err) {
4004                 ext4_msg(sb, KERN_ERR, "error loading journal");
4005                 jbd2_journal_destroy(journal);
4006                 return err;
4007         }
4008
4009         EXT4_SB(sb)->s_journal = journal;
4010         ext4_clear_journal_err(sb, es);
4011
4012         if (!really_read_only && journal_devnum &&
4013             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4014                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4015
4016                 /* Make sure we flush the recovery flag to disk. */
4017                 ext4_commit_super(sb, 1);
4018         }
4019
4020         return 0;
4021 }
4022
4023 static int ext4_commit_super(struct super_block *sb, int sync)
4024 {
4025         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4026         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4027         int error = 0;
4028
4029         if (!sbh)
4030                 return error;
4031         if (buffer_write_io_error(sbh)) {
4032                 /*
4033                  * Oh, dear.  A previous attempt to write the
4034                  * superblock failed.  This could happen because the
4035                  * USB device was yanked out.  Or it could happen to
4036                  * be a transient write error and maybe the block will
4037                  * be remapped.  Nothing we can do but to retry the
4038                  * write and hope for the best.
4039                  */
4040                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4041                        "superblock detected");
4042                 clear_buffer_write_io_error(sbh);
4043                 set_buffer_uptodate(sbh);
4044         }
4045         /*
4046          * If the file system is mounted read-only, don't update the
4047          * superblock write time.  This avoids updating the superblock
4048          * write time when we are mounting the root file system
4049          * read/only but we need to replay the journal; at that point,
4050          * for people who are east of GMT and who make their clock
4051          * tick in localtime for Windows bug-for-bug compatibility,
4052          * the clock is set in the future, and this will cause e2fsck
4053          * to complain and force a full file system check.
4054          */
4055         if (!(sb->s_flags & MS_RDONLY))
4056                 es->s_wtime = cpu_to_le32(get_seconds());
4057         if (sb->s_bdev->bd_part)
4058                 es->s_kbytes_written =
4059                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4060                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4061                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4062         else
4063                 es->s_kbytes_written =
4064                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4065         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
4066                                            &EXT4_SB(sb)->s_freeblocks_counter));
4067         es->s_free_inodes_count =
4068                 cpu_to_le32(percpu_counter_sum_positive(
4069                                 &EXT4_SB(sb)->s_freeinodes_counter));
4070         sb->s_dirt = 0;
4071         BUFFER_TRACE(sbh, "marking dirty");
4072         mark_buffer_dirty(sbh);
4073         if (sync) {
4074                 error = sync_dirty_buffer(sbh);
4075                 if (error)
4076                         return error;
4077
4078                 error = buffer_write_io_error(sbh);
4079                 if (error) {
4080                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4081                                "superblock");
4082                         clear_buffer_write_io_error(sbh);
4083                         set_buffer_uptodate(sbh);
4084                 }
4085         }
4086         return error;
4087 }
4088
4089 /*
4090  * Have we just finished recovery?  If so, and if we are mounting (or
4091  * remounting) the filesystem readonly, then we will end up with a
4092  * consistent fs on disk.  Record that fact.
4093  */
4094 static void ext4_mark_recovery_complete(struct super_block *sb,
4095                                         struct ext4_super_block *es)
4096 {
4097         journal_t *journal = EXT4_SB(sb)->s_journal;
4098
4099         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4100                 BUG_ON(journal != NULL);
4101                 return;
4102         }
4103         jbd2_journal_lock_updates(journal);
4104         if (jbd2_journal_flush(journal) < 0)
4105                 goto out;
4106
4107         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4108             sb->s_flags & MS_RDONLY) {
4109                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4110                 ext4_commit_super(sb, 1);
4111         }
4112
4113 out:
4114         jbd2_journal_unlock_updates(journal);
4115 }
4116
4117 /*
4118  * If we are mounting (or read-write remounting) a filesystem whose journal
4119  * has recorded an error from a previous lifetime, move that error to the
4120  * main filesystem now.
4121  */
4122 static void ext4_clear_journal_err(struct super_block *sb,
4123                                    struct ext4_super_block *es)
4124 {
4125         journal_t *journal;
4126         int j_errno;
4127         const char *errstr;
4128
4129         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4130
4131         journal = EXT4_SB(sb)->s_journal;
4132
4133         /*
4134          * Now check for any error status which may have been recorded in the
4135          * journal by a prior ext4_error() or ext4_abort()
4136          */
4137
4138         j_errno = jbd2_journal_errno(journal);
4139         if (j_errno) {
4140                 char nbuf[16];
4141
4142                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4143                 ext4_warning(sb, "Filesystem error recorded "
4144                              "from previous mount: %s", errstr);
4145                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4146
4147                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4148                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4149                 ext4_commit_super(sb, 1);
4150
4151                 jbd2_journal_clear_err(journal);
4152         }
4153 }
4154
4155 /*
4156  * Force the running and committing transactions to commit,
4157  * and wait on the commit.
4158  */
4159 int ext4_force_commit(struct super_block *sb)
4160 {
4161         journal_t *journal;
4162         int ret = 0;
4163
4164         if (sb->s_flags & MS_RDONLY)
4165                 return 0;
4166
4167         journal = EXT4_SB(sb)->s_journal;
4168         if (journal) {
4169                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4170                 ret = ext4_journal_force_commit(journal);
4171         }
4172
4173         return ret;
4174 }
4175
4176 static void ext4_write_super(struct super_block *sb)
4177 {
4178         lock_super(sb);
4179         ext4_commit_super(sb, 1);
4180         unlock_super(sb);
4181 }
4182
4183 static int ext4_sync_fs(struct super_block *sb, int wait)
4184 {
4185         int ret = 0;
4186         tid_t target;
4187         struct ext4_sb_info *sbi = EXT4_SB(sb);
4188
4189         trace_ext4_sync_fs(sb, wait);
4190         flush_workqueue(sbi->dio_unwritten_wq);
4191         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4192                 if (wait)
4193                         jbd2_log_wait_commit(sbi->s_journal, target);
4194         }
4195         return ret;
4196 }
4197
4198 /*
4199  * LVM calls this function before a (read-only) snapshot is created.  This
4200  * gives us a chance to flush the journal completely and mark the fs clean.
4201  *
4202  * Note that only this function cannot bring a filesystem to be in a clean
4203  * state independently, because ext4 prevents a new handle from being started
4204  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4205  * the upper layer.
4206  */
4207 static int ext4_freeze(struct super_block *sb)
4208 {
4209         int error = 0;
4210         journal_t *journal;
4211
4212         if (sb->s_flags & MS_RDONLY)
4213                 return 0;
4214
4215         journal = EXT4_SB(sb)->s_journal;
4216
4217         /* Now we set up the journal barrier. */
4218         jbd2_journal_lock_updates(journal);
4219
4220         /*
4221          * Don't clear the needs_recovery flag if we failed to flush
4222          * the journal.
4223          */
4224         error = jbd2_journal_flush(journal);
4225         if (error < 0)
4226                 goto out;
4227
4228         /* Journal blocked and flushed, clear needs_recovery flag. */
4229         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4230         error = ext4_commit_super(sb, 1);
4231 out:
4232         /* we rely on s_frozen to stop further updates */
4233         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4234         return error;
4235 }
4236
4237 /*
4238  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4239  * flag here, even though the filesystem is not technically dirty yet.
4240  */
4241 static int ext4_unfreeze(struct super_block *sb)
4242 {
4243         if (sb->s_flags & MS_RDONLY)
4244                 return 0;
4245
4246         lock_super(sb);
4247         /* Reset the needs_recovery flag before the fs is unlocked. */
4248         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4249         ext4_commit_super(sb, 1);
4250         unlock_super(sb);
4251         return 0;
4252 }
4253
4254 /*
4255  * Structure to save mount options for ext4_remount's benefit
4256  */
4257 struct ext4_mount_options {
4258         unsigned long s_mount_opt;
4259         unsigned long s_mount_opt2;
4260         uid_t s_resuid;
4261         gid_t s_resgid;
4262         unsigned long s_commit_interval;
4263         u32 s_min_batch_time, s_max_batch_time;
4264 #ifdef CONFIG_QUOTA
4265         int s_jquota_fmt;
4266         char *s_qf_names[MAXQUOTAS];
4267 #endif
4268 };
4269
4270 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4271 {
4272         struct ext4_super_block *es;
4273         struct ext4_sb_info *sbi = EXT4_SB(sb);
4274         ext4_fsblk_t n_blocks_count = 0;
4275         unsigned long old_sb_flags;
4276         struct ext4_mount_options old_opts;
4277         int enable_quota = 0;
4278         ext4_group_t g;
4279         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4280         int err = 0;
4281 #ifdef CONFIG_QUOTA
4282         int i;
4283 #endif
4284         char *orig_data = kstrdup(data, GFP_KERNEL);
4285
4286         /* Store the original options */
4287         lock_super(sb);
4288         old_sb_flags = sb->s_flags;
4289         old_opts.s_mount_opt = sbi->s_mount_opt;
4290         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4291         old_opts.s_resuid = sbi->s_resuid;
4292         old_opts.s_resgid = sbi->s_resgid;
4293         old_opts.s_commit_interval = sbi->s_commit_interval;
4294         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4295         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4296 #ifdef CONFIG_QUOTA
4297         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4298         for (i = 0; i < MAXQUOTAS; i++)
4299                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4300 #endif
4301         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4302                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4303
4304         /*
4305          * Allow the "check" option to be passed as a remount option.
4306          */
4307         if (!parse_options(data, sb, NULL, &journal_ioprio,
4308                            &n_blocks_count, 1)) {
4309                 err = -EINVAL;
4310                 goto restore_opts;
4311         }
4312
4313         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4314                 ext4_abort(sb, "Abort forced by user");
4315
4316         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4317                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4318
4319         es = sbi->s_es;
4320
4321         if (sbi->s_journal) {
4322                 ext4_init_journal_params(sb, sbi->s_journal);
4323                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4324         }
4325
4326         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4327                 n_blocks_count > ext4_blocks_count(es)) {
4328                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4329                         err = -EROFS;
4330                         goto restore_opts;
4331                 }
4332
4333                 if (*flags & MS_RDONLY) {
4334                         err = dquot_suspend(sb, -1);
4335                         if (err < 0)
4336                                 goto restore_opts;
4337
4338                         /*
4339                          * First of all, the unconditional stuff we have to do
4340                          * to disable replay of the journal when we next remount
4341                          */
4342                         sb->s_flags |= MS_RDONLY;
4343
4344                         /*
4345                          * OK, test if we are remounting a valid rw partition
4346                          * readonly, and if so set the rdonly flag and then
4347                          * mark the partition as valid again.
4348                          */
4349                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4350                             (sbi->s_mount_state & EXT4_VALID_FS))
4351                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4352
4353                         if (sbi->s_journal)
4354                                 ext4_mark_recovery_complete(sb, es);
4355                 } else {
4356                         /* Make sure we can mount this feature set readwrite */
4357                         if (!ext4_feature_set_ok(sb, 0)) {
4358                                 err = -EROFS;
4359                                 goto restore_opts;
4360                         }
4361                         /*
4362                          * Make sure the group descriptor checksums
4363                          * are sane.  If they aren't, refuse to remount r/w.
4364                          */
4365                         for (g = 0; g < sbi->s_groups_count; g++) {
4366                                 struct ext4_group_desc *gdp =
4367                                         ext4_get_group_desc(sb, g, NULL);
4368
4369                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4370                                         ext4_msg(sb, KERN_ERR,
4371                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4372                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4373                                                le16_to_cpu(gdp->bg_checksum));
4374                                         err = -EINVAL;
4375                                         goto restore_opts;
4376                                 }
4377                         }
4378
4379                         /*
4380                          * If we have an unprocessed orphan list hanging
4381                          * around from a previously readonly bdev mount,
4382                          * require a full umount/remount for now.
4383                          */
4384                         if (es->s_last_orphan) {
4385                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4386                                        "remount RDWR because of unprocessed "
4387                                        "orphan inode list.  Please "
4388                                        "umount/remount instead");
4389                                 err = -EINVAL;
4390                                 goto restore_opts;
4391                         }
4392
4393                         /*
4394                          * Mounting a RDONLY partition read-write, so reread
4395                          * and store the current valid flag.  (It may have
4396                          * been changed by e2fsck since we originally mounted
4397                          * the partition.)
4398                          */
4399                         if (sbi->s_journal)
4400                                 ext4_clear_journal_err(sb, es);
4401                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4402                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4403                                 goto restore_opts;
4404                         if (!ext4_setup_super(sb, es, 0))
4405                                 sb->s_flags &= ~MS_RDONLY;
4406                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4407                                                      EXT4_FEATURE_INCOMPAT_MMP))
4408                                 if (ext4_multi_mount_protect(sb,
4409                                                 le64_to_cpu(es->s_mmp_block))) {
4410                                         err = -EROFS;
4411                                         goto restore_opts;
4412                                 }
4413                         enable_quota = 1;
4414                 }
4415         }
4416
4417         /*
4418          * Reinitialize lazy itable initialization thread based on
4419          * current settings
4420          */
4421         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4422                 ext4_unregister_li_request(sb);
4423         else {
4424                 ext4_group_t first_not_zeroed;
4425                 first_not_zeroed = ext4_has_uninit_itable(sb);
4426                 ext4_register_li_request(sb, first_not_zeroed);
4427         }
4428
4429         ext4_setup_system_zone(sb);
4430         if (sbi->s_journal == NULL)
4431                 ext4_commit_super(sb, 1);
4432
4433 #ifdef CONFIG_QUOTA
4434         /* Release old quota file names */
4435         for (i = 0; i < MAXQUOTAS; i++)
4436                 if (old_opts.s_qf_names[i] &&
4437                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4438                         kfree(old_opts.s_qf_names[i]);
4439 #endif
4440         unlock_super(sb);
4441         if (enable_quota)
4442                 dquot_resume(sb, -1);
4443
4444         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4445         kfree(orig_data);
4446         return 0;
4447
4448 restore_opts:
4449         sb->s_flags = old_sb_flags;
4450         sbi->s_mount_opt = old_opts.s_mount_opt;
4451         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4452         sbi->s_resuid = old_opts.s_resuid;
4453         sbi->s_resgid = old_opts.s_resgid;
4454         sbi->s_commit_interval = old_opts.s_commit_interval;
4455         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4456         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4457 #ifdef CONFIG_QUOTA
4458         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4459         for (i = 0; i < MAXQUOTAS; i++) {
4460                 if (sbi->s_qf_names[i] &&
4461                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4462                         kfree(sbi->s_qf_names[i]);
4463                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4464         }
4465 #endif
4466         unlock_super(sb);
4467         kfree(orig_data);
4468         return err;
4469 }
4470
4471 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4472 {
4473         struct super_block *sb = dentry->d_sb;
4474         struct ext4_sb_info *sbi = EXT4_SB(sb);
4475         struct ext4_super_block *es = sbi->s_es;
4476         u64 fsid;
4477         s64 bfree;
4478
4479         if (test_opt(sb, MINIX_DF)) {
4480                 sbi->s_overhead_last = 0;
4481         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4482                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4483                 ext4_fsblk_t overhead = 0;
4484
4485                 /*
4486                  * Compute the overhead (FS structures).  This is constant
4487                  * for a given filesystem unless the number of block groups
4488                  * changes so we cache the previous value until it does.
4489                  */
4490
4491                 /*
4492                  * All of the blocks before first_data_block are
4493                  * overhead
4494                  */
4495                 overhead = le32_to_cpu(es->s_first_data_block);
4496
4497                 /*
4498                  * Add the overhead attributed to the superblock and
4499                  * block group descriptors.  If the sparse superblocks
4500                  * feature is turned on, then not all groups have this.
4501                  */
4502                 for (i = 0; i < ngroups; i++) {
4503                         overhead += ext4_bg_has_super(sb, i) +
4504                                 ext4_bg_num_gdb(sb, i);
4505                         cond_resched();
4506                 }
4507
4508                 /*
4509                  * Every block group has an inode bitmap, a block
4510                  * bitmap, and an inode table.
4511                  */
4512                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4513                 sbi->s_overhead_last = overhead;
4514                 smp_wmb();
4515                 sbi->s_blocks_last = ext4_blocks_count(es);
4516         }
4517
4518         buf->f_type = EXT4_SUPER_MAGIC;
4519         buf->f_bsize = sb->s_blocksize;
4520         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4521         bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4522                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4523         /* prevent underflow in case that few free space is available */
4524         buf->f_bfree = max_t(s64, bfree, 0);
4525         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4526         if (buf->f_bfree < ext4_r_blocks_count(es))
4527                 buf->f_bavail = 0;
4528         buf->f_files = le32_to_cpu(es->s_inodes_count);
4529         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4530         buf->f_namelen = EXT4_NAME_LEN;
4531         fsid = le64_to_cpup((void *)es->s_uuid) ^
4532                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4533         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4534         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4535
4536         return 0;
4537 }
4538
4539 /* Helper function for writing quotas on sync - we need to start transaction
4540  * before quota file is locked for write. Otherwise the are possible deadlocks:
4541  * Process 1                         Process 2
4542  * ext4_create()                     quota_sync()
4543  *   jbd2_journal_start()                  write_dquot()
4544  *   dquot_initialize()                         down(dqio_mutex)
4545  *     down(dqio_mutex)                    jbd2_journal_start()
4546  *
4547  */
4548
4549 #ifdef CONFIG_QUOTA
4550
4551 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4552 {
4553         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4554 }
4555
4556 static int ext4_write_dquot(struct dquot *dquot)
4557 {
4558         int ret, err;
4559         handle_t *handle;
4560         struct inode *inode;
4561
4562         inode = dquot_to_inode(dquot);
4563         handle = ext4_journal_start(inode,
4564                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4565         if (IS_ERR(handle))
4566                 return PTR_ERR(handle);
4567         ret = dquot_commit(dquot);
4568         err = ext4_journal_stop(handle);
4569         if (!ret)
4570                 ret = err;
4571         return ret;
4572 }
4573
4574 static int ext4_acquire_dquot(struct dquot *dquot)
4575 {
4576         int ret, err;
4577         handle_t *handle;
4578
4579         handle = ext4_journal_start(dquot_to_inode(dquot),
4580                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4581         if (IS_ERR(handle))
4582                 return PTR_ERR(handle);
4583         ret = dquot_acquire(dquot);
4584         err = ext4_journal_stop(handle);
4585         if (!ret)
4586                 ret = err;
4587         return ret;
4588 }
4589
4590 static int ext4_release_dquot(struct dquot *dquot)
4591 {
4592         int ret, err;
4593         handle_t *handle;
4594
4595         handle = ext4_journal_start(dquot_to_inode(dquot),
4596                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4597         if (IS_ERR(handle)) {
4598                 /* Release dquot anyway to avoid endless cycle in dqput() */
4599                 dquot_release(dquot);
4600                 return PTR_ERR(handle);
4601         }
4602         ret = dquot_release(dquot);
4603         err = ext4_journal_stop(handle);
4604         if (!ret)
4605                 ret = err;
4606         return ret;
4607 }
4608
4609 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4610 {
4611         /* Are we journaling quotas? */
4612         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4613             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4614                 dquot_mark_dquot_dirty(dquot);
4615                 return ext4_write_dquot(dquot);
4616         } else {
4617                 return dquot_mark_dquot_dirty(dquot);
4618         }
4619 }
4620
4621 static int ext4_write_info(struct super_block *sb, int type)
4622 {
4623         int ret, err;
4624         handle_t *handle;
4625
4626         /* Data block + inode block */
4627         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4628         if (IS_ERR(handle))
4629                 return PTR_ERR(handle);
4630         ret = dquot_commit_info(sb, type);
4631         err = ext4_journal_stop(handle);
4632         if (!ret)
4633                 ret = err;
4634         return ret;
4635 }
4636
4637 /*
4638  * Turn on quotas during mount time - we need to find
4639  * the quota file and such...
4640  */
4641 static int ext4_quota_on_mount(struct super_block *sb, int type)
4642 {
4643         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4644                                         EXT4_SB(sb)->s_jquota_fmt, type);
4645 }
4646
4647 /*
4648  * Standard function to be called on quota_on
4649  */
4650 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4651                          struct path *path)
4652 {
4653         int err;
4654
4655         if (!test_opt(sb, QUOTA))
4656                 return -EINVAL;
4657
4658         /* Quotafile not on the same filesystem? */
4659         if (path->mnt->mnt_sb != sb)
4660                 return -EXDEV;
4661         /* Journaling quota? */
4662         if (EXT4_SB(sb)->s_qf_names[type]) {
4663                 /* Quotafile not in fs root? */
4664                 if (path->dentry->d_parent != sb->s_root)
4665                         ext4_msg(sb, KERN_WARNING,
4666                                 "Quota file not on filesystem root. "
4667                                 "Journaled quota will not work");
4668         }
4669
4670         /*
4671          * When we journal data on quota file, we have to flush journal to see
4672          * all updates to the file when we bypass pagecache...
4673          */
4674         if (EXT4_SB(sb)->s_journal &&
4675             ext4_should_journal_data(path->dentry->d_inode)) {
4676                 /*
4677                  * We don't need to lock updates but journal_flush() could
4678                  * otherwise be livelocked...
4679                  */
4680                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4681                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4682                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4683                 if (err)
4684                         return err;
4685         }
4686
4687         return dquot_quota_on(sb, type, format_id, path);
4688 }
4689
4690 static int ext4_quota_off(struct super_block *sb, int type)
4691 {
4692         struct inode *inode = sb_dqopt(sb)->files[type];
4693         handle_t *handle;
4694
4695         /* Force all delayed allocation blocks to be allocated.
4696          * Caller already holds s_umount sem */
4697         if (test_opt(sb, DELALLOC))
4698                 sync_filesystem(sb);
4699
4700         if (!inode)
4701                 goto out;
4702
4703         /* Update modification times of quota files when userspace can
4704          * start looking at them */
4705         handle = ext4_journal_start(inode, 1);
4706         if (IS_ERR(handle))
4707                 goto out;
4708         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4709         ext4_mark_inode_dirty(handle, inode);
4710         ext4_journal_stop(handle);
4711
4712 out:
4713         return dquot_quota_off(sb, type);
4714 }
4715
4716 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4717  * acquiring the locks... As quota files are never truncated and quota code
4718  * itself serializes the operations (and no one else should touch the files)
4719  * we don't have to be afraid of races */
4720 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4721                                size_t len, loff_t off)
4722 {
4723         struct inode *inode = sb_dqopt(sb)->files[type];
4724         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4725         int err = 0;
4726         int offset = off & (sb->s_blocksize - 1);
4727         int tocopy;
4728         size_t toread;
4729         struct buffer_head *bh;
4730         loff_t i_size = i_size_read(inode);
4731
4732         if (off > i_size)
4733                 return 0;
4734         if (off+len > i_size)
4735                 len = i_size-off;
4736         toread = len;
4737         while (toread > 0) {
4738                 tocopy = sb->s_blocksize - offset < toread ?
4739                                 sb->s_blocksize - offset : toread;
4740                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4741                 if (err)
4742                         return err;
4743                 if (!bh)        /* A hole? */
4744                         memset(data, 0, tocopy);
4745                 else
4746                         memcpy(data, bh->b_data+offset, tocopy);
4747                 brelse(bh);
4748                 offset = 0;
4749                 toread -= tocopy;
4750                 data += tocopy;
4751                 blk++;
4752         }
4753         return len;
4754 }
4755
4756 /* Write to quotafile (we know the transaction is already started and has
4757  * enough credits) */
4758 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4759                                 const char *data, size_t len, loff_t off)
4760 {
4761         struct inode *inode = sb_dqopt(sb)->files[type];
4762         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4763         int err = 0;
4764         int offset = off & (sb->s_blocksize - 1);
4765         struct buffer_head *bh;
4766         handle_t *handle = journal_current_handle();
4767
4768         if (EXT4_SB(sb)->s_journal && !handle) {
4769                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4770                         " cancelled because transaction is not started",
4771                         (unsigned long long)off, (unsigned long long)len);
4772                 return -EIO;
4773         }
4774         /*
4775          * Since we account only one data block in transaction credits,
4776          * then it is impossible to cross a block boundary.
4777          */
4778         if (sb->s_blocksize - offset < len) {
4779                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4780                         " cancelled because not block aligned",
4781                         (unsigned long long)off, (unsigned long long)len);
4782                 return -EIO;
4783         }
4784
4785         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4786         bh = ext4_bread(handle, inode, blk, 1, &err);
4787         if (!bh)
4788                 goto out;
4789         err = ext4_journal_get_write_access(handle, bh);
4790         if (err) {
4791                 brelse(bh);
4792                 goto out;
4793         }
4794         lock_buffer(bh);
4795         memcpy(bh->b_data+offset, data, len);
4796         flush_dcache_page(bh->b_page);
4797         unlock_buffer(bh);
4798         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4799         brelse(bh);
4800 out:
4801         if (err) {
4802                 mutex_unlock(&inode->i_mutex);
4803                 return err;
4804         }
4805         if (inode->i_size < off + len) {
4806                 i_size_write(inode, off + len);
4807                 EXT4_I(inode)->i_disksize = inode->i_size;
4808                 ext4_mark_inode_dirty(handle, inode);
4809         }
4810         mutex_unlock(&inode->i_mutex);
4811         return len;
4812 }
4813
4814 #endif
4815
4816 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4817                        const char *dev_name, void *data)
4818 {
4819         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4820 }
4821
4822 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4823 static inline void register_as_ext2(void)
4824 {
4825         int err = register_filesystem(&ext2_fs_type);
4826         if (err)
4827                 printk(KERN_WARNING
4828                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4829 }
4830
4831 static inline void unregister_as_ext2(void)
4832 {
4833         unregister_filesystem(&ext2_fs_type);
4834 }
4835
4836 static inline int ext2_feature_set_ok(struct super_block *sb)
4837 {
4838         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4839                 return 0;
4840         if (sb->s_flags & MS_RDONLY)
4841                 return 1;
4842         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4843                 return 0;
4844         return 1;
4845 }
4846 MODULE_ALIAS("ext2");
4847 #else
4848 static inline void register_as_ext2(void) { }
4849 static inline void unregister_as_ext2(void) { }
4850 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4851 #endif
4852
4853 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4854 static inline void register_as_ext3(void)
4855 {
4856         int err = register_filesystem(&ext3_fs_type);
4857         if (err)
4858                 printk(KERN_WARNING
4859                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4860 }
4861
4862 static inline void unregister_as_ext3(void)
4863 {
4864         unregister_filesystem(&ext3_fs_type);
4865 }
4866
4867 static inline int ext3_feature_set_ok(struct super_block *sb)
4868 {
4869         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4870                 return 0;
4871         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4872                 return 0;
4873         if (sb->s_flags & MS_RDONLY)
4874                 return 1;
4875         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4876                 return 0;
4877         return 1;
4878 }
4879 MODULE_ALIAS("ext3");
4880 #else
4881 static inline void register_as_ext3(void) { }
4882 static inline void unregister_as_ext3(void) { }
4883 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4884 #endif
4885
4886 static struct file_system_type ext4_fs_type = {
4887         .owner          = THIS_MODULE,
4888         .name           = "ext4",
4889         .mount          = ext4_mount,
4890         .kill_sb        = kill_block_super,
4891         .fs_flags       = FS_REQUIRES_DEV,
4892 };
4893
4894 static int __init ext4_init_feat_adverts(void)
4895 {
4896         struct ext4_features *ef;
4897         int ret = -ENOMEM;
4898
4899         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4900         if (!ef)
4901                 goto out;
4902
4903         ef->f_kobj.kset = ext4_kset;
4904         init_completion(&ef->f_kobj_unregister);
4905         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4906                                    "features");
4907         if (ret) {
4908                 kfree(ef);
4909                 goto out;
4910         }
4911
4912         ext4_feat = ef;
4913         ret = 0;
4914 out:
4915         return ret;
4916 }
4917
4918 static void ext4_exit_feat_adverts(void)
4919 {
4920         kobject_put(&ext4_feat->f_kobj);
4921         wait_for_completion(&ext4_feat->f_kobj_unregister);
4922         kfree(ext4_feat);
4923 }
4924
4925 /* Shared across all ext4 file systems */
4926 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4927 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4928
4929 static int __init ext4_init_fs(void)
4930 {
4931         int i, err;
4932
4933         ext4_check_flag_values();
4934
4935         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4936                 mutex_init(&ext4__aio_mutex[i]);
4937                 init_waitqueue_head(&ext4__ioend_wq[i]);
4938         }
4939
4940         err = ext4_init_pageio();
4941         if (err)
4942                 return err;
4943         err = ext4_init_system_zone();
4944         if (err)
4945                 goto out7;
4946         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4947         if (!ext4_kset)
4948                 goto out6;
4949         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4950         if (!ext4_proc_root)
4951                 goto out5;
4952
4953         err = ext4_init_feat_adverts();
4954         if (err)
4955                 goto out4;
4956
4957         err = ext4_init_mballoc();
4958         if (err)
4959                 goto out3;
4960
4961         err = ext4_init_xattr();
4962         if (err)
4963                 goto out2;
4964         err = init_inodecache();
4965         if (err)
4966                 goto out1;
4967         register_as_ext3();
4968         register_as_ext2();
4969         err = register_filesystem(&ext4_fs_type);
4970         if (err)
4971                 goto out;
4972
4973         ext4_li_info = NULL;
4974         mutex_init(&ext4_li_mtx);
4975         return 0;
4976 out:
4977         unregister_as_ext2();
4978         unregister_as_ext3();
4979         destroy_inodecache();
4980 out1:
4981         ext4_exit_xattr();
4982 out2:
4983         ext4_exit_mballoc();
4984 out3:
4985         ext4_exit_feat_adverts();
4986 out4:
4987         remove_proc_entry("fs/ext4", NULL);
4988 out5:
4989         kset_unregister(ext4_kset);
4990 out6:
4991         ext4_exit_system_zone();
4992 out7:
4993         ext4_exit_pageio();
4994         return err;
4995 }
4996
4997 static void __exit ext4_exit_fs(void)
4998 {
4999         ext4_destroy_lazyinit_thread();
5000         unregister_as_ext2();
5001         unregister_as_ext3();
5002         unregister_filesystem(&ext4_fs_type);
5003         destroy_inodecache();
5004         ext4_exit_xattr();
5005         ext4_exit_mballoc();
5006         ext4_exit_feat_adverts();
5007         remove_proc_entry("fs/ext4", NULL);
5008         kset_unregister(ext4_kset);
5009         ext4_exit_system_zone();
5010         ext4_exit_pageio();
5011 }
5012
5013 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5014 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5015 MODULE_LICENSE("GPL");
5016 module_init(ext4_init_fs)
5017 module_exit(ext4_exit_fs)