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 struct ext4_lazy_init *ext4_li_info;
  58 struct mutex ext4_li_mtx;
  59 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 void ext4_destroy_lazyinit_thread(void);
  79 static void ext4_unregister_li_request(struct super_block *sb);
  80
  81 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  82 static struct file_system_type ext3_fs_type = {
  83         .owner          = THIS_MODULE,
  84         .name           = "ext3",
  85         .mount          = ext4_mount,
  86         .kill_sb        = kill_block_super,
  87         .fs_flags       = FS_REQUIRES_DEV,
  88 };
  89 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  90 #else
  91 #define IS_EXT3_SB(sb) (0)
  92 #endif
  93
  94 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  95                                struct ext4_group_desc *bg)
  96 {
  97         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  98                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  99                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 100 }
 101
 102 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 103                                struct ext4_group_desc *bg)
 104 {
 105         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 106                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 107                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 108 }
 109
 110 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 111                               struct ext4_group_desc *bg)
 112 {
 113         return le32_to_cpu(bg->bg_inode_table_lo) |
 114                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 115                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 116 }
 117
 118 __u32 ext4_free_blks_count(struct super_block *sb,
 119                               struct ext4_group_desc *bg)
 120 {
 121         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 122                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 123                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 124 }
 125
 126 __u32 ext4_free_inodes_count(struct super_block *sb,
 127                               struct ext4_group_desc *bg)
 128 {
 129         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 130                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 131                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 132 }
 133
 134 __u32 ext4_used_dirs_count(struct super_block *sb,
 135                               struct ext4_group_desc *bg)
 136 {
 137         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 138                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 139                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 140 }
 141
 142 __u32 ext4_itable_unused_count(struct super_block *sb,
 143                               struct ext4_group_desc *bg)
 144 {
 145         return le16_to_cpu(bg->bg_itable_unused_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 148 }
 149
 150 void ext4_block_bitmap_set(struct super_block *sb,
 151                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 152 {
 153         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 154         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 155                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 156 }
 157
 158 void ext4_inode_bitmap_set(struct super_block *sb,
 159                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 160 {
 161         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 162         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 163                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 164 }
 165
 166 void ext4_inode_table_set(struct super_block *sb,
 167                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 168 {
 169         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 170         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 171                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 172 }
 173
 174 void ext4_free_blks_set(struct super_block *sb,
 175                           struct ext4_group_desc *bg, __u32 count)
 176 {
 177         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 178         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 179                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 180 }
 181
 182 void ext4_free_inodes_set(struct super_block *sb,
 183                           struct ext4_group_desc *bg, __u32 count)
 184 {
 185         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 186         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 187                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 188 }
 189
 190 void ext4_used_dirs_set(struct super_block *sb,
 191                           struct ext4_group_desc *bg, __u32 count)
 192 {
 193         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 194         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 195                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 196 }
 197
 198 void ext4_itable_unused_set(struct super_block *sb,
 199                           struct ext4_group_desc *bg, __u32 count)
 200 {
 201         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 204 }
 205
 206
 207 /* Just increment the non-pointer handle value */
 208 static handle_t *ext4_get_nojournal(void)
 209 {
 210         handle_t *handle = current->journal_info;
 211         unsigned long ref_cnt = (unsigned long)handle;
 212
 213         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 214
 215         ref_cnt++;
 216         handle = (handle_t *)ref_cnt;
 217
 218         current->journal_info = handle;
 219         return handle;
 220 }
 221
 222
 223 /* Decrement the non-pointer handle value */
 224 static void ext4_put_nojournal(handle_t *handle)
 225 {
 226         unsigned long ref_cnt = (unsigned long)handle;
 227
 228         BUG_ON(ref_cnt == 0);
 229
 230         ref_cnt--;
 231         handle = (handle_t *)ref_cnt;
 232
 233         current->journal_info = handle;
 234 }
 235
 236 /*
 237  * Wrappers for jbd2_journal_start/end.
 238  *
 239  * The only special thing we need to do here is to make sure that all
 240  * journal_end calls result in the superblock being marked dirty, so
 241  * that sync() will call the filesystem's write_super callback if
 242  * appropriate.
 243  */
 244 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 245 {
 246         journal_t *journal;
 247
 248         if (sb->s_flags & MS_RDONLY)
 249                 return ERR_PTR(-EROFS);
 250
 251         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 252         /* Special case here: if the journal has aborted behind our
 253          * backs (eg. EIO in the commit thread), then we still need to
 254          * take the FS itself readonly cleanly. */
 255         journal = EXT4_SB(sb)->s_journal;
 256         if (journal) {
 257                 if (is_journal_aborted(journal)) {
 258                         ext4_abort(sb, "Detected aborted journal");
 259                         return ERR_PTR(-EROFS);
 260                 }
 261                 return jbd2_journal_start(journal, nblocks);
 262         }
 263         return ext4_get_nojournal();
 264 }
 265
 266 /*
 267  * The only special thing we need to do here is to make sure that all
 268  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 269  * that sync() will call the filesystem's write_super callback if
 270  * appropriate.
 271  */
 272 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 273 {
 274         struct super_block *sb;
 275         int err;
 276         int rc;
 277
 278         if (!ext4_handle_valid(handle)) {
 279                 ext4_put_nojournal(handle);
 280                 return 0;
 281         }
 282         sb = handle->h_transaction->t_journal->j_private;
 283         err = handle->h_err;
 284         rc = jbd2_journal_stop(handle);
 285
 286         if (!err)
 287                 err = rc;
 288         if (err)
 289                 __ext4_std_error(sb, where, line, err);
 290         return err;
 291 }
 292
 293 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 294                                const char *err_fn, struct buffer_head *bh,
 295                                handle_t *handle, int err)
 296 {
 297         char nbuf[16];
 298         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 299
 300         BUG_ON(!ext4_handle_valid(handle));
 301
 302         if (bh)
 303                 BUFFER_TRACE(bh, "abort");
 304
 305         if (!handle->h_err)
 306                 handle->h_err = err;
 307
 308         if (is_handle_aborted(handle))
 309                 return;
 310
 311         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 312                caller, line, errstr, err_fn);
 313
 314         jbd2_journal_abort_handle(handle);
 315 }
 316
 317 static void __save_error_info(struct super_block *sb, const char *func,
 318                             unsigned int line)
 319 {
 320         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 321
 322         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 324         es->s_last_error_time = cpu_to_le32(get_seconds());
 325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 326         es->s_last_error_line = cpu_to_le32(line);
 327         if (!es->s_first_error_time) {
 328                 es->s_first_error_time = es->s_last_error_time;
 329                 strncpy(es->s_first_error_func, func,
 330                         sizeof(es->s_first_error_func));
 331                 es->s_first_error_line = cpu_to_le32(line);
 332                 es->s_first_error_ino = es->s_last_error_ino;
 333                 es->s_first_error_block = es->s_last_error_block;
 334         }
 335         /*
 336          * Start the daily error reporting function if it hasn't been
 337          * started already
 338          */
 339         if (!es->s_error_count)
 340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 341         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 342 }
 343
 344 static void save_error_info(struct super_block *sb, const char *func,
 345                             unsigned int line)
 346 {
 347         __save_error_info(sb, func, line);
 348         ext4_commit_super(sb, 1);
 349 }
 350
 351
 352 /* Deal with the reporting of failure conditions on a filesystem such as
 353  * inconsistencies detected or read IO failures.
 354  *
 355  * On ext2, we can store the error state of the filesystem in the
 356  * superblock.  That is not possible on ext4, because we may have other
 357  * write ordering constraints on the superblock which prevent us from
 358  * writing it out straight away; and given that the journal is about to
 359  * be aborted, we can't rely on the current, or future, transactions to
 360  * write out the superblock safely.
 361  *
 362  * We'll just use the jbd2_journal_abort() error code to record an error in
 363  * the journal instead.  On recovery, the journal will complain about
 364  * that error until we've noted it down and cleared it.
 365  */
 366
 367 static void ext4_handle_error(struct super_block *sb)
 368 {
 369         if (sb->s_flags & MS_RDONLY)
 370                 return;
 371
 372         if (!test_opt(sb, ERRORS_CONT)) {
 373                 journal_t *journal = EXT4_SB(sb)->s_journal;
 374
 375                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 376                 if (journal)
 377                         jbd2_journal_abort(journal, -EIO);
 378         }
 379         if (test_opt(sb, ERRORS_RO)) {
 380                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 381                 sb->s_flags |= MS_RDONLY;
 382         }
 383         if (test_opt(sb, ERRORS_PANIC))
 384                 panic("EXT4-fs (device %s): panic forced after error\n",
 385                         sb->s_id);
 386 }
 387
 388 void __ext4_error(struct super_block *sb, const char *function,
 389                   unsigned int line, const char *fmt, ...)
 390 {
 391         va_list args;
 392
 393         va_start(args, fmt);
 394         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: ",
 395                sb->s_id, function, line, current->comm);
 396         vprintk(fmt, args);
 397         printk("\n");
 398         va_end(args);
 399
 400         ext4_handle_error(sb);
 401 }
 402
 403 void ext4_error_inode(struct inode *inode, const char *function,
 404                       unsigned int line, ext4_fsblk_t block,
 405                       const char *fmt, ...)
 406 {
 407         va_list args;
 408         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 409
 410         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 411         es->s_last_error_block = cpu_to_le64(block);
 412         save_error_info(inode->i_sb, function, line);
 413         va_start(args, fmt);
 414         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 415                inode->i_sb->s_id, function, line, inode->i_ino);
 416         if (block)
 417                 printk("block %llu: ", block);
 418         printk("comm %s: ", current->comm);
 419         vprintk(fmt, args);
 420         printk("\n");
 421         va_end(args);
 422
 423         ext4_handle_error(inode->i_sb);
 424 }
 425
 426 void ext4_error_file(struct file *file, const char *function,
 427                      unsigned int line, const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct ext4_super_block *es;
 431         struct inode *inode = file->f_dentry->d_inode;
 432         char pathname[80], *path;
 433
 434         es = EXT4_SB(inode->i_sb)->s_es;
 435         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 436         save_error_info(inode->i_sb, function, line);
 437         va_start(args, fmt);
 438         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 439         if (!path)
 440                 path = "(unknown)";
 441         printk(KERN_CRIT
 442                "EXT4-fs error (device %s): %s:%d: inode #%lu "
 443                "(comm %s path %s): ",
 444                inode->i_sb->s_id, function, line, inode->i_ino,
 445                current->comm, path);
 446         vprintk(fmt, args);
 447         printk("\n");
 448         va_end(args);
 449
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 static const char *ext4_decode_error(struct super_block *sb, int errno,
 454                                      char nbuf[16])
 455 {
 456         char *errstr = NULL;
 457
 458         switch (errno) {
 459         case -EIO:
 460                 errstr = "IO failure";
 461                 break;
 462         case -ENOMEM:
 463                 errstr = "Out of memory";
 464                 break;
 465         case -EROFS:
 466                 if (!sb || (EXT4_SB(sb)->s_journal &&
 467                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 468                         errstr = "Journal has aborted";
 469                 else
 470                         errstr = "Readonly filesystem";
 471                 break;
 472         default:
 473                 /* If the caller passed in an extra buffer for unknown
 474                  * errors, textualise them now.  Else we just return
 475                  * NULL. */
 476                 if (nbuf) {
 477                         /* Check for truncated error codes... */
 478                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 479                                 errstr = nbuf;
 480                 }
 481                 break;
 482         }
 483
 484         return errstr;
 485 }
 486
 487 /* __ext4_std_error decodes expected errors from journaling functions
 488  * automatically and invokes the appropriate error response.  */
 489
 490 void __ext4_std_error(struct super_block *sb, const char *function,
 491                       unsigned int line, int errno)
 492 {
 493         char nbuf[16];
 494         const char *errstr;
 495
 496         /* Special case: if the error is EROFS, and we're not already
 497          * inside a transaction, then there's really no point in logging
 498          * an error. */
 499         if (errno == -EROFS && journal_current_handle() == NULL &&
 500             (sb->s_flags & MS_RDONLY))
 501                 return;
 502
 503         errstr = ext4_decode_error(sb, errno, nbuf);
 504         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 505                sb->s_id, function, line, errstr);
 506         save_error_info(sb, function, line);
 507
 508         ext4_handle_error(sb);
 509 }
 510
 511 /*
 512  * ext4_abort is a much stronger failure handler than ext4_error.  The
 513  * abort function may be used to deal with unrecoverable failures such
 514  * as journal IO errors or ENOMEM at a critical moment in log management.
 515  *
 516  * We unconditionally force the filesystem into an ABORT|READONLY state,
 517  * unless the error response on the fs has been set to panic in which
 518  * case we take the easy way out and panic immediately.
 519  */
 520
 521 void __ext4_abort(struct super_block *sb, const char *function,
 522                 unsigned int line, const char *fmt, ...)
 523 {
 524         va_list args;
 525
 526         save_error_info(sb, function, line);
 527         va_start(args, fmt);
 528         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 529                function, line);
 530         vprintk(fmt, args);
 531         printk("\n");
 532         va_end(args);
 533
 534         if ((sb->s_flags & MS_RDONLY) == 0) {
 535                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 536                 sb->s_flags |= MS_RDONLY;
 537                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 538                 if (EXT4_SB(sb)->s_journal)
 539                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 540                 save_error_info(sb, function, line);
 541         }
 542         if (test_opt(sb, ERRORS_PANIC))
 543                 panic("EXT4-fs panic from previous error\n");
 544 }
 545
 546 void ext4_msg (struct super_block * sb, const char *prefix,
 547                    const char *fmt, ...)
 548 {
 549         va_list args;
 550
 551         va_start(args, fmt);
 552         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 553         vprintk(fmt, args);
 554         printk("\n");
 555         va_end(args);
 556 }
 557
 558 void __ext4_warning(struct super_block *sb, const char *function,
 559                     unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         va_start(args, fmt);
 564         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: ",
 565                sb->s_id, function, line);
 566         vprintk(fmt, args);
 567         printk("\n");
 568         va_end(args);
 569 }
 570
 571 void __ext4_grp_locked_error(const char *function, unsigned int line,
 572                              struct super_block *sb, ext4_group_t grp,
 573                              unsigned long ino, ext4_fsblk_t block,
 574                              const char *fmt, ...)
 575 __releases(bitlock)
 576 __acquires(bitlock)
 577 {
 578         va_list args;
 579         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 580
 581         es->s_last_error_ino = cpu_to_le32(ino);
 582         es->s_last_error_block = cpu_to_le64(block);
 583         __save_error_info(sb, function, line);
 584         va_start(args, fmt);
 585         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u",
 586                sb->s_id, function, line, grp);
 587         if (ino)
 588                 printk("inode %lu: ", ino);
 589         if (block)
 590                 printk("block %llu:", (unsigned long long) block);
 591         vprintk(fmt, args);
 592         printk("\n");
 593         va_end(args);
 594
 595         if (test_opt(sb, ERRORS_CONT)) {
 596                 ext4_commit_super(sb, 0);
 597                 return;
 598         }
 599
 600         ext4_unlock_group(sb, grp);
 601         ext4_handle_error(sb);
 602         /*
 603          * We only get here in the ERRORS_RO case; relocking the group
 604          * may be dangerous, but nothing bad will happen since the
 605          * filesystem will have already been marked read/only and the
 606          * journal has been aborted.  We return 1 as a hint to callers
 607          * who might what to use the return value from
 608          * ext4_grp_locked_error() to distinguish beween the
 609          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 610          * aggressively from the ext4 function in question, with a
 611          * more appropriate error code.
 612          */
 613         ext4_lock_group(sb, grp);
 614         return;
 615 }
 616
 617 void ext4_update_dynamic_rev(struct super_block *sb)
 618 {
 619         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 620
 621         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 622                 return;
 623
 624         ext4_warning(sb,
 625                      "updating to rev %d because of new feature flag, "
 626                      "running e2fsck is recommended",
 627                      EXT4_DYNAMIC_REV);
 628
 629         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 630         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 631         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 632         /* leave es->s_feature_*compat flags alone */
 633         /* es->s_uuid will be set by e2fsck if empty */
 634
 635         /*
 636          * The rest of the superblock fields should be zero, and if not it
 637          * means they are likely already in use, so leave them alone.  We
 638          * can leave it up to e2fsck to clean up any inconsistencies there.
 639          */
 640 }
 641
 642 /*
 643  * Open the external journal device
 644  */
 645 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 646 {
 647         struct block_device *bdev;
 648         char b[BDEVNAME_SIZE];
 649
 650         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 651         if (IS_ERR(bdev))
 652                 goto fail;
 653         return bdev;
 654
 655 fail:
 656         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 657                         __bdevname(dev, b), PTR_ERR(bdev));
 658         return NULL;
 659 }
 660
 661 /*
 662  * Release the journal device
 663  */
 664 static int ext4_blkdev_put(struct block_device *bdev)
 665 {
 666         bd_release(bdev);
 667         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 668 }
 669
 670 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 671 {
 672         struct block_device *bdev;
 673         int ret = -ENODEV;
 674
 675         bdev = sbi->journal_bdev;
 676         if (bdev) {
 677                 ret = ext4_blkdev_put(bdev);
 678                 sbi->journal_bdev = NULL;
 679         }
 680         return ret;
 681 }
 682
 683 static inline struct inode *orphan_list_entry(struct list_head *l)
 684 {
 685         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 686 }
 687
 688 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 689 {
 690         struct list_head *l;
 691
 692         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 693                  le32_to_cpu(sbi->s_es->s_last_orphan));
 694
 695         printk(KERN_ERR "sb_info orphan list:\n");
 696         list_for_each(l, &sbi->s_orphan) {
 697                 struct inode *inode = orphan_list_entry(l);
 698                 printk(KERN_ERR "  "
 699                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 700                        inode->i_sb->s_id, inode->i_ino, inode,
 701                        inode->i_mode, inode->i_nlink,
 702                        NEXT_ORPHAN(inode));
 703         }
 704 }
 705
 706 static void ext4_put_super(struct super_block *sb)
 707 {
 708         struct ext4_sb_info *sbi = EXT4_SB(sb);
 709         struct ext4_super_block *es = sbi->s_es;
 710         int i, err;
 711
 712         ext4_unregister_li_request(sb);
 713         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 714
 715         flush_workqueue(sbi->dio_unwritten_wq);
 716         destroy_workqueue(sbi->dio_unwritten_wq);
 717
 718         lock_super(sb);
 719         if (sb->s_dirt)
 720                 ext4_commit_super(sb, 1);
 721
 722         if (sbi->s_journal) {
 723                 err = jbd2_journal_destroy(sbi->s_journal);
 724                 sbi->s_journal = NULL;
 725                 if (err < 0)
 726                         ext4_abort(sb, "Couldn't clean up the journal");
 727         }
 728
 729         del_timer(&sbi->s_err_report);
 730         ext4_release_system_zone(sb);
 731         ext4_mb_release(sb);
 732         ext4_ext_release(sb);
 733         ext4_xattr_put_super(sb);
 734
 735         if (!(sb->s_flags & MS_RDONLY)) {
 736                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 737                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 738                 ext4_commit_super(sb, 1);
 739         }
 740         if (sbi->s_proc) {
 741                 remove_proc_entry(sb->s_id, ext4_proc_root);
 742         }
 743         kobject_del(&sbi->s_kobj);
 744
 745         for (i = 0; i < sbi->s_gdb_count; i++)
 746                 brelse(sbi->s_group_desc[i]);
 747         kfree(sbi->s_group_desc);
 748         if (is_vmalloc_addr(sbi->s_flex_groups))
 749                 vfree(sbi->s_flex_groups);
 750         else
 751                 kfree(sbi->s_flex_groups);
 752         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 753         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 754         percpu_counter_destroy(&sbi->s_dirs_counter);
 755         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 756         brelse(sbi->s_sbh);
 757 #ifdef CONFIG_QUOTA
 758         for (i = 0; i < MAXQUOTAS; i++)
 759                 kfree(sbi->s_qf_names[i]);
 760 #endif
 761
 762         /* Debugging code just in case the in-memory inode orphan list
 763          * isn't empty.  The on-disk one can be non-empty if we've
 764          * detected an error and taken the fs readonly, but the
 765          * in-memory list had better be clean by this point. */
 766         if (!list_empty(&sbi->s_orphan))
 767                 dump_orphan_list(sb, sbi);
 768         J_ASSERT(list_empty(&sbi->s_orphan));
 769
 770         invalidate_bdev(sb->s_bdev);
 771         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 772                 /*
 773                  * Invalidate the journal device's buffers.  We don't want them
 774                  * floating about in memory - the physical journal device may
 775                  * hotswapped, and it breaks the `ro-after' testing code.
 776                  */
 777                 sync_blockdev(sbi->journal_bdev);
 778                 invalidate_bdev(sbi->journal_bdev);
 779                 ext4_blkdev_remove(sbi);
 780         }
 781         sb->s_fs_info = NULL;
 782         /*
 783          * Now that we are completely done shutting down the
 784          * superblock, we need to actually destroy the kobject.
 785          */
 786         unlock_super(sb);
 787         kobject_put(&sbi->s_kobj);
 788         wait_for_completion(&sbi->s_kobj_unregister);
 789         kfree(sbi->s_blockgroup_lock);
 790         kfree(sbi);
 791 }
 792
 793 static struct kmem_cache *ext4_inode_cachep;
 794
 795 /*
 796  * Called inside transaction, so use GFP_NOFS
 797  */
 798 static struct inode *ext4_alloc_inode(struct super_block *sb)
 799 {
 800         struct ext4_inode_info *ei;
 801
 802         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 803         if (!ei)
 804                 return NULL;
 805
 806         ei->vfs_inode.i_version = 1;
 807         ei->vfs_inode.i_data.writeback_index = 0;
 808         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 809         INIT_LIST_HEAD(&ei->i_prealloc_list);
 810         spin_lock_init(&ei->i_prealloc_lock);
 811         /*
 812          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 813          * therefore it can be null here.  Don't check it, just initialize
 814          * jinode.
 815          */
 816         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 817         ei->i_reserved_data_blocks = 0;
 818         ei->i_reserved_meta_blocks = 0;
 819         ei->i_allocated_meta_blocks = 0;
 820         ei->i_da_metadata_calc_len = 0;
 821         ei->i_delalloc_reserved_flag = 0;
 822         spin_lock_init(&(ei->i_block_reservation_lock));
 823 #ifdef CONFIG_QUOTA
 824         ei->i_reserved_quota = 0;
 825 #endif
 826         INIT_LIST_HEAD(&ei->i_completed_io_list);
 827         spin_lock_init(&ei->i_completed_io_lock);
 828         ei->cur_aio_dio = NULL;
 829         ei->i_sync_tid = 0;
 830         ei->i_datasync_tid = 0;
 831         atomic_set(&ei->i_ioend_count, 0);
 832
 833         return &ei->vfs_inode;
 834 }
 835
 836 static int ext4_drop_inode(struct inode *inode)
 837 {
 838         int drop = generic_drop_inode(inode);
 839
 840         trace_ext4_drop_inode(inode, drop);
 841         return drop;
 842 }
 843
 844 static void ext4_destroy_inode(struct inode *inode)
 845 {
 846         ext4_ioend_wait(inode);
 847         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 848                 ext4_msg(inode->i_sb, KERN_ERR,
 849                          "Inode %lu (%p): orphan list check failed!",
 850                          inode->i_ino, EXT4_I(inode));
 851                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 852                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 853                                 true);
 854                 dump_stack();
 855         }
 856         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 857 }
 858
 859 static void init_once(void *foo)
 860 {
 861         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 862
 863         INIT_LIST_HEAD(&ei->i_orphan);
 864 #ifdef CONFIG_EXT4_FS_XATTR
 865         init_rwsem(&ei->xattr_sem);
 866 #endif
 867         init_rwsem(&ei->i_data_sem);
 868         inode_init_once(&ei->vfs_inode);
 869 }
 870
 871 static int init_inodecache(void)
 872 {
 873         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 874                                              sizeof(struct ext4_inode_info),
 875                                              0, (SLAB_RECLAIM_ACCOUNT|
 876                                                 SLAB_MEM_SPREAD),
 877                                              init_once);
 878         if (ext4_inode_cachep == NULL)
 879                 return -ENOMEM;
 880         return 0;
 881 }
 882
 883 static void destroy_inodecache(void)
 884 {
 885         kmem_cache_destroy(ext4_inode_cachep);
 886 }
 887
 888 void ext4_clear_inode(struct inode *inode)
 889 {
 890         invalidate_inode_buffers(inode);
 891         end_writeback(inode);
 892         dquot_drop(inode);
 893         ext4_discard_preallocations(inode);
 894         if (EXT4_JOURNAL(inode))
 895                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 896                                        &EXT4_I(inode)->jinode);
 897 }
 898
 899 static inline void ext4_show_quota_options(struct seq_file *seq,
 900                                            struct super_block *sb)
 901 {
 902 #if defined(CONFIG_QUOTA)
 903         struct ext4_sb_info *sbi = EXT4_SB(sb);
 904
 905         if (sbi->s_jquota_fmt) {
 906                 char *fmtname = "";
 907
 908                 switch (sbi->s_jquota_fmt) {
 909                 case QFMT_VFS_OLD:
 910                         fmtname = "vfsold";
 911                         break;
 912                 case QFMT_VFS_V0:
 913                         fmtname = "vfsv0";
 914                         break;
 915                 case QFMT_VFS_V1:
 916                         fmtname = "vfsv1";
 917                         break;
 918                 }
 919                 seq_printf(seq, ",jqfmt=%s", fmtname);
 920         }
 921
 922         if (sbi->s_qf_names[USRQUOTA])
 923                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 924
 925         if (sbi->s_qf_names[GRPQUOTA])
 926                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 927
 928         if (test_opt(sb, USRQUOTA))
 929                 seq_puts(seq, ",usrquota");
 930
 931         if (test_opt(sb, GRPQUOTA))
 932                 seq_puts(seq, ",grpquota");
 933 #endif
 934 }
 935
 936 /*
 937  * Show an option if
 938  *  - it's set to a non-default value OR
 939  *  - if the per-sb default is different from the global default
 940  */
 941 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 942 {
 943         int def_errors;
 944         unsigned long def_mount_opts;
 945         struct super_block *sb = vfs->mnt_sb;
 946         struct ext4_sb_info *sbi = EXT4_SB(sb);
 947         struct ext4_super_block *es = sbi->s_es;
 948
 949         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 950         def_errors     = le16_to_cpu(es->s_errors);
 951
 952         if (sbi->s_sb_block != 1)
 953                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 954         if (test_opt(sb, MINIX_DF))
 955                 seq_puts(seq, ",minixdf");
 956         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 957                 seq_puts(seq, ",grpid");
 958         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 959                 seq_puts(seq, ",nogrpid");
 960         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 961             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 962                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 963         }
 964         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 965             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 966                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 967         }
 968         if (test_opt(sb, ERRORS_RO)) {
 969                 if (def_errors == EXT4_ERRORS_PANIC ||
 970                     def_errors == EXT4_ERRORS_CONTINUE) {
 971                         seq_puts(seq, ",errors=remount-ro");
 972                 }
 973         }
 974         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 975                 seq_puts(seq, ",errors=continue");
 976         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 977                 seq_puts(seq, ",errors=panic");
 978         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 979                 seq_puts(seq, ",nouid32");
 980         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 981                 seq_puts(seq, ",debug");
 982         if (test_opt(sb, OLDALLOC))
 983                 seq_puts(seq, ",oldalloc");
 984 #ifdef CONFIG_EXT4_FS_XATTR
 985         if (test_opt(sb, XATTR_USER) &&
 986                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 987                 seq_puts(seq, ",user_xattr");
 988         if (!test_opt(sb, XATTR_USER) &&
 989             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 990                 seq_puts(seq, ",nouser_xattr");
 991         }
 992 #endif
 993 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 994         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 995                 seq_puts(seq, ",acl");
 996         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 997                 seq_puts(seq, ",noacl");
 998 #endif
 999         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1000                 seq_printf(seq, ",commit=%u",
1001                            (unsigned) (sbi->s_commit_interval / HZ));
1002         }
1003         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1004                 seq_printf(seq, ",min_batch_time=%u",
1005                            (unsigned) sbi->s_min_batch_time);
1006         }
1007         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1008                 seq_printf(seq, ",max_batch_time=%u",
1009                            (unsigned) sbi->s_min_batch_time);
1010         }
1011
1012         /*
1013          * We're changing the default of barrier mount option, so
1014          * let's always display its mount state so it's clear what its
1015          * status is.
1016          */
1017         seq_puts(seq, ",barrier=");
1018         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1019         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1020                 seq_puts(seq, ",journal_async_commit");
1021         else if (test_opt(sb, JOURNAL_CHECKSUM))
1022                 seq_puts(seq, ",journal_checksum");
1023         if (test_opt(sb, I_VERSION))
1024                 seq_puts(seq, ",i_version");
1025         if (!test_opt(sb, DELALLOC) &&
1026             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1027                 seq_puts(seq, ",nodelalloc");
1028
1029         if (sbi->s_stripe)
1030                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1031         /*
1032          * journal mode get enabled in different ways
1033          * So just print the value even if we didn't specify it
1034          */
1035         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1036                 seq_puts(seq, ",data=journal");
1037         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1038                 seq_puts(seq, ",data=ordered");
1039         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1040                 seq_puts(seq, ",data=writeback");
1041
1042         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1043                 seq_printf(seq, ",inode_readahead_blks=%u",
1044                            sbi->s_inode_readahead_blks);
1045
1046         if (test_opt(sb, DATA_ERR_ABORT))
1047                 seq_puts(seq, ",data_err=abort");
1048
1049         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1050                 seq_puts(seq, ",noauto_da_alloc");
1051
1052         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1053                 seq_puts(seq, ",discard");
1054
1055         if (test_opt(sb, NOLOAD))
1056                 seq_puts(seq, ",norecovery");
1057
1058         if (test_opt(sb, DIOREAD_NOLOCK))
1059                 seq_puts(seq, ",dioread_nolock");
1060
1061         if (test_opt(sb, BLOCK_VALIDITY) &&
1062             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1063                 seq_puts(seq, ",block_validity");
1064
1065         if (!test_opt(sb, INIT_INODE_TABLE))
1066                 seq_puts(seq, ",noinit_inode_table");
1067         else if (sbi->s_li_wait_mult)
1068                 seq_printf(seq, ",init_inode_table=%u",
1069                            (unsigned) sbi->s_li_wait_mult);
1070
1071         ext4_show_quota_options(seq, sb);
1072
1073         return 0;
1074 }
1075
1076 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1077                                         u64 ino, u32 generation)
1078 {
1079         struct inode *inode;
1080
1081         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1082                 return ERR_PTR(-ESTALE);
1083         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1084                 return ERR_PTR(-ESTALE);
1085
1086         /* iget isn't really right if the inode is currently unallocated!!
1087          *
1088          * ext4_read_inode will return a bad_inode if the inode had been
1089          * deleted, so we should be safe.
1090          *
1091          * Currently we don't know the generation for parent directory, so
1092          * a generation of 0 means "accept any"
1093          */
1094         inode = ext4_iget(sb, ino);
1095         if (IS_ERR(inode))
1096                 return ERR_CAST(inode);
1097         if (generation && inode->i_generation != generation) {
1098                 iput(inode);
1099                 return ERR_PTR(-ESTALE);
1100         }
1101
1102         return inode;
1103 }
1104
1105 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1106                                         int fh_len, int fh_type)
1107 {
1108         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1109                                     ext4_nfs_get_inode);
1110 }
1111
1112 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1113                                         int fh_len, int fh_type)
1114 {
1115         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1116                                     ext4_nfs_get_inode);
1117 }
1118
1119 /*
1120  * Try to release metadata pages (indirect blocks, directories) which are
1121  * mapped via the block device.  Since these pages could have journal heads
1122  * which would prevent try_to_free_buffers() from freeing them, we must use
1123  * jbd2 layer's try_to_free_buffers() function to release them.
1124  */
1125 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1126                                  gfp_t wait)
1127 {
1128         journal_t *journal = EXT4_SB(sb)->s_journal;
1129
1130         WARN_ON(PageChecked(page));
1131         if (!page_has_buffers(page))
1132                 return 0;
1133         if (journal)
1134                 return jbd2_journal_try_to_free_buffers(journal, page,
1135                                                         wait & ~__GFP_WAIT);
1136         return try_to_free_buffers(page);
1137 }
1138
1139 #ifdef CONFIG_QUOTA
1140 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1141 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1142
1143 static int ext4_write_dquot(struct dquot *dquot);
1144 static int ext4_acquire_dquot(struct dquot *dquot);
1145 static int ext4_release_dquot(struct dquot *dquot);
1146 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1147 static int ext4_write_info(struct super_block *sb, int type);
1148 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1149                                 char *path);
1150 static int ext4_quota_off(struct super_block *sb, int type);
1151 static int ext4_quota_on_mount(struct super_block *sb, int type);
1152 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1153                                size_t len, loff_t off);
1154 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1155                                 const char *data, size_t len, loff_t off);
1156
1157 static const struct dquot_operations ext4_quota_operations = {
1158 #ifdef CONFIG_QUOTA
1159         .get_reserved_space = ext4_get_reserved_space,
1160 #endif
1161         .write_dquot    = ext4_write_dquot,
1162         .acquire_dquot  = ext4_acquire_dquot,
1163         .release_dquot  = ext4_release_dquot,
1164         .mark_dirty     = ext4_mark_dquot_dirty,
1165         .write_info     = ext4_write_info,
1166         .alloc_dquot    = dquot_alloc,
1167         .destroy_dquot  = dquot_destroy,
1168 };
1169
1170 static const struct quotactl_ops ext4_qctl_operations = {
1171         .quota_on       = ext4_quota_on,
1172         .quota_off      = ext4_quota_off,
1173         .quota_sync     = dquot_quota_sync,
1174         .get_info       = dquot_get_dqinfo,
1175         .set_info       = dquot_set_dqinfo,
1176         .get_dqblk      = dquot_get_dqblk,
1177         .set_dqblk      = dquot_set_dqblk
1178 };
1179 #endif
1180
1181 static const struct super_operations ext4_sops = {
1182         .alloc_inode    = ext4_alloc_inode,
1183         .destroy_inode  = ext4_destroy_inode,
1184         .write_inode    = ext4_write_inode,
1185         .dirty_inode    = ext4_dirty_inode,
1186         .drop_inode     = ext4_drop_inode,
1187         .evict_inode    = ext4_evict_inode,
1188         .put_super      = ext4_put_super,
1189         .sync_fs        = ext4_sync_fs,
1190         .freeze_fs      = ext4_freeze,
1191         .unfreeze_fs    = ext4_unfreeze,
1192         .statfs         = ext4_statfs,
1193         .remount_fs     = ext4_remount,
1194         .show_options   = ext4_show_options,
1195 #ifdef CONFIG_QUOTA
1196         .quota_read     = ext4_quota_read,
1197         .quota_write    = ext4_quota_write,
1198 #endif
1199         .bdev_try_to_free_page = bdev_try_to_free_page,
1200         .trim_fs        = ext4_trim_fs
1201 };
1202
1203 static const struct super_operations ext4_nojournal_sops = {
1204         .alloc_inode    = ext4_alloc_inode,
1205         .destroy_inode  = ext4_destroy_inode,
1206         .write_inode    = ext4_write_inode,
1207         .dirty_inode    = ext4_dirty_inode,
1208         .drop_inode     = ext4_drop_inode,
1209         .evict_inode    = ext4_evict_inode,
1210         .write_super    = ext4_write_super,
1211         .put_super      = ext4_put_super,
1212         .statfs         = ext4_statfs,
1213         .remount_fs     = ext4_remount,
1214         .show_options   = ext4_show_options,
1215 #ifdef CONFIG_QUOTA
1216         .quota_read     = ext4_quota_read,
1217         .quota_write    = ext4_quota_write,
1218 #endif
1219         .bdev_try_to_free_page = bdev_try_to_free_page,
1220 };
1221
1222 static const struct export_operations ext4_export_ops = {
1223         .fh_to_dentry = ext4_fh_to_dentry,
1224         .fh_to_parent = ext4_fh_to_parent,
1225         .get_parent = ext4_get_parent,
1226 };
1227
1228 enum {
1229         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1230         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1231         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1232         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1233         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1234         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1235         Opt_journal_update, Opt_journal_dev,
1236         Opt_journal_checksum, Opt_journal_async_commit,
1237         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1238         Opt_data_err_abort, Opt_data_err_ignore,
1239         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1240         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1241         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1242         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1243         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1244         Opt_block_validity, Opt_noblock_validity,
1245         Opt_inode_readahead_blks, Opt_journal_ioprio,
1246         Opt_dioread_nolock, Opt_dioread_lock,
1247         Opt_discard, Opt_nodiscard,
1248         Opt_init_inode_table, Opt_noinit_inode_table,
1249 };
1250
1251 static const match_table_t tokens = {
1252         {Opt_bsd_df, "bsddf"},
1253         {Opt_minix_df, "minixdf"},
1254         {Opt_grpid, "grpid"},
1255         {Opt_grpid, "bsdgroups"},
1256         {Opt_nogrpid, "nogrpid"},
1257         {Opt_nogrpid, "sysvgroups"},
1258         {Opt_resgid, "resgid=%u"},
1259         {Opt_resuid, "resuid=%u"},
1260         {Opt_sb, "sb=%u"},
1261         {Opt_err_cont, "errors=continue"},
1262         {Opt_err_panic, "errors=panic"},
1263         {Opt_err_ro, "errors=remount-ro"},
1264         {Opt_nouid32, "nouid32"},
1265         {Opt_debug, "debug"},
1266         {Opt_oldalloc, "oldalloc"},
1267         {Opt_orlov, "orlov"},
1268         {Opt_user_xattr, "user_xattr"},
1269         {Opt_nouser_xattr, "nouser_xattr"},
1270         {Opt_acl, "acl"},
1271         {Opt_noacl, "noacl"},
1272         {Opt_noload, "noload"},
1273         {Opt_noload, "norecovery"},
1274         {Opt_nobh, "nobh"},
1275         {Opt_bh, "bh"},
1276         {Opt_commit, "commit=%u"},
1277         {Opt_min_batch_time, "min_batch_time=%u"},
1278         {Opt_max_batch_time, "max_batch_time=%u"},
1279         {Opt_journal_update, "journal=update"},
1280         {Opt_journal_dev, "journal_dev=%u"},
1281         {Opt_journal_checksum, "journal_checksum"},
1282         {Opt_journal_async_commit, "journal_async_commit"},
1283         {Opt_abort, "abort"},
1284         {Opt_data_journal, "data=journal"},
1285         {Opt_data_ordered, "data=ordered"},
1286         {Opt_data_writeback, "data=writeback"},
1287         {Opt_data_err_abort, "data_err=abort"},
1288         {Opt_data_err_ignore, "data_err=ignore"},
1289         {Opt_offusrjquota, "usrjquota="},
1290         {Opt_usrjquota, "usrjquota=%s"},
1291         {Opt_offgrpjquota, "grpjquota="},
1292         {Opt_grpjquota, "grpjquota=%s"},
1293         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1294         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1295         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1296         {Opt_grpquota, "grpquota"},
1297         {Opt_noquota, "noquota"},
1298         {Opt_quota, "quota"},
1299         {Opt_usrquota, "usrquota"},
1300         {Opt_barrier, "barrier=%u"},
1301         {Opt_barrier, "barrier"},
1302         {Opt_nobarrier, "nobarrier"},
1303         {Opt_i_version, "i_version"},
1304         {Opt_stripe, "stripe=%u"},
1305         {Opt_resize, "resize"},
1306         {Opt_delalloc, "delalloc"},
1307         {Opt_nodelalloc, "nodelalloc"},
1308         {Opt_block_validity, "block_validity"},
1309         {Opt_noblock_validity, "noblock_validity"},
1310         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1311         {Opt_journal_ioprio, "journal_ioprio=%u"},
1312         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1313         {Opt_auto_da_alloc, "auto_da_alloc"},
1314         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1315         {Opt_dioread_nolock, "dioread_nolock"},
1316         {Opt_dioread_lock, "dioread_lock"},
1317         {Opt_discard, "discard"},
1318         {Opt_nodiscard, "nodiscard"},
1319         {Opt_init_inode_table, "init_itable=%u"},
1320         {Opt_init_inode_table, "init_itable"},
1321         {Opt_noinit_inode_table, "noinit_itable"},
1322         {Opt_err, NULL},
1323 };
1324
1325 static ext4_fsblk_t get_sb_block(void **data)
1326 {
1327         ext4_fsblk_t    sb_block;
1328         char            *options = (char *) *data;
1329
1330         if (!options || strncmp(options, "sb=", 3) != 0)
1331                 return 1;       /* Default location */
1332
1333         options += 3;
1334         /* TODO: use simple_strtoll with >32bit ext4 */
1335         sb_block = simple_strtoul(options, &options, 0);
1336         if (*options && *options != ',') {
1337                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1338                        (char *) *data);
1339                 return 1;
1340         }
1341         if (*options == ',')
1342                 options++;
1343         *data = (void *) options;
1344
1345         return sb_block;
1346 }
1347
1348 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1349 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1350         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1351
1352 #ifdef CONFIG_QUOTA
1353 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1354 {
1355         struct ext4_sb_info *sbi = EXT4_SB(sb);
1356         char *qname;
1357
1358         if (sb_any_quota_loaded(sb) &&
1359                 !sbi->s_qf_names[qtype]) {
1360                 ext4_msg(sb, KERN_ERR,
1361                         "Cannot change journaled "
1362                         "quota options when quota turned on");
1363                 return 0;
1364         }
1365         qname = match_strdup(args);
1366         if (!qname) {
1367                 ext4_msg(sb, KERN_ERR,
1368                         "Not enough memory for storing quotafile name");
1369                 return 0;
1370         }
1371         if (sbi->s_qf_names[qtype] &&
1372                 strcmp(sbi->s_qf_names[qtype], qname)) {
1373                 ext4_msg(sb, KERN_ERR,
1374                         "%s quota file already specified", QTYPE2NAME(qtype));
1375                 kfree(qname);
1376                 return 0;
1377         }
1378         sbi->s_qf_names[qtype] = qname;
1379         if (strchr(sbi->s_qf_names[qtype], '/')) {
1380                 ext4_msg(sb, KERN_ERR,
1381                         "quotafile must be on filesystem root");
1382                 kfree(sbi->s_qf_names[qtype]);
1383                 sbi->s_qf_names[qtype] = NULL;
1384                 return 0;
1385         }
1386         set_opt(sbi->s_mount_opt, QUOTA);
1387         return 1;
1388 }
1389
1390 static int clear_qf_name(struct super_block *sb, int qtype)
1391 {
1392
1393         struct ext4_sb_info *sbi = EXT4_SB(sb);
1394
1395         if (sb_any_quota_loaded(sb) &&
1396                 sbi->s_qf_names[qtype]) {
1397                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1398                         " when quota turned on");
1399                 return 0;
1400         }
1401         /*
1402          * The space will be released later when all options are confirmed
1403          * to be correct
1404          */
1405         sbi->s_qf_names[qtype] = NULL;
1406         return 1;
1407 }
1408 #endif
1409
1410 static int parse_options(char *options, struct super_block *sb,
1411                          unsigned long *journal_devnum,
1412                          unsigned int *journal_ioprio,
1413                          ext4_fsblk_t *n_blocks_count, int is_remount)
1414 {
1415         struct ext4_sb_info *sbi = EXT4_SB(sb);
1416         char *p;
1417         substring_t args[MAX_OPT_ARGS];
1418         int data_opt = 0;
1419         int option;
1420 #ifdef CONFIG_QUOTA
1421         int qfmt;
1422 #endif
1423
1424         if (!options)
1425                 return 1;
1426
1427         while ((p = strsep(&options, ",")) != NULL) {
1428                 int token;
1429                 if (!*p)
1430                         continue;
1431
1432                 /*
1433                  * Initialize args struct so we know whether arg was
1434                  * found; some options take optional arguments.
1435                  */
1436                 args[0].to = args[0].from = 0;
1437                 token = match_token(p, tokens, args);
1438                 switch (token) {
1439                 case Opt_bsd_df:
1440                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1441                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1442                         break;
1443                 case Opt_minix_df:
1444                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1445                         set_opt(sbi->s_mount_opt, MINIX_DF);
1446
1447                         break;
1448                 case Opt_grpid:
1449                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1450                         set_opt(sbi->s_mount_opt, GRPID);
1451
1452                         break;
1453                 case Opt_nogrpid:
1454                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1455                         clear_opt(sbi->s_mount_opt, GRPID);
1456
1457                         break;
1458                 case Opt_resuid:
1459                         if (match_int(&args[0], &option))
1460                                 return 0;
1461                         sbi->s_resuid = option;
1462                         break;
1463                 case Opt_resgid:
1464                         if (match_int(&args[0], &option))
1465                                 return 0;
1466                         sbi->s_resgid = option;
1467                         break;
1468                 case Opt_sb:
1469                         /* handled by get_sb_block() instead of here */
1470                         /* *sb_block = match_int(&args[0]); */
1471                         break;
1472                 case Opt_err_panic:
1473                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1474                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1475                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1476                         break;
1477                 case Opt_err_ro:
1478                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1479                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1480                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1481                         break;
1482                 case Opt_err_cont:
1483                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1484                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1485                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1486                         break;
1487                 case Opt_nouid32:
1488                         set_opt(sbi->s_mount_opt, NO_UID32);
1489                         break;
1490                 case Opt_debug:
1491                         set_opt(sbi->s_mount_opt, DEBUG);
1492                         break;
1493                 case Opt_oldalloc:
1494                         set_opt(sbi->s_mount_opt, OLDALLOC);
1495                         break;
1496                 case Opt_orlov:
1497                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1498                         break;
1499 #ifdef CONFIG_EXT4_FS_XATTR
1500                 case Opt_user_xattr:
1501                         set_opt(sbi->s_mount_opt, XATTR_USER);
1502                         break;
1503                 case Opt_nouser_xattr:
1504                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1505                         break;
1506 #else
1507                 case Opt_user_xattr:
1508                 case Opt_nouser_xattr:
1509                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1510                         break;
1511 #endif
1512 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1513                 case Opt_acl:
1514                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1515                         break;
1516                 case Opt_noacl:
1517                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1518                         break;
1519 #else
1520                 case Opt_acl:
1521                 case Opt_noacl:
1522                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1523                         break;
1524 #endif
1525                 case Opt_journal_update:
1526                         /* @@@ FIXME */
1527                         /* Eventually we will want to be able to create
1528                            a journal file here.  For now, only allow the
1529                            user to specify an existing inode to be the
1530                            journal file. */
1531                         if (is_remount) {
1532                                 ext4_msg(sb, KERN_ERR,
1533                                          "Cannot specify journal on remount");
1534                                 return 0;
1535                         }
1536                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1537                         break;
1538                 case Opt_journal_dev:
1539                         if (is_remount) {
1540                                 ext4_msg(sb, KERN_ERR,
1541                                         "Cannot specify journal on remount");
1542                                 return 0;
1543                         }
1544                         if (match_int(&args[0], &option))
1545                                 return 0;
1546                         *journal_devnum = option;
1547                         break;
1548                 case Opt_journal_checksum:
1549                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1550                         break;
1551                 case Opt_journal_async_commit:
1552                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1553                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1554                         break;
1555                 case Opt_noload:
1556                         set_opt(sbi->s_mount_opt, NOLOAD);
1557                         break;
1558                 case Opt_commit:
1559                         if (match_int(&args[0], &option))
1560                                 return 0;
1561                         if (option < 0)
1562                                 return 0;
1563                         if (option == 0)
1564                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1565                         sbi->s_commit_interval = HZ * option;
1566                         break;
1567                 case Opt_max_batch_time:
1568                         if (match_int(&args[0], &option))
1569                                 return 0;
1570                         if (option < 0)
1571                                 return 0;
1572                         if (option == 0)
1573                                 option = EXT4_DEF_MAX_BATCH_TIME;
1574                         sbi->s_max_batch_time = option;
1575                         break;
1576                 case Opt_min_batch_time:
1577                         if (match_int(&args[0], &option))
1578                                 return 0;
1579                         if (option < 0)
1580                                 return 0;
1581                         sbi->s_min_batch_time = option;
1582                         break;
1583                 case Opt_data_journal:
1584                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1585                         goto datacheck;
1586                 case Opt_data_ordered:
1587                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1588                         goto datacheck;
1589                 case Opt_data_writeback:
1590                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1591                 datacheck:
1592                         if (is_remount) {
1593                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1594                                         ext4_msg(sb, KERN_ERR,
1595                                                 "Cannot change data mode on remount");
1596                                         return 0;
1597                                 }
1598                         } else {
1599                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1600                                 sbi->s_mount_opt |= data_opt;
1601                         }
1602                         break;
1603                 case Opt_data_err_abort:
1604                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1605                         break;
1606                 case Opt_data_err_ignore:
1607                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1608                         break;
1609 #ifdef CONFIG_QUOTA
1610                 case Opt_usrjquota:
1611                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1612                                 return 0;
1613                         break;
1614                 case Opt_grpjquota:
1615                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1616                                 return 0;
1617                         break;
1618                 case Opt_offusrjquota:
1619                         if (!clear_qf_name(sb, USRQUOTA))
1620                                 return 0;
1621                         break;
1622                 case Opt_offgrpjquota:
1623                         if (!clear_qf_name(sb, GRPQUOTA))
1624                                 return 0;
1625                         break;
1626
1627                 case Opt_jqfmt_vfsold:
1628                         qfmt = QFMT_VFS_OLD;
1629                         goto set_qf_format;
1630                 case Opt_jqfmt_vfsv0:
1631                         qfmt = QFMT_VFS_V0;
1632                         goto set_qf_format;
1633                 case Opt_jqfmt_vfsv1:
1634                         qfmt = QFMT_VFS_V1;
1635 set_qf_format:
1636                         if (sb_any_quota_loaded(sb) &&
1637                             sbi->s_jquota_fmt != qfmt) {
1638                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1639                                         "journaled quota options when "
1640                                         "quota turned on");
1641                                 return 0;
1642                         }
1643                         sbi->s_jquota_fmt = qfmt;
1644                         break;
1645                 case Opt_quota:
1646                 case Opt_usrquota:
1647                         set_opt(sbi->s_mount_opt, QUOTA);
1648                         set_opt(sbi->s_mount_opt, USRQUOTA);
1649                         break;
1650                 case Opt_grpquota:
1651                         set_opt(sbi->s_mount_opt, QUOTA);
1652                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1653                         break;
1654                 case Opt_noquota:
1655                         if (sb_any_quota_loaded(sb)) {
1656                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1657                                         "options when quota turned on");
1658                                 return 0;
1659                         }
1660                         clear_opt(sbi->s_mount_opt, QUOTA);
1661                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1662                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1663                         break;
1664 #else
1665                 case Opt_quota:
1666                 case Opt_usrquota:
1667                 case Opt_grpquota:
1668                         ext4_msg(sb, KERN_ERR,
1669                                 "quota options not supported");
1670                         break;
1671                 case Opt_usrjquota:
1672                 case Opt_grpjquota:
1673                 case Opt_offusrjquota:
1674                 case Opt_offgrpjquota:
1675                 case Opt_jqfmt_vfsold:
1676                 case Opt_jqfmt_vfsv0:
1677                 case Opt_jqfmt_vfsv1:
1678                         ext4_msg(sb, KERN_ERR,
1679                                 "journaled quota options not supported");
1680                         break;
1681                 case Opt_noquota:
1682                         break;
1683 #endif
1684                 case Opt_abort:
1685                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1686                         break;
1687                 case Opt_nobarrier:
1688                         clear_opt(sbi->s_mount_opt, BARRIER);
1689                         break;
1690                 case Opt_barrier:
1691                         if (args[0].from) {
1692                                 if (match_int(&args[0], &option))
1693                                         return 0;
1694                         } else
1695                                 option = 1;     /* No argument, default to 1 */
1696                         if (option)
1697                                 set_opt(sbi->s_mount_opt, BARRIER);
1698                         else
1699                                 clear_opt(sbi->s_mount_opt, BARRIER);
1700                         break;
1701                 case Opt_ignore:
1702                         break;
1703                 case Opt_resize:
1704                         if (!is_remount) {
1705                                 ext4_msg(sb, KERN_ERR,
1706                                         "resize option only available "
1707                                         "for remount");
1708                                 return 0;
1709                         }
1710                         if (match_int(&args[0], &option) != 0)
1711                                 return 0;
1712                         *n_blocks_count = option;
1713                         break;
1714                 case Opt_nobh:
1715                         ext4_msg(sb, KERN_WARNING,
1716                                  "Ignoring deprecated nobh option");
1717                         break;
1718                 case Opt_bh:
1719                         ext4_msg(sb, KERN_WARNING,
1720                                  "Ignoring deprecated bh option");
1721                         break;
1722                 case Opt_i_version:
1723                         set_opt(sbi->s_mount_opt, I_VERSION);
1724                         sb->s_flags |= MS_I_VERSION;
1725                         break;
1726                 case Opt_nodelalloc:
1727                         clear_opt(sbi->s_mount_opt, DELALLOC);
1728                         break;
1729                 case Opt_stripe:
1730                         if (match_int(&args[0], &option))
1731                                 return 0;
1732                         if (option < 0)
1733                                 return 0;
1734                         sbi->s_stripe = option;
1735                         break;
1736                 case Opt_delalloc:
1737                         set_opt(sbi->s_mount_opt, DELALLOC);
1738                         break;
1739                 case Opt_block_validity:
1740                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1741                         break;
1742                 case Opt_noblock_validity:
1743                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1744                         break;
1745                 case Opt_inode_readahead_blks:
1746                         if (match_int(&args[0], &option))
1747                                 return 0;
1748                         if (option < 0 || option > (1 << 30))
1749                                 return 0;
1750                         if (!is_power_of_2(option)) {
1751                                 ext4_msg(sb, KERN_ERR,
1752                                          "EXT4-fs: inode_readahead_blks"
1753                                          " must be a power of 2");
1754                                 return 0;
1755                         }
1756                         sbi->s_inode_readahead_blks = option;
1757                         break;
1758                 case Opt_journal_ioprio:
1759                         if (match_int(&args[0], &option))
1760                                 return 0;
1761                         if (option < 0 || option > 7)
1762                                 break;
1763                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1764                                                             option);
1765                         break;
1766                 case Opt_noauto_da_alloc:
1767                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1768                         break;
1769                 case Opt_auto_da_alloc:
1770                         if (args[0].from) {
1771                                 if (match_int(&args[0], &option))
1772                                         return 0;
1773                         } else
1774                                 option = 1;     /* No argument, default to 1 */
1775                         if (option)
1776                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1777                         else
1778                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1779                         break;
1780                 case Opt_discard:
1781                         set_opt(sbi->s_mount_opt, DISCARD);
1782                         break;
1783                 case Opt_nodiscard:
1784                         clear_opt(sbi->s_mount_opt, DISCARD);
1785                         break;
1786                 case Opt_dioread_nolock:
1787                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1788                         break;
1789                 case Opt_dioread_lock:
1790                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1791                         break;
1792                 case Opt_init_inode_table:
1793                         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1794                         if (args[0].from) {
1795                                 if (match_int(&args[0], &option))
1796                                         return 0;
1797                         } else
1798                                 option = EXT4_DEF_LI_WAIT_MULT;
1799                         if (option < 0)
1800                                 return 0;
1801                         sbi->s_li_wait_mult = option;
1802                         break;
1803                 case Opt_noinit_inode_table:
1804                         clear_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
1805                         break;
1806                 default:
1807                         ext4_msg(sb, KERN_ERR,
1808                                "Unrecognized mount option \"%s\" "
1809                                "or missing value", p);
1810                         return 0;
1811                 }
1812         }
1813 #ifdef CONFIG_QUOTA
1814         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1815                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1816                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1817
1818                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1819                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1820
1821                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1822                         ext4_msg(sb, KERN_ERR, "old and new quota "
1823                                         "format mixing");
1824                         return 0;
1825                 }
1826
1827                 if (!sbi->s_jquota_fmt) {
1828                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1829                                         "not specified");
1830                         return 0;
1831                 }
1832         } else {
1833                 if (sbi->s_jquota_fmt) {
1834                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1835                                         "specified with no journaling "
1836                                         "enabled");
1837                         return 0;
1838                 }
1839         }
1840 #endif
1841         return 1;
1842 }
1843
1844 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1845                             int read_only)
1846 {
1847         struct ext4_sb_info *sbi = EXT4_SB(sb);
1848         int res = 0;
1849
1850         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1851                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1852                          "forcing read-only mode");
1853                 res = MS_RDONLY;
1854         }
1855         if (read_only)
1856                 return res;
1857         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1858                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1859                          "running e2fsck is recommended");
1860         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1861                 ext4_msg(sb, KERN_WARNING,
1862                          "warning: mounting fs with errors, "
1863                          "running e2fsck is recommended");
1864         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1865                  le16_to_cpu(es->s_mnt_count) >=
1866                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1867                 ext4_msg(sb, KERN_WARNING,
1868                          "warning: maximal mount count reached, "
1869                          "running e2fsck is recommended");
1870         else if (le32_to_cpu(es->s_checkinterval) &&
1871                 (le32_to_cpu(es->s_lastcheck) +
1872                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1873                 ext4_msg(sb, KERN_WARNING,
1874                          "warning: checktime reached, "
1875                          "running e2fsck is recommended");
1876         if (!sbi->s_journal)
1877                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1878         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1879                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1880         le16_add_cpu(&es->s_mnt_count, 1);
1881         es->s_mtime = cpu_to_le32(get_seconds());
1882         ext4_update_dynamic_rev(sb);
1883         if (sbi->s_journal)
1884                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1885
1886         ext4_commit_super(sb, 1);
1887         if (test_opt(sb, DEBUG))
1888                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1889                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1890                         sb->s_blocksize,
1891                         sbi->s_groups_count,
1892                         EXT4_BLOCKS_PER_GROUP(sb),
1893                         EXT4_INODES_PER_GROUP(sb),
1894                         sbi->s_mount_opt);
1895
1896         return res;
1897 }
1898
1899 static int ext4_fill_flex_info(struct super_block *sb)
1900 {
1901         struct ext4_sb_info *sbi = EXT4_SB(sb);
1902         struct ext4_group_desc *gdp = NULL;
1903         ext4_group_t flex_group_count;
1904         ext4_group_t flex_group;
1905         int groups_per_flex = 0;
1906         size_t size;
1907         int i;
1908
1909         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1910         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1911
1912         if (groups_per_flex < 2) {
1913                 sbi->s_log_groups_per_flex = 0;
1914                 return 1;
1915         }
1916
1917         /* We allocate both existing and potentially added groups */
1918         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1919                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1920                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1921         size = flex_group_count * sizeof(struct flex_groups);
1922         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1923         if (sbi->s_flex_groups == NULL) {
1924                 sbi->s_flex_groups = vmalloc(size);
1925                 if (sbi->s_flex_groups)
1926                         memset(sbi->s_flex_groups, 0, size);
1927         }
1928         if (sbi->s_flex_groups == NULL) {
1929                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1930                                 "%u flex groups", flex_group_count);
1931                 goto failed;
1932         }
1933
1934         for (i = 0; i < sbi->s_groups_count; i++) {
1935                 gdp = ext4_get_group_desc(sb, i, NULL);
1936
1937                 flex_group = ext4_flex_group(sbi, i);
1938                 atomic_add(ext4_free_inodes_count(sb, gdp),
1939                            &sbi->s_flex_groups[flex_group].free_inodes);
1940                 atomic_add(ext4_free_blks_count(sb, gdp),
1941                            &sbi->s_flex_groups[flex_group].free_blocks);
1942                 atomic_add(ext4_used_dirs_count(sb, gdp),
1943                            &sbi->s_flex_groups[flex_group].used_dirs);
1944         }
1945
1946         return 1;
1947 failed:
1948         return 0;
1949 }
1950
1951 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1952                             struct ext4_group_desc *gdp)
1953 {
1954         __u16 crc = 0;
1955
1956         if (sbi->s_es->s_feature_ro_compat &
1957             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1958                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1959                 __le32 le_group = cpu_to_le32(block_group);
1960
1961                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1962                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1963                 crc = crc16(crc, (__u8 *)gdp, offset);
1964                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1965                 /* for checksum of struct ext4_group_desc do the rest...*/
1966                 if ((sbi->s_es->s_feature_incompat &
1967                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1968                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1969                         crc = crc16(crc, (__u8 *)gdp + offset,
1970                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1971                                         offset);
1972         }
1973
1974         return cpu_to_le16(crc);
1975 }
1976
1977 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1978                                 struct ext4_group_desc *gdp)
1979 {
1980         if ((sbi->s_es->s_feature_ro_compat &
1981              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1982             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1983                 return 0;
1984
1985         return 1;
1986 }
1987
1988 /* Called at mount-time, super-block is locked */
1989 static int ext4_check_descriptors(struct super_block *sb,
1990                                   ext4_group_t *first_not_zeroed)
1991 {
1992         struct ext4_sb_info *sbi = EXT4_SB(sb);
1993         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1994         ext4_fsblk_t last_block;
1995         ext4_fsblk_t block_bitmap;
1996         ext4_fsblk_t inode_bitmap;
1997         ext4_fsblk_t inode_table;
1998         int flexbg_flag = 0;
1999         ext4_group_t i, grp = sbi->s_groups_count;
2000
2001         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2002                 flexbg_flag = 1;
2003
2004         ext4_debug("Checking group descriptors");
2005
2006         for (i = 0; i < sbi->s_groups_count; i++) {
2007                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2008
2009                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2010                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2011                 else
2012                         last_block = first_block +
2013                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2014
2015                 if ((grp == sbi->s_groups_count) &&
2016                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2017                         grp = i;
2018
2019                 block_bitmap = ext4_block_bitmap(sb, gdp);
2020                 if (block_bitmap < first_block || block_bitmap > last_block) {
2021                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2022                                "Block bitmap for group %u not in group "
2023                                "(block %llu)!", i, block_bitmap);
2024                         return 0;
2025                 }
2026                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2027                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2028                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2029                                "Inode bitmap for group %u not in group "
2030                                "(block %llu)!", i, inode_bitmap);
2031                         return 0;
2032                 }
2033                 inode_table = ext4_inode_table(sb, gdp);
2034                 if (inode_table < first_block ||
2035                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2036                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2037                                "Inode table for group %u not in group "
2038                                "(block %llu)!", i, inode_table);
2039                         return 0;
2040                 }
2041                 ext4_lock_group(sb, i);
2042                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2043                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2044                                  "Checksum for group %u failed (%u!=%u)",
2045                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2046                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2047                         if (!(sb->s_flags & MS_RDONLY)) {
2048                                 ext4_unlock_group(sb, i);
2049                                 return 0;
2050                         }
2051                 }
2052                 ext4_unlock_group(sb, i);
2053                 if (!flexbg_flag)
2054                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2055         }
2056         if (NULL != first_not_zeroed)
2057                 *first_not_zeroed = grp;
2058
2059         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2060         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2061         return 1;
2062 }
2063
2064 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2065  * the superblock) which were deleted from all directories, but held open by
2066  * a process at the time of a crash.  We walk the list and try to delete these
2067  * inodes at recovery time (only with a read-write filesystem).
2068  *
2069  * In order to keep the orphan inode chain consistent during traversal (in
2070  * case of crash during recovery), we link each inode into the superblock
2071  * orphan list_head and handle it the same way as an inode deletion during
2072  * normal operation (which journals the operations for us).
2073  *
2074  * We only do an iget() and an iput() on each inode, which is very safe if we
2075  * accidentally point at an in-use or already deleted inode.  The worst that
2076  * can happen in this case is that we get a "bit already cleared" message from
2077  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2078  * e2fsck was run on this filesystem, and it must have already done the orphan
2079  * inode cleanup for us, so we can safely abort without any further action.
2080  */
2081 static void ext4_orphan_cleanup(struct super_block *sb,
2082                                 struct ext4_super_block *es)
2083 {
2084         unsigned int s_flags = sb->s_flags;
2085         int nr_orphans = 0, nr_truncates = 0;
2086 #ifdef CONFIG_QUOTA
2087         int i;
2088 #endif
2089         if (!es->s_last_orphan) {
2090                 jbd_debug(4, "no orphan inodes to clean up\n");
2091                 return;
2092         }
2093
2094         if (bdev_read_only(sb->s_bdev)) {
2095                 ext4_msg(sb, KERN_ERR, "write access "
2096                         "unavailable, skipping orphan cleanup");
2097                 return;
2098         }
2099
2100         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2101                 if (es->s_last_orphan)
2102                         jbd_debug(1, "Errors on filesystem, "
2103                                   "clearing orphan list.\n");
2104                 es->s_last_orphan = 0;
2105                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2106                 return;
2107         }
2108
2109         if (s_flags & MS_RDONLY) {
2110                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2111                 sb->s_flags &= ~MS_RDONLY;
2112         }
2113 #ifdef CONFIG_QUOTA
2114         /* Needed for iput() to work correctly and not trash data */
2115         sb->s_flags |= MS_ACTIVE;
2116         /* Turn on quotas so that they are updated correctly */
2117         for (i = 0; i < MAXQUOTAS; i++) {
2118                 if (EXT4_SB(sb)->s_qf_names[i]) {
2119                         int ret = ext4_quota_on_mount(sb, i);
2120                         if (ret < 0)
2121                                 ext4_msg(sb, KERN_ERR,
2122                                         "Cannot turn on journaled "
2123                                         "quota: error %d", ret);
2124                 }
2125         }
2126 #endif
2127
2128         while (es->s_last_orphan) {
2129                 struct inode *inode;
2130
2131                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2132                 if (IS_ERR(inode)) {
2133                         es->s_last_orphan = 0;
2134                         break;
2135                 }
2136
2137                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2138                 dquot_initialize(inode);
2139                 if (inode->i_nlink) {
2140                         ext4_msg(sb, KERN_DEBUG,
2141                                 "%s: truncating inode %lu to %lld bytes",
2142                                 __func__, inode->i_ino, inode->i_size);
2143                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2144                                   inode->i_ino, inode->i_size);
2145                         ext4_truncate(inode);
2146                         nr_truncates++;
2147                 } else {
2148                         ext4_msg(sb, KERN_DEBUG,
2149                                 "%s: deleting unreferenced inode %lu",
2150                                 __func__, inode->i_ino);
2151                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2152                                   inode->i_ino);
2153                         nr_orphans++;
2154                 }
2155                 iput(inode);  /* The delete magic happens here! */
2156         }
2157
2158 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2159
2160         if (nr_orphans)
2161                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2162                        PLURAL(nr_orphans));
2163         if (nr_truncates)
2164                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2165                        PLURAL(nr_truncates));
2166 #ifdef CONFIG_QUOTA
2167         /* Turn quotas off */
2168         for (i = 0; i < MAXQUOTAS; i++) {
2169                 if (sb_dqopt(sb)->files[i])
2170                         dquot_quota_off(sb, i);
2171         }
2172 #endif
2173         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2174 }
2175
2176 /*
2177  * Maximal extent format file size.
2178  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2179  * extent format containers, within a sector_t, and within i_blocks
2180  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2181  * so that won't be a limiting factor.
2182  *
2183  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2184  */
2185 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2186 {
2187         loff_t res;
2188         loff_t upper_limit = MAX_LFS_FILESIZE;
2189
2190         /* small i_blocks in vfs inode? */
2191         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2192                 /*
2193                  * CONFIG_LBDAF is not enabled implies the inode
2194                  * i_block represent total blocks in 512 bytes
2195                  * 32 == size of vfs inode i_blocks * 8
2196                  */
2197                 upper_limit = (1LL << 32) - 1;
2198
2199                 /* total blocks in file system block size */
2200                 upper_limit >>= (blkbits - 9);
2201                 upper_limit <<= blkbits;
2202         }
2203
2204         /* 32-bit extent-start container, ee_block */
2205         res = 1LL << 32;
2206         res <<= blkbits;
2207         res -= 1;
2208
2209         /* Sanity check against vm- & vfs- imposed limits */
2210         if (res > upper_limit)
2211                 res = upper_limit;
2212
2213         return res;
2214 }
2215
2216 /*
2217  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2218  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2219  * We need to be 1 filesystem block less than the 2^48 sector limit.
2220  */
2221 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2222 {
2223         loff_t res = EXT4_NDIR_BLOCKS;
2224         int meta_blocks;
2225         loff_t upper_limit;
2226         /* This is calculated to be the largest file size for a dense, block
2227          * mapped file such that the file's total number of 512-byte sectors,
2228          * including data and all indirect blocks, does not exceed (2^48 - 1).
2229          *
2230          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2231          * number of 512-byte sectors of the file.
2232          */
2233
2234         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2235                 /*
2236                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2237                  * the inode i_block field represents total file blocks in
2238                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2239                  */
2240                 upper_limit = (1LL << 32) - 1;
2241
2242                 /* total blocks in file system block size */
2243                 upper_limit >>= (bits - 9);
2244
2245         } else {
2246                 /*
2247                  * We use 48 bit ext4_inode i_blocks
2248                  * With EXT4_HUGE_FILE_FL set the i_blocks
2249                  * represent total number of blocks in
2250                  * file system block size
2251                  */
2252                 upper_limit = (1LL << 48) - 1;
2253
2254         }
2255
2256         /* indirect blocks */
2257         meta_blocks = 1;
2258         /* double indirect blocks */
2259         meta_blocks += 1 + (1LL << (bits-2));
2260         /* tripple indirect blocks */
2261         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2262
2263         upper_limit -= meta_blocks;
2264         upper_limit <<= bits;
2265
2266         res += 1LL << (bits-2);
2267         res += 1LL << (2*(bits-2));
2268         res += 1LL << (3*(bits-2));
2269         res <<= bits;
2270         if (res > upper_limit)
2271                 res = upper_limit;
2272
2273         if (res > MAX_LFS_FILESIZE)
2274                 res = MAX_LFS_FILESIZE;
2275
2276         return res;
2277 }
2278
2279 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2280                                    ext4_fsblk_t logical_sb_block, int nr)
2281 {
2282         struct ext4_sb_info *sbi = EXT4_SB(sb);
2283         ext4_group_t bg, first_meta_bg;
2284         int has_super = 0;
2285
2286         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2287
2288         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2289             nr < first_meta_bg)
2290                 return logical_sb_block + nr + 1;
2291         bg = sbi->s_desc_per_block * nr;
2292         if (ext4_bg_has_super(sb, bg))
2293                 has_super = 1;
2294
2295         return (has_super + ext4_group_first_block_no(sb, bg));
2296 }
2297
2298 /**
2299  * ext4_get_stripe_size: Get the stripe size.
2300  * @sbi: In memory super block info
2301  *
2302  * If we have specified it via mount option, then
2303  * use the mount option value. If the value specified at mount time is
2304  * greater than the blocks per group use the super block value.
2305  * If the super block value is greater than blocks per group return 0.
2306  * Allocator needs it be less than blocks per group.
2307  *
2308  */
2309 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2310 {
2311         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2312         unsigned long stripe_width =
2313                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2314
2315         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2316                 return sbi->s_stripe;
2317
2318         if (stripe_width <= sbi->s_blocks_per_group)
2319                 return stripe_width;
2320
2321         if (stride <= sbi->s_blocks_per_group)
2322                 return stride;
2323
2324         return 0;
2325 }
2326
2327 /* sysfs supprt */
2328
2329 struct ext4_attr {
2330         struct attribute attr;
2331         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2332         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2333                          const char *, size_t);
2334         int offset;
2335 };
2336
2337 static int parse_strtoul(const char *buf,
2338                 unsigned long max, unsigned long *value)
2339 {
2340         char *endp;
2341
2342         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2343         endp = skip_spaces(endp);
2344         if (*endp || *value > max)
2345                 return -EINVAL;
2346
2347         return 0;
2348 }
2349
2350 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2351                                               struct ext4_sb_info *sbi,
2352                                               char *buf)
2353 {
2354         return snprintf(buf, PAGE_SIZE, "%llu\n",
2355                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2356 }
2357
2358 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2359                                          struct ext4_sb_info *sbi, char *buf)
2360 {
2361         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2362
2363         if (!sb->s_bdev->bd_part)
2364                 return snprintf(buf, PAGE_SIZE, "0\n");
2365         return snprintf(buf, PAGE_SIZE, "%lu\n",
2366                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2367                          sbi->s_sectors_written_start) >> 1);
2368 }
2369
2370 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2371                                           struct ext4_sb_info *sbi, char *buf)
2372 {
2373         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2374
2375         if (!sb->s_bdev->bd_part)
2376                 return snprintf(buf, PAGE_SIZE, "0\n");
2377         return snprintf(buf, PAGE_SIZE, "%llu\n",
2378                         (unsigned long long)(sbi->s_kbytes_written +
2379                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2380                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2381 }
2382
2383 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2384                                           struct ext4_sb_info *sbi,
2385                                           const char *buf, size_t count)
2386 {
2387         unsigned long t;
2388
2389         if (parse_strtoul(buf, 0x40000000, &t))
2390                 return -EINVAL;
2391
2392         if (!is_power_of_2(t))
2393                 return -EINVAL;
2394
2395         sbi->s_inode_readahead_blks = t;
2396         return count;
2397 }
2398
2399 static ssize_t sbi_ui_show(struct ext4_attr *a,
2400                            struct ext4_sb_info *sbi, char *buf)
2401 {
2402         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2403
2404         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2405 }
2406
2407 static ssize_t sbi_ui_store(struct ext4_attr *a,
2408                             struct ext4_sb_info *sbi,
2409                             const char *buf, size_t count)
2410 {
2411         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2412         unsigned long t;
2413
2414         if (parse_strtoul(buf, 0xffffffff, &t))
2415                 return -EINVAL;
2416         *ui = t;
2417         return count;
2418 }
2419
2420 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2421 static struct ext4_attr ext4_attr_##_name = {                   \
2422         .attr = {.name = __stringify(_name), .mode = _mode },   \
2423         .show   = _show,                                        \
2424         .store  = _store,                                       \
2425         .offset = offsetof(struct ext4_sb_info, _elname),       \
2426 }
2427 #define EXT4_ATTR(name, mode, show, store) \
2428 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2429
2430 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2431 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2432 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2433 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2434         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2435 #define ATTR_LIST(name) &ext4_attr_##name.attr
2436
2437 EXT4_RO_ATTR(delayed_allocation_blocks);
2438 EXT4_RO_ATTR(session_write_kbytes);
2439 EXT4_RO_ATTR(lifetime_write_kbytes);
2440 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2441                  inode_readahead_blks_store, s_inode_readahead_blks);
2442 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2443 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2444 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2445 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2446 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2447 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2448 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2449 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2450
2451 static struct attribute *ext4_attrs[] = {
2452         ATTR_LIST(delayed_allocation_blocks),
2453         ATTR_LIST(session_write_kbytes),
2454         ATTR_LIST(lifetime_write_kbytes),
2455         ATTR_LIST(inode_readahead_blks),
2456         ATTR_LIST(inode_goal),
2457         ATTR_LIST(mb_stats),
2458         ATTR_LIST(mb_max_to_scan),
2459         ATTR_LIST(mb_min_to_scan),
2460         ATTR_LIST(mb_order2_req),
2461         ATTR_LIST(mb_stream_req),
2462         ATTR_LIST(mb_group_prealloc),
2463         ATTR_LIST(max_writeback_mb_bump),
2464         NULL,
2465 };
2466
2467 /* Features this copy of ext4 supports */
2468 EXT4_INFO_ATTR(lazy_itable_init);
2469 EXT4_INFO_ATTR(batched_discard);
2470
2471 static struct attribute *ext4_feat_attrs[] = {
2472         ATTR_LIST(lazy_itable_init),
2473         ATTR_LIST(batched_discard),
2474         NULL,
2475 };
2476
2477 static ssize_t ext4_attr_show(struct kobject *kobj,
2478                               struct attribute *attr, char *buf)
2479 {
2480         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2481                                                 s_kobj);
2482         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2483
2484         return a->show ? a->show(a, sbi, buf) : 0;
2485 }
2486
2487 static ssize_t ext4_attr_store(struct kobject *kobj,
2488                                struct attribute *attr,
2489                                const char *buf, size_t len)
2490 {
2491         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2492                                                 s_kobj);
2493         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2494
2495         return a->store ? a->store(a, sbi, buf, len) : 0;
2496 }
2497
2498 static void ext4_sb_release(struct kobject *kobj)
2499 {
2500         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2501                                                 s_kobj);
2502         complete(&sbi->s_kobj_unregister);
2503 }
2504
2505 static const struct sysfs_ops ext4_attr_ops = {
2506         .show   = ext4_attr_show,
2507         .store  = ext4_attr_store,
2508 };
2509
2510 static struct kobj_type ext4_ktype = {
2511         .default_attrs  = ext4_attrs,
2512         .sysfs_ops      = &ext4_attr_ops,
2513         .release        = ext4_sb_release,
2514 };
2515
2516 static void ext4_feat_release(struct kobject *kobj)
2517 {
2518         complete(&ext4_feat->f_kobj_unregister);
2519 }
2520
2521 static struct kobj_type ext4_feat_ktype = {
2522         .default_attrs  = ext4_feat_attrs,
2523         .sysfs_ops      = &ext4_attr_ops,
2524         .release        = ext4_feat_release,
2525 };
2526
2527 /*
2528  * Check whether this filesystem can be mounted based on
2529  * the features present and the RDONLY/RDWR mount requested.
2530  * Returns 1 if this filesystem can be mounted as requested,
2531  * 0 if it cannot be.
2532  */
2533 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2534 {
2535         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2536                 ext4_msg(sb, KERN_ERR,
2537                         "Couldn't mount because of "
2538                         "unsupported optional features (%x)",
2539                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2540                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2541                 return 0;
2542         }
2543
2544         if (readonly)
2545                 return 1;
2546
2547         /* Check that feature set is OK for a read-write mount */
2548         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2549                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2550                          "unsupported optional features (%x)",
2551                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2552                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2553                 return 0;
2554         }
2555         /*
2556          * Large file size enabled file system can only be mounted
2557          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2558          */
2559         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2560                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2561                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2562                                  "cannot be mounted RDWR without "
2563                                  "CONFIG_LBDAF");
2564                         return 0;
2565                 }
2566         }
2567         return 1;
2568 }
2569
2570 /*
2571  * This function is called once a day if we have errors logged
2572  * on the file system
2573  */
2574 static void print_daily_error_info(unsigned long arg)
2575 {
2576         struct super_block *sb = (struct super_block *) arg;
2577         struct ext4_sb_info *sbi;
2578         struct ext4_super_block *es;
2579
2580         sbi = EXT4_SB(sb);
2581         es = sbi->s_es;
2582
2583         if (es->s_error_count)
2584                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2585                          le32_to_cpu(es->s_error_count));
2586         if (es->s_first_error_time) {
2587                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2588                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2589                        (int) sizeof(es->s_first_error_func),
2590                        es->s_first_error_func,
2591                        le32_to_cpu(es->s_first_error_line));
2592                 if (es->s_first_error_ino)
2593                         printk(": inode %u",
2594                                le32_to_cpu(es->s_first_error_ino));
2595                 if (es->s_first_error_block)
2596                         printk(": block %llu", (unsigned long long)
2597                                le64_to_cpu(es->s_first_error_block));
2598                 printk("\n");
2599         }
2600         if (es->s_last_error_time) {
2601                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2602                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2603                        (int) sizeof(es->s_last_error_func),
2604                        es->s_last_error_func,
2605                        le32_to_cpu(es->s_last_error_line));
2606                 if (es->s_last_error_ino)
2607                         printk(": inode %u",
2608                                le32_to_cpu(es->s_last_error_ino));
2609                 if (es->s_last_error_block)
2610                         printk(": block %llu", (unsigned long long)
2611                                le64_to_cpu(es->s_last_error_block));
2612                 printk("\n");
2613         }
2614         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2615 }
2616
2617 static void ext4_lazyinode_timeout(unsigned long data)
2618 {
2619         struct task_struct *p = (struct task_struct *)data;
2620         wake_up_process(p);
2621 }
2622
2623 /* Find next suitable group and run ext4_init_inode_table */
2624 static int ext4_run_li_request(struct ext4_li_request *elr)
2625 {
2626         struct ext4_group_desc *gdp = NULL;
2627         ext4_group_t group, ngroups;
2628         struct super_block *sb;
2629         unsigned long timeout = 0;
2630         int ret = 0;
2631
2632         sb = elr->lr_super;
2633         ngroups = EXT4_SB(sb)->s_groups_count;
2634
2635         for (group = elr->lr_next_group; group < ngroups; group++) {
2636                 gdp = ext4_get_group_desc(sb, group, NULL);
2637                 if (!gdp) {
2638                         ret = 1;
2639                         break;
2640                 }
2641
2642                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2643                         break;
2644         }
2645
2646         if (group == ngroups)
2647                 ret = 1;
2648
2649         if (!ret) {
2650                 timeout = jiffies;
2651                 ret = ext4_init_inode_table(sb, group,
2652                                             elr->lr_timeout ? 0 : 1);
2653                 if (elr->lr_timeout == 0) {
2654                         timeout = jiffies - timeout;
2655                         if (elr->lr_sbi->s_li_wait_mult)
2656                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2657                         else
2658                                 timeout *= 20;
2659                         elr->lr_timeout = timeout;
2660                 }
2661                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2662                 elr->lr_next_group = group + 1;
2663         }
2664
2665         return ret;
2666 }
2667
2668 /*
2669  * Remove lr_request from the list_request and free the
2670  * request tructure. Should be called with li_list_mtx held
2671  */
2672 static void ext4_remove_li_request(struct ext4_li_request *elr)
2673 {
2674         struct ext4_sb_info *sbi;
2675
2676         if (!elr)
2677                 return;
2678
2679         sbi = elr->lr_sbi;
2680
2681         list_del(&elr->lr_request);
2682         sbi->s_li_request = NULL;
2683         kfree(elr);
2684 }
2685
2686 static void ext4_unregister_li_request(struct super_block *sb)
2687 {
2688         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2689
2690         if (!ext4_li_info)
2691                 return;
2692
2693         mutex_lock(&ext4_li_info->li_list_mtx);
2694         ext4_remove_li_request(elr);
2695         mutex_unlock(&ext4_li_info->li_list_mtx);
2696 }
2697
2698 /*
2699  * This is the function where ext4lazyinit thread lives. It walks
2700  * through the request list searching for next scheduled filesystem.
2701  * When such a fs is found, run the lazy initialization request
2702  * (ext4_rn_li_request) and keep track of the time spend in this
2703  * function. Based on that time we compute next schedule time of
2704  * the request. When walking through the list is complete, compute
2705  * next waking time and put itself into sleep.
2706  */
2707 static int ext4_lazyinit_thread(void *arg)
2708 {
2709         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2710         struct list_head *pos, *n;
2711         struct ext4_li_request *elr;
2712         unsigned long next_wakeup;
2713         DEFINE_WAIT(wait);
2714
2715         BUG_ON(NULL == eli);
2716
2717         eli->li_timer.data = (unsigned long)current;
2718         eli->li_timer.function = ext4_lazyinode_timeout;
2719
2720         eli->li_task = current;
2721         wake_up(&eli->li_wait_task);
2722
2723 cont_thread:
2724         while (true) {
2725                 next_wakeup = MAX_JIFFY_OFFSET;
2726
2727                 mutex_lock(&eli->li_list_mtx);
2728                 if (list_empty(&eli->li_request_list)) {
2729                         mutex_unlock(&eli->li_list_mtx);
2730                         goto exit_thread;
2731                 }
2732
2733                 list_for_each_safe(pos, n, &eli->li_request_list) {
2734                         elr = list_entry(pos, struct ext4_li_request,
2735                                          lr_request);
2736
2737                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2738                                 if (ext4_run_li_request(elr) != 0) {
2739                                         /* error, remove the lazy_init job */
2740                                         ext4_remove_li_request(elr);
2741                                         continue;
2742                                 }
2743                         }
2744
2745                         if (time_before(elr->lr_next_sched, next_wakeup))
2746                                 next_wakeup = elr->lr_next_sched;
2747                 }
2748                 mutex_unlock(&eli->li_list_mtx);
2749
2750                 if (freezing(current))
2751                         refrigerator();
2752
2753                 if ((time_after_eq(jiffies, next_wakeup)) ||
2754                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2755                         cond_resched();
2756                         continue;
2757                 }
2758
2759                 eli->li_timer.expires = next_wakeup;
2760                 add_timer(&eli->li_timer);
2761                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2762                                 TASK_INTERRUPTIBLE);
2763                 if (time_before(jiffies, next_wakeup))
2764                         schedule();
2765                 finish_wait(&eli->li_wait_daemon, &wait);
2766         }
2767
2768 exit_thread:
2769         /*
2770          * It looks like the request list is empty, but we need
2771          * to check it under the li_list_mtx lock, to prevent any
2772          * additions into it, and of course we should lock ext4_li_mtx
2773          * to atomically free the list and ext4_li_info, because at
2774          * this point another ext4 filesystem could be registering
2775          * new one.
2776          */
2777         mutex_lock(&ext4_li_mtx);
2778         mutex_lock(&eli->li_list_mtx);
2779         if (!list_empty(&eli->li_request_list)) {
2780                 mutex_unlock(&eli->li_list_mtx);
2781                 mutex_unlock(&ext4_li_mtx);
2782                 goto cont_thread;
2783         }
2784         mutex_unlock(&eli->li_list_mtx);
2785         del_timer_sync(&ext4_li_info->li_timer);
2786         eli->li_task = NULL;
2787         wake_up(&eli->li_wait_task);
2788
2789         kfree(ext4_li_info);
2790         ext4_li_info = NULL;
2791         mutex_unlock(&ext4_li_mtx);
2792
2793         return 0;
2794 }
2795
2796 static void ext4_clear_request_list(void)
2797 {
2798         struct list_head *pos, *n;
2799         struct ext4_li_request *elr;
2800
2801         mutex_lock(&ext4_li_info->li_list_mtx);
2802         if (list_empty(&ext4_li_info->li_request_list))
2803                 return;
2804
2805         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2806                 elr = list_entry(pos, struct ext4_li_request,
2807                                  lr_request);
2808                 ext4_remove_li_request(elr);
2809         }
2810         mutex_unlock(&ext4_li_info->li_list_mtx);
2811 }
2812
2813 static int ext4_run_lazyinit_thread(void)
2814 {
2815         struct task_struct *t;
2816
2817         t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2818         if (IS_ERR(t)) {
2819                 int err = PTR_ERR(t);
2820                 ext4_clear_request_list();
2821                 del_timer_sync(&ext4_li_info->li_timer);
2822                 kfree(ext4_li_info);
2823                 ext4_li_info = NULL;
2824                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2825                                  "initialization thread\n",
2826                                  err);
2827                 return err;
2828         }
2829         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2830
2831         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2832         return 0;
2833 }
2834
2835 /*
2836  * Check whether it make sense to run itable init. thread or not.
2837  * If there is at least one uninitialized inode table, return
2838  * corresponding group number, else the loop goes through all
2839  * groups and return total number of groups.
2840  */
2841 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2842 {
2843         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2844         struct ext4_group_desc *gdp = NULL;
2845
2846         for (group = 0; group < ngroups; group++) {
2847                 gdp = ext4_get_group_desc(sb, group, NULL);
2848                 if (!gdp)
2849                         continue;
2850
2851                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2852                         break;
2853         }
2854
2855         return group;
2856 }
2857
2858 static int ext4_li_info_new(void)
2859 {
2860         struct ext4_lazy_init *eli = NULL;
2861
2862         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2863         if (!eli)
2864                 return -ENOMEM;
2865
2866         eli->li_task = NULL;
2867         INIT_LIST_HEAD(&eli->li_request_list);
2868         mutex_init(&eli->li_list_mtx);
2869
2870         init_waitqueue_head(&eli->li_wait_daemon);
2871         init_waitqueue_head(&eli->li_wait_task);
2872         init_timer(&eli->li_timer);
2873         eli->li_state |= EXT4_LAZYINIT_QUIT;
2874
2875         ext4_li_info = eli;
2876
2877         return 0;
2878 }
2879
2880 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2881                                             ext4_group_t start)
2882 {
2883         struct ext4_sb_info *sbi = EXT4_SB(sb);
2884         struct ext4_li_request *elr;
2885         unsigned long rnd;
2886
2887         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2888         if (!elr)
2889                 return NULL;
2890
2891         elr->lr_super = sb;
2892         elr->lr_sbi = sbi;
2893         elr->lr_next_group = start;
2894
2895         /*
2896          * Randomize first schedule time of the request to
2897          * spread the inode table initialization requests
2898          * better.
2899          */
2900         get_random_bytes(&rnd, sizeof(rnd));
2901         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2902                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2903
2904         return elr;
2905 }
2906
2907 static int ext4_register_li_request(struct super_block *sb,
2908                                     ext4_group_t first_not_zeroed)
2909 {
2910         struct ext4_sb_info *sbi = EXT4_SB(sb);
2911         struct ext4_li_request *elr;
2912         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2913         int ret;
2914
2915         if (sbi->s_li_request != NULL)
2916                 return 0;
2917
2918         if (first_not_zeroed == ngroups ||
2919             (sb->s_flags & MS_RDONLY) ||
2920             !test_opt(sb, INIT_INODE_TABLE)) {
2921                 sbi->s_li_request = NULL;
2922                 return 0;
2923         }
2924
2925         if (first_not_zeroed == ngroups) {
2926                 sbi->s_li_request = NULL;
2927                 return 0;
2928         }
2929
2930         elr = ext4_li_request_new(sb, first_not_zeroed);
2931         if (!elr)
2932                 return -ENOMEM;
2933
2934         mutex_lock(&ext4_li_mtx);
2935
2936         if (NULL == ext4_li_info) {
2937                 ret = ext4_li_info_new();
2938                 if (ret)
2939                         goto out;
2940         }
2941
2942         mutex_lock(&ext4_li_info->li_list_mtx);
2943         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2944         mutex_unlock(&ext4_li_info->li_list_mtx);
2945
2946         sbi->s_li_request = elr;
2947
2948         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2949                 ret = ext4_run_lazyinit_thread();
2950                 if (ret)
2951                         goto out;
2952         }
2953 out:
2954         mutex_unlock(&ext4_li_mtx);
2955         if (ret)
2956                 kfree(elr);
2957         return ret;
2958 }
2959
2960 /*
2961  * We do not need to lock anything since this is called on
2962  * module unload.
2963  */
2964 static void ext4_destroy_lazyinit_thread(void)
2965 {
2966         /*
2967          * If thread exited earlier
2968          * there's nothing to be done.
2969          */
2970         if (!ext4_li_info)
2971                 return;
2972
2973         ext4_clear_request_list();
2974
2975         while (ext4_li_info->li_task) {
2976                 wake_up(&ext4_li_info->li_wait_daemon);
2977                 wait_event(ext4_li_info->li_wait_task,
2978                            ext4_li_info->li_task == NULL);
2979         }
2980 }
2981
2982 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2983                                 __releases(kernel_lock)
2984                                 __acquires(kernel_lock)
2985 {
2986         char *orig_data = kstrdup(data, GFP_KERNEL);
2987         struct buffer_head *bh;
2988         struct ext4_super_block *es = NULL;
2989         struct ext4_sb_info *sbi;
2990         ext4_fsblk_t block;
2991         ext4_fsblk_t sb_block = get_sb_block(&data);
2992         ext4_fsblk_t logical_sb_block;
2993         unsigned long offset = 0;
2994         unsigned long journal_devnum = 0;
2995         unsigned long def_mount_opts;
2996         struct inode *root;
2997         char *cp;
2998         const char *descr;
2999         int ret = -ENOMEM;
3000         int blocksize;
3001         unsigned int db_count;
3002         unsigned int i;
3003         int needs_recovery, has_huge_files;
3004         __u64 blocks_count;
3005         int err;
3006         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3007         ext4_group_t first_not_zeroed;
3008
3009         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3010         if (!sbi)
3011                 goto out_free_orig;
3012
3013         sbi->s_blockgroup_lock =
3014                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3015         if (!sbi->s_blockgroup_lock) {
3016                 kfree(sbi);
3017                 goto out_free_orig;
3018         }
3019         sb->s_fs_info = sbi;
3020         sbi->s_mount_opt = 0;
3021         sbi->s_resuid = EXT4_DEF_RESUID;
3022         sbi->s_resgid = EXT4_DEF_RESGID;
3023         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3024         sbi->s_sb_block = sb_block;
3025         if (sb->s_bdev->bd_part)
3026                 sbi->s_sectors_written_start =
3027                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3028
3029         /* Cleanup superblock name */
3030         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3031                 *cp = '!';
3032
3033         ret = -EINVAL;
3034         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3035         if (!blocksize) {
3036                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3037                 goto out_fail;
3038         }
3039
3040         /*
3041          * The ext4 superblock will not be buffer aligned for other than 1kB
3042          * block sizes.  We need to calculate the offset from buffer start.
3043          */
3044         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3045                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3046                 offset = do_div(logical_sb_block, blocksize);
3047         } else {
3048                 logical_sb_block = sb_block;
3049         }
3050
3051         if (!(bh = sb_bread(sb, logical_sb_block))) {
3052                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3053                 goto out_fail;
3054         }
3055         /*
3056          * Note: s_es must be initialized as soon as possible because
3057          *       some ext4 macro-instructions depend on its value
3058          */
3059         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3060         sbi->s_es = es;
3061         sb->s_magic = le16_to_cpu(es->s_magic);
3062         if (sb->s_magic != EXT4_SUPER_MAGIC)
3063                 goto cantfind_ext4;
3064         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3065
3066         /* Set defaults before we parse the mount options */
3067         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3068         set_opt(sbi->s_mount_opt, INIT_INODE_TABLE);
3069         if (def_mount_opts & EXT4_DEFM_DEBUG)
3070                 set_opt(sbi->s_mount_opt, DEBUG);
3071         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3072                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3073                         "2.6.38");
3074                 set_opt(sbi->s_mount_opt, GRPID);
3075         }
3076         if (def_mount_opts & EXT4_DEFM_UID16)
3077                 set_opt(sbi->s_mount_opt, NO_UID32);
3078 #ifdef CONFIG_EXT4_FS_XATTR
3079         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3080                 set_opt(sbi->s_mount_opt, XATTR_USER);
3081 #endif
3082 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3083         if (def_mount_opts & EXT4_DEFM_ACL)
3084                 set_opt(sbi->s_mount_opt, POSIX_ACL);
3085 #endif
3086         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3087                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3088         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3089                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
3090         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3091                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3092
3093         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3094                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
3095         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3096                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
3097         else
3098                 set_opt(sbi->s_mount_opt, ERRORS_RO);
3099         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3100                 set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
3101         if (def_mount_opts & EXT4_DEFM_DISCARD)
3102                 set_opt(sbi->s_mount_opt, DISCARD);
3103
3104         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3105         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3106         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3107         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3108         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3109
3110         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3111                 set_opt(sbi->s_mount_opt, BARRIER);
3112
3113         /*
3114          * enable delayed allocation by default
3115          * Use -o nodelalloc to turn it off
3116          */
3117         if (!IS_EXT3_SB(sb) &&
3118             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3119                 set_opt(sbi->s_mount_opt, DELALLOC);
3120
3121         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3122                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3123                 ext4_msg(sb, KERN_WARNING,
3124                          "failed to parse options in superblock: %s",
3125                          sbi->s_es->s_mount_opts);
3126         }
3127         if (!parse_options((char *) data, sb, &journal_devnum,
3128                            &journal_ioprio, NULL, 0))
3129                 goto failed_mount;
3130
3131         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3132                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3133
3134         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3135             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3136              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3137              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3138                 ext4_msg(sb, KERN_WARNING,
3139                        "feature flags set on rev 0 fs, "
3140                        "running e2fsck is recommended");
3141
3142         /*
3143          * Check feature flags regardless of the revision level, since we
3144          * previously didn't change the revision level when setting the flags,
3145          * so there is a chance incompat flags are set on a rev 0 filesystem.
3146          */
3147         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3148                 goto failed_mount;
3149
3150         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3151
3152         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3153             blocksize > EXT4_MAX_BLOCK_SIZE) {
3154                 ext4_msg(sb, KERN_ERR,
3155                        "Unsupported filesystem blocksize %d", blocksize);
3156                 goto failed_mount;
3157         }
3158
3159         if (sb->s_blocksize != blocksize) {
3160                 /* Validate the filesystem blocksize */
3161                 if (!sb_set_blocksize(sb, blocksize)) {
3162                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3163                                         blocksize);
3164                         goto failed_mount;
3165                 }
3166
3167                 brelse(bh);
3168                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3169                 offset = do_div(logical_sb_block, blocksize);
3170                 bh = sb_bread(sb, logical_sb_block);
3171                 if (!bh) {
3172                         ext4_msg(sb, KERN_ERR,
3173                                "Can't read superblock on 2nd try");
3174                         goto failed_mount;
3175                 }
3176                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3177                 sbi->s_es = es;
3178                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3179                         ext4_msg(sb, KERN_ERR,
3180                                "Magic mismatch, very weird!");
3181                         goto failed_mount;
3182                 }
3183         }
3184
3185         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3186                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3187         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3188                                                       has_huge_files);
3189         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3190
3191         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3192                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3193                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3194         } else {
3195                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3196                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3197                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3198                     (!is_power_of_2(sbi->s_inode_size)) ||
3199                     (sbi->s_inode_size > blocksize)) {
3200                         ext4_msg(sb, KERN_ERR,
3201                                "unsupported inode size: %d",
3202                                sbi->s_inode_size);
3203                         goto failed_mount;
3204                 }
3205                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3206                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3207         }
3208
3209         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3210         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3211                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3212                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3213                     !is_power_of_2(sbi->s_desc_size)) {
3214                         ext4_msg(sb, KERN_ERR,
3215                                "unsupported descriptor size %lu",
3216                                sbi->s_desc_size);
3217                         goto failed_mount;
3218                 }
3219         } else
3220                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3221
3222         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3223         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3224         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3225                 goto cantfind_ext4;
3226
3227         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3228         if (sbi->s_inodes_per_block == 0)
3229                 goto cantfind_ext4;
3230         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3231                                         sbi->s_inodes_per_block;
3232         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3233         sbi->s_sbh = bh;
3234         sbi->s_mount_state = le16_to_cpu(es->s_state);
3235         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3236         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3237
3238         for (i = 0; i < 4; i++)
3239                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3240         sbi->s_def_hash_version = es->s_def_hash_version;
3241         i = le32_to_cpu(es->s_flags);
3242         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3243                 sbi->s_hash_unsigned = 3;
3244         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3245 #ifdef __CHAR_UNSIGNED__
3246                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3247                 sbi->s_hash_unsigned = 3;
3248 #else
3249                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3250 #endif
3251                 sb->s_dirt = 1;
3252         }
3253
3254         if (sbi->s_blocks_per_group > blocksize * 8) {
3255                 ext4_msg(sb, KERN_ERR,
3256                        "#blocks per group too big: %lu",
3257                        sbi->s_blocks_per_group);
3258                 goto failed_mount;
3259         }
3260         if (sbi->s_inodes_per_group > blocksize * 8) {
3261                 ext4_msg(sb, KERN_ERR,
3262                        "#inodes per group too big: %lu",
3263                        sbi->s_inodes_per_group);
3264                 goto failed_mount;
3265         }
3266
3267         /*
3268          * Test whether we have more sectors than will fit in sector_t,
3269          * and whether the max offset is addressable by the page cache.
3270          */
3271         ret = generic_check_addressable(sb->s_blocksize_bits,
3272                                         ext4_blocks_count(es));
3273         if (ret) {
3274                 ext4_msg(sb, KERN_ERR, "filesystem"
3275                          " too large to mount safely on this system");
3276                 if (sizeof(sector_t) < 8)
3277                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3278                 goto failed_mount;
3279         }
3280
3281         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3282                 goto cantfind_ext4;
3283
3284         /* check blocks count against device size */
3285         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3286         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3287                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3288                        "exceeds size of device (%llu blocks)",
3289                        ext4_blocks_count(es), blocks_count);
3290                 goto failed_mount;
3291         }
3292
3293         /*
3294          * It makes no sense for the first data block to be beyond the end
3295          * of the filesystem.
3296          */
3297         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3298                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3299                          "block %u is beyond end of filesystem (%llu)",
3300                          le32_to_cpu(es->s_first_data_block),
3301                          ext4_blocks_count(es));
3302                 goto failed_mount;
3303         }
3304         blocks_count = (ext4_blocks_count(es) -
3305                         le32_to_cpu(es->s_first_data_block) +
3306                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3307         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3308         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3309                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3310                        "(block count %llu, first data block %u, "
3311                        "blocks per group %lu)", sbi->s_groups_count,
3312                        ext4_blocks_count(es),
3313                        le32_to_cpu(es->s_first_data_block),
3314                        EXT4_BLOCKS_PER_GROUP(sb));
3315                 goto failed_mount;
3316         }
3317         sbi->s_groups_count = blocks_count;
3318         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3319                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3320         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3321                    EXT4_DESC_PER_BLOCK(sb);
3322         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3323                                     GFP_KERNEL);
3324         if (sbi->s_group_desc == NULL) {
3325                 ext4_msg(sb, KERN_ERR, "not enough memory");
3326                 goto failed_mount;
3327         }
3328
3329 #ifdef CONFIG_PROC_FS
3330         if (ext4_proc_root)
3331                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3332 #endif
3333
3334         bgl_lock_init(sbi->s_blockgroup_lock);
3335
3336         for (i = 0; i < db_count; i++) {
3337                 block = descriptor_loc(sb, logical_sb_block, i);
3338                 sbi->s_group_desc[i] = sb_bread(sb, block);
3339                 if (!sbi->s_group_desc[i]) {
3340                         ext4_msg(sb, KERN_ERR,
3341                                "can't read group descriptor %d", i);
3342                         db_count = i;
3343                         goto failed_mount2;
3344                 }
3345         }
3346         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3347                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3348                 goto failed_mount2;
3349         }
3350         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3351                 if (!ext4_fill_flex_info(sb)) {
3352                         ext4_msg(sb, KERN_ERR,
3353                                "unable to initialize "
3354                                "flex_bg meta info!");
3355                         goto failed_mount2;
3356                 }
3357
3358         sbi->s_gdb_count = db_count;
3359         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3360         spin_lock_init(&sbi->s_next_gen_lock);
3361
3362         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3363                         ext4_count_free_blocks(sb));
3364         if (!err) {
3365                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3366                                 ext4_count_free_inodes(sb));
3367         }
3368         if (!err) {
3369                 err = percpu_counter_init(&sbi->s_dirs_counter,
3370                                 ext4_count_dirs(sb));
3371         }
3372         if (!err) {
3373                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3374         }
3375         if (err) {
3376                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3377                 goto failed_mount3;
3378         }
3379
3380         sbi->s_stripe = ext4_get_stripe_size(sbi);
3381         sbi->s_max_writeback_mb_bump = 128;
3382
3383         /*
3384          * set up enough so that it can read an inode
3385          */
3386         if (!test_opt(sb, NOLOAD) &&
3387             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3388                 sb->s_op = &ext4_sops;
3389         else
3390                 sb->s_op = &ext4_nojournal_sops;
3391         sb->s_export_op = &ext4_export_ops;
3392         sb->s_xattr = ext4_xattr_handlers;
3393 #ifdef CONFIG_QUOTA
3394         sb->s_qcop = &ext4_qctl_operations;
3395         sb->dq_op = &ext4_quota_operations;
3396 #endif
3397         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3398         mutex_init(&sbi->s_orphan_lock);
3399         mutex_init(&sbi->s_resize_lock);
3400
3401         sb->s_root = NULL;
3402
3403         needs_recovery = (es->s_last_orphan != 0 ||
3404                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3405                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3406
3407         /*
3408          * The first inode we look at is the journal inode.  Don't try
3409          * root first: it may be modified in the journal!
3410          */
3411         if (!test_opt(sb, NOLOAD) &&
3412             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3413                 if (ext4_load_journal(sb, es, journal_devnum))
3414                         goto failed_mount3;
3415         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3416               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3417                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3418                        "suppressed and not mounted read-only");
3419                 goto failed_mount_wq;
3420         } else {
3421                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
3422                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
3423                 sbi->s_journal = NULL;
3424                 needs_recovery = 0;
3425                 goto no_journal;
3426         }
3427
3428         if (ext4_blocks_count(es) > 0xffffffffULL &&
3429             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3430                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3431                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3432                 goto failed_mount_wq;
3433         }
3434
3435         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3436                 jbd2_journal_set_features(sbi->s_journal,
3437                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3438                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3439         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3440                 jbd2_journal_set_features(sbi->s_journal,
3441                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3442                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3443                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3444         } else {
3445                 jbd2_journal_clear_features(sbi->s_journal,
3446                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3447                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3448         }
3449
3450         /* We have now updated the journal if required, so we can
3451          * validate the data journaling mode. */
3452         switch (test_opt(sb, DATA_FLAGS)) {
3453         case 0:
3454                 /* No mode set, assume a default based on the journal
3455                  * capabilities: ORDERED_DATA if the journal can
3456                  * cope, else JOURNAL_DATA
3457                  */
3458                 if (jbd2_journal_check_available_features
3459                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3460                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
3461                 else
3462                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
3463                 break;
3464
3465         case EXT4_MOUNT_ORDERED_DATA:
3466         case EXT4_MOUNT_WRITEBACK_DATA:
3467                 if (!jbd2_journal_check_available_features
3468                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3469                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3470                                "requested data journaling mode");
3471                         goto failed_mount_wq;
3472                 }
3473         default:
3474                 break;
3475         }
3476         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3477
3478         /*
3479          * The journal may have updated the bg summary counts, so we
3480          * need to update the global counters.
3481          */
3482         percpu_counter_set(&sbi->s_freeblocks_counter,
3483                            ext4_count_free_blocks(sb));
3484         percpu_counter_set(&sbi->s_freeinodes_counter,
3485                            ext4_count_free_inodes(sb));
3486         percpu_counter_set(&sbi->s_dirs_counter,
3487                            ext4_count_dirs(sb));
3488         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3489
3490 no_journal:
3491         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3492         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3493                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3494                 goto failed_mount_wq;
3495         }
3496
3497         /*
3498          * The jbd2_journal_load will have done any necessary log recovery,
3499          * so we can safely mount the rest of the filesystem now.
3500          */
3501
3502         root = ext4_iget(sb, EXT4_ROOT_INO);
3503         if (IS_ERR(root)) {
3504                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3505                 ret = PTR_ERR(root);
3506                 goto failed_mount4;
3507         }
3508         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3509                 iput(root);
3510                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3511                 goto failed_mount4;
3512         }
3513         sb->s_root = d_alloc_root(root);
3514         if (!sb->s_root) {
3515                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3516                 iput(root);
3517                 ret = -ENOMEM;
3518                 goto failed_mount4;
3519         }
3520
3521         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3522
3523         /* determine the minimum size of new large inodes, if present */
3524         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3525                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3526                                                      EXT4_GOOD_OLD_INODE_SIZE;
3527                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3528                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3529                         if (sbi->s_want_extra_isize <
3530                             le16_to_cpu(es->s_want_extra_isize))
3531                                 sbi->s_want_extra_isize =
3532                                         le16_to_cpu(es->s_want_extra_isize);
3533                         if (sbi->s_want_extra_isize <
3534                             le16_to_cpu(es->s_min_extra_isize))
3535                                 sbi->s_want_extra_isize =
3536                                         le16_to_cpu(es->s_min_extra_isize);
3537                 }
3538         }
3539         /* Check if enough inode space is available */
3540         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3541                                                         sbi->s_inode_size) {
3542                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3543                                                        EXT4_GOOD_OLD_INODE_SIZE;
3544                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3545                          "available");
3546         }
3547
3548         if (test_opt(sb, DELALLOC) &&
3549             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3550                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3551                          "requested data journaling mode");
3552                 clear_opt(sbi->s_mount_opt, DELALLOC);
3553         }
3554         if (test_opt(sb, DIOREAD_NOLOCK)) {
3555                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3556                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3557                                 "option - requested data journaling mode");
3558                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3559                 }
3560                 if (sb->s_blocksize < PAGE_SIZE) {
3561                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3562                                 "option - block size is too small");
3563                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3564                 }
3565         }
3566
3567         err = ext4_setup_system_zone(sb);
3568         if (err) {
3569                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3570                          "zone (%d)", err);
3571                 goto failed_mount4;
3572         }
3573
3574         ext4_ext_init(sb);
3575         err = ext4_mb_init(sb, needs_recovery);
3576         if (err) {
3577                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3578                          err);
3579                 goto failed_mount4;
3580         }
3581
3582         err = ext4_register_li_request(sb, first_not_zeroed);
3583         if (err)
3584                 goto failed_mount4;
3585
3586         sbi->s_kobj.kset = ext4_kset;
3587         init_completion(&sbi->s_kobj_unregister);
3588         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3589                                    "%s", sb->s_id);
3590         if (err) {
3591                 ext4_mb_release(sb);
3592                 ext4_ext_release(sb);
3593                 goto failed_mount4;
3594         };
3595
3596         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3597         ext4_orphan_cleanup(sb, es);
3598         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3599         if (needs_recovery) {
3600                 ext4_msg(sb, KERN_INFO, "recovery complete");
3601                 ext4_mark_recovery_complete(sb, es);
3602         }
3603         if (EXT4_SB(sb)->s_journal) {
3604                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3605                         descr = " journalled data mode";
3606                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3607                         descr = " ordered data mode";
3608                 else
3609                         descr = " writeback data mode";
3610         } else
3611                 descr = "out journal";
3612
3613         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3614                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3615                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3616
3617         init_timer(&sbi->s_err_report);
3618         sbi->s_err_report.function = print_daily_error_info;
3619         sbi->s_err_report.data = (unsigned long) sb;
3620         if (es->s_error_count)
3621                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3622
3623         kfree(orig_data);
3624         return 0;
3625
3626 cantfind_ext4:
3627         if (!silent)
3628                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3629         goto failed_mount;
3630
3631 failed_mount4:
3632         ext4_msg(sb, KERN_ERR, "mount failed");
3633         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3634 failed_mount_wq:
3635         ext4_release_system_zone(sb);
3636         if (sbi->s_journal) {
3637                 jbd2_journal_destroy(sbi->s_journal);
3638                 sbi->s_journal = NULL;
3639         }
3640 failed_mount3:
3641         if (sbi->s_flex_groups) {
3642                 if (is_vmalloc_addr(sbi->s_flex_groups))
3643                         vfree(sbi->s_flex_groups);
3644                 else
3645                         kfree(sbi->s_flex_groups);
3646         }
3647         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3648         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3649         percpu_counter_destroy(&sbi->s_dirs_counter);
3650         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3651 failed_mount2:
3652         for (i = 0; i < db_count; i++)
3653                 brelse(sbi->s_group_desc[i]);
3654         kfree(sbi->s_group_desc);
3655 failed_mount:
3656         if (sbi->s_proc) {
3657                 remove_proc_entry(sb->s_id, ext4_proc_root);
3658         }
3659 #ifdef CONFIG_QUOTA
3660         for (i = 0; i < MAXQUOTAS; i++)
3661                 kfree(sbi->s_qf_names[i]);
3662 #endif
3663         ext4_blkdev_remove(sbi);
3664         brelse(bh);
3665 out_fail:
3666         sb->s_fs_info = NULL;
3667         kfree(sbi->s_blockgroup_lock);
3668         kfree(sbi);
3669 out_free_orig:
3670         kfree(orig_data);
3671         return ret;
3672 }
3673
3674 /*
3675  * Setup any per-fs journal parameters now.  We'll do this both on
3676  * initial mount, once the journal has been initialised but before we've
3677  * done any recovery; and again on any subsequent remount.
3678  */
3679 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3680 {
3681         struct ext4_sb_info *sbi = EXT4_SB(sb);
3682
3683         journal->j_commit_interval = sbi->s_commit_interval;
3684         journal->j_min_batch_time = sbi->s_min_batch_time;
3685         journal->j_max_batch_time = sbi->s_max_batch_time;
3686
3687         write_lock(&journal->j_state_lock);
3688         if (test_opt(sb, BARRIER))
3689                 journal->j_flags |= JBD2_BARRIER;
3690         else
3691                 journal->j_flags &= ~JBD2_BARRIER;
3692         if (test_opt(sb, DATA_ERR_ABORT))
3693                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3694         else
3695                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3696         write_unlock(&journal->j_state_lock);
3697 }
3698
3699 static journal_t *ext4_get_journal(struct super_block *sb,
3700                                    unsigned int journal_inum)
3701 {
3702         struct inode *journal_inode;
3703         journal_t *journal;
3704
3705         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3706
3707         /* First, test for the existence of a valid inode on disk.  Bad
3708          * things happen if we iget() an unused inode, as the subsequent
3709          * iput() will try to delete it. */
3710
3711         journal_inode = ext4_iget(sb, journal_inum);
3712         if (IS_ERR(journal_inode)) {
3713                 ext4_msg(sb, KERN_ERR, "no journal found");
3714                 return NULL;
3715         }
3716         if (!journal_inode->i_nlink) {
3717                 make_bad_inode(journal_inode);
3718                 iput(journal_inode);
3719                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3720                 return NULL;
3721         }
3722
3723         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3724                   journal_inode, journal_inode->i_size);
3725         if (!S_ISREG(journal_inode->i_mode)) {
3726                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3727                 iput(journal_inode);
3728                 return NULL;
3729         }
3730
3731         journal = jbd2_journal_init_inode(journal_inode);
3732         if (!journal) {
3733                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3734                 iput(journal_inode);
3735                 return NULL;
3736         }
3737         journal->j_private = sb;
3738         ext4_init_journal_params(sb, journal);
3739         return journal;
3740 }
3741
3742 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3743                                        dev_t j_dev)
3744 {
3745         struct buffer_head *bh;
3746         journal_t *journal;
3747         ext4_fsblk_t start;
3748         ext4_fsblk_t len;
3749         int hblock, blocksize;
3750         ext4_fsblk_t sb_block;
3751         unsigned long offset;
3752         struct ext4_super_block *es;
3753         struct block_device *bdev;
3754
3755         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3756
3757         bdev = ext4_blkdev_get(j_dev, sb);
3758         if (bdev == NULL)
3759                 return NULL;
3760
3761         if (bd_claim(bdev, sb)) {
3762                 ext4_msg(sb, KERN_ERR,
3763                         "failed to claim external journal device");
3764                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3765                 return NULL;
3766         }
3767
3768         blocksize = sb->s_blocksize;
3769         hblock = bdev_logical_block_size(bdev);
3770         if (blocksize < hblock) {
3771                 ext4_msg(sb, KERN_ERR,
3772                         "blocksize too small for journal device");
3773                 goto out_bdev;
3774         }
3775
3776         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3777         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3778         set_blocksize(bdev, blocksize);
3779         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3780                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3781                        "external journal");
3782                 goto out_bdev;
3783         }
3784
3785         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3786         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3787             !(le32_to_cpu(es->s_feature_incompat) &
3788               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3789                 ext4_msg(sb, KERN_ERR, "external journal has "
3790                                         "bad superblock");
3791                 brelse(bh);
3792                 goto out_bdev;
3793         }
3794
3795         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3796                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3797                 brelse(bh);
3798                 goto out_bdev;
3799         }
3800
3801         len = ext4_blocks_count(es);
3802         start = sb_block + 1;
3803         brelse(bh);     /* we're done with the superblock */
3804
3805         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3806                                         start, len, blocksize);
3807         if (!journal) {
3808                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3809                 goto out_bdev;
3810         }
3811         journal->j_private = sb;
3812         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3813         wait_on_buffer(journal->j_sb_buffer);
3814         if (!buffer_uptodate(journal->j_sb_buffer)) {
3815                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3816                 goto out_journal;
3817         }
3818         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3819                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3820                                         "user (unsupported) - %d",
3821                         be32_to_cpu(journal->j_superblock->s_nr_users));
3822                 goto out_journal;
3823         }
3824         EXT4_SB(sb)->journal_bdev = bdev;
3825         ext4_init_journal_params(sb, journal);
3826         return journal;
3827
3828 out_journal:
3829         jbd2_journal_destroy(journal);
3830 out_bdev:
3831         ext4_blkdev_put(bdev);
3832         return NULL;
3833 }
3834
3835 static int ext4_load_journal(struct super_block *sb,
3836                              struct ext4_super_block *es,
3837                              unsigned long journal_devnum)
3838 {
3839         journal_t *journal;
3840         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3841         dev_t journal_dev;
3842         int err = 0;
3843         int really_read_only;
3844
3845         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3846
3847         if (journal_devnum &&
3848             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3849                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3850                         "numbers have changed");
3851                 journal_dev = new_decode_dev(journal_devnum);
3852         } else
3853                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3854
3855         really_read_only = bdev_read_only(sb->s_bdev);
3856
3857         /*
3858          * Are we loading a blank journal or performing recovery after a
3859          * crash?  For recovery, we need to check in advance whether we
3860          * can get read-write access to the device.
3861          */
3862         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3863                 if (sb->s_flags & MS_RDONLY) {
3864                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3865                                         "required on readonly filesystem");
3866                         if (really_read_only) {
3867                                 ext4_msg(sb, KERN_ERR, "write access "
3868                                         "unavailable, cannot proceed");
3869                                 return -EROFS;
3870                         }
3871                         ext4_msg(sb, KERN_INFO, "write access will "
3872                                "be enabled during recovery");
3873                 }
3874         }
3875
3876         if (journal_inum && journal_dev) {
3877                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3878                        "and inode journals!");
3879                 return -EINVAL;
3880         }
3881
3882         if (journal_inum) {
3883                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3884                         return -EINVAL;
3885         } else {
3886                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3887                         return -EINVAL;
3888         }
3889
3890         if (!(journal->j_flags & JBD2_BARRIER))
3891                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3892
3893         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3894                 err = jbd2_journal_update_format(journal);
3895                 if (err)  {
3896                         ext4_msg(sb, KERN_ERR, "error updating journal");
3897                         jbd2_journal_destroy(journal);
3898                         return err;
3899                 }
3900         }
3901
3902         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3903                 err = jbd2_journal_wipe(journal, !really_read_only);
3904         if (!err) {
3905                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3906                 if (save)
3907                         memcpy(save, ((char *) es) +
3908                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3909                 err = jbd2_journal_load(journal);
3910                 if (save)
3911                         memcpy(((char *) es) + EXT4_S_ERR_START,
3912                                save, EXT4_S_ERR_LEN);
3913                 kfree(save);
3914         }
3915
3916         if (err) {
3917                 ext4_msg(sb, KERN_ERR, "error loading journal");
3918                 jbd2_journal_destroy(journal);
3919                 return err;
3920         }
3921
3922         EXT4_SB(sb)->s_journal = journal;
3923         ext4_clear_journal_err(sb, es);
3924
3925         if (!really_read_only && journal_devnum &&
3926             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3927                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3928
3929                 /* Make sure we flush the recovery flag to disk. */
3930                 ext4_commit_super(sb, 1);
3931         }
3932
3933         return 0;
3934 }
3935
3936 static int ext4_commit_super(struct super_block *sb, int sync)
3937 {
3938         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3939         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3940         int error = 0;
3941
3942         if (!sbh)
3943                 return error;
3944         if (buffer_write_io_error(sbh)) {
3945                 /*
3946                  * Oh, dear.  A previous attempt to write the
3947                  * superblock failed.  This could happen because the
3948                  * USB device was yanked out.  Or it could happen to
3949                  * be a transient write error and maybe the block will
3950                  * be remapped.  Nothing we can do but to retry the
3951                  * write and hope for the best.
3952                  */
3953                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3954                        "superblock detected");
3955                 clear_buffer_write_io_error(sbh);
3956                 set_buffer_uptodate(sbh);
3957         }
3958         /*
3959          * If the file system is mounted read-only, don't update the
3960          * superblock write time.  This avoids updating the superblock
3961          * write time when we are mounting the root file system
3962          * read/only but we need to replay the journal; at that point,
3963          * for people who are east of GMT and who make their clock
3964          * tick in localtime for Windows bug-for-bug compatibility,
3965          * the clock is set in the future, and this will cause e2fsck
3966          * to complain and force a full file system check.
3967          */
3968         if (!(sb->s_flags & MS_RDONLY))
3969                 es->s_wtime = cpu_to_le32(get_seconds());
3970         if (sb->s_bdev->bd_part)
3971                 es->s_kbytes_written =
3972                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3973                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3974                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3975         else
3976                 es->s_kbytes_written =
3977                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3978         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3979                                            &EXT4_SB(sb)->s_freeblocks_counter));
3980         es->s_free_inodes_count =
3981                 cpu_to_le32(percpu_counter_sum_positive(
3982                                 &EXT4_SB(sb)->s_freeinodes_counter));
3983         sb->s_dirt = 0;
3984         BUFFER_TRACE(sbh, "marking dirty");
3985         mark_buffer_dirty(sbh);
3986         if (sync) {
3987                 error = sync_dirty_buffer(sbh);
3988                 if (error)
3989                         return error;
3990
3991                 error = buffer_write_io_error(sbh);
3992                 if (error) {
3993                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3994                                "superblock");
3995                         clear_buffer_write_io_error(sbh);
3996                         set_buffer_uptodate(sbh);
3997                 }
3998         }
3999         return error;
4000 }
4001
4002 /*
4003  * Have we just finished recovery?  If so, and if we are mounting (or
4004  * remounting) the filesystem readonly, then we will end up with a
4005  * consistent fs on disk.  Record that fact.
4006  */
4007 static void ext4_mark_recovery_complete(struct super_block *sb,
4008                                         struct ext4_super_block *es)
4009 {
4010         journal_t *journal = EXT4_SB(sb)->s_journal;
4011
4012         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4013                 BUG_ON(journal != NULL);
4014                 return;
4015         }
4016         jbd2_journal_lock_updates(journal);
4017         if (jbd2_journal_flush(journal) < 0)
4018                 goto out;
4019
4020         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4021             sb->s_flags & MS_RDONLY) {
4022                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4023                 ext4_commit_super(sb, 1);
4024         }
4025
4026 out:
4027         jbd2_journal_unlock_updates(journal);
4028 }
4029
4030 /*
4031  * If we are mounting (or read-write remounting) a filesystem whose journal
4032  * has recorded an error from a previous lifetime, move that error to the
4033  * main filesystem now.
4034  */
4035 static void ext4_clear_journal_err(struct super_block *sb,
4036                                    struct ext4_super_block *es)
4037 {
4038         journal_t *journal;
4039         int j_errno;
4040         const char *errstr;
4041
4042         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4043
4044         journal = EXT4_SB(sb)->s_journal;
4045
4046         /*
4047          * Now check for any error status which may have been recorded in the
4048          * journal by a prior ext4_error() or ext4_abort()
4049          */
4050
4051         j_errno = jbd2_journal_errno(journal);
4052         if (j_errno) {
4053                 char nbuf[16];
4054
4055                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4056                 ext4_warning(sb, "Filesystem error recorded "
4057                              "from previous mount: %s", errstr);
4058                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4059
4060                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4061                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4062                 ext4_commit_super(sb, 1);
4063
4064                 jbd2_journal_clear_err(journal);
4065         }
4066 }
4067
4068 /*
4069  * Force the running and committing transactions to commit,
4070  * and wait on the commit.
4071  */
4072 int ext4_force_commit(struct super_block *sb)
4073 {
4074         journal_t *journal;
4075         int ret = 0;
4076
4077         if (sb->s_flags & MS_RDONLY)
4078                 return 0;
4079
4080         journal = EXT4_SB(sb)->s_journal;
4081         if (journal) {
4082                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4083                 ret = ext4_journal_force_commit(journal);
4084         }
4085
4086         return ret;
4087 }
4088
4089 static void ext4_write_super(struct super_block *sb)
4090 {
4091         lock_super(sb);
4092         ext4_commit_super(sb, 1);
4093         unlock_super(sb);
4094 }
4095
4096 static int ext4_sync_fs(struct super_block *sb, int wait)
4097 {
4098         int ret = 0;
4099         tid_t target;
4100         struct ext4_sb_info *sbi = EXT4_SB(sb);
4101
4102         trace_ext4_sync_fs(sb, wait);
4103         flush_workqueue(sbi->dio_unwritten_wq);
4104         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4105                 if (wait)
4106                         jbd2_log_wait_commit(sbi->s_journal, target);
4107         }
4108         return ret;
4109 }
4110
4111 /*
4112  * LVM calls this function before a (read-only) snapshot is created.  This
4113  * gives us a chance to flush the journal completely and mark the fs clean.
4114  */
4115 static int ext4_freeze(struct super_block *sb)
4116 {
4117         int error = 0;
4118         journal_t *journal;
4119
4120         if (sb->s_flags & MS_RDONLY)
4121                 return 0;
4122
4123         journal = EXT4_SB(sb)->s_journal;
4124
4125         /* Now we set up the journal barrier. */
4126         jbd2_journal_lock_updates(journal);
4127
4128         /*
4129          * Don't clear the needs_recovery flag if we failed to flush
4130          * the journal.
4131          */
4132         error = jbd2_journal_flush(journal);
4133         if (error < 0)
4134                 goto out;
4135
4136         /* Journal blocked and flushed, clear needs_recovery flag. */
4137         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4138         error = ext4_commit_super(sb, 1);
4139 out:
4140         /* we rely on s_frozen to stop further updates */
4141         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4142         return error;
4143 }
4144
4145 /*
4146  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4147  * flag here, even though the filesystem is not technically dirty yet.
4148  */
4149 static int ext4_unfreeze(struct super_block *sb)
4150 {
4151         if (sb->s_flags & MS_RDONLY)
4152                 return 0;
4153
4154         lock_super(sb);
4155         /* Reset the needs_recovery flag before the fs is unlocked. */
4156         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4157         ext4_commit_super(sb, 1);
4158         unlock_super(sb);
4159         return 0;
4160 }
4161
4162 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4163 {
4164         struct ext4_super_block *es;
4165         struct ext4_sb_info *sbi = EXT4_SB(sb);
4166         ext4_fsblk_t n_blocks_count = 0;
4167         unsigned long old_sb_flags;
4168         struct ext4_mount_options old_opts;
4169         int enable_quota = 0;
4170         ext4_group_t g;
4171         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4172         int err;
4173 #ifdef CONFIG_QUOTA
4174         int i;
4175 #endif
4176         char *orig_data = kstrdup(data, GFP_KERNEL);
4177
4178         /* Store the original options */
4179         lock_super(sb);
4180         old_sb_flags = sb->s_flags;
4181         old_opts.s_mount_opt = sbi->s_mount_opt;
4182         old_opts.s_resuid = sbi->s_resuid;
4183         old_opts.s_resgid = sbi->s_resgid;
4184         old_opts.s_commit_interval = sbi->s_commit_interval;
4185         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4186         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4187 #ifdef CONFIG_QUOTA
4188         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4189         for (i = 0; i < MAXQUOTAS; i++)
4190                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4191 #endif
4192         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4193                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4194
4195         /*
4196          * Allow the "check" option to be passed as a remount option.
4197          */
4198         if (!parse_options(data, sb, NULL, &journal_ioprio,
4199                            &n_blocks_count, 1)) {
4200                 err = -EINVAL;
4201                 goto restore_opts;
4202         }
4203
4204         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4205                 ext4_abort(sb, "Abort forced by user");
4206
4207         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4208                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4209
4210         es = sbi->s_es;
4211
4212         if (sbi->s_journal) {
4213                 ext4_init_journal_params(sb, sbi->s_journal);
4214                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4215         }
4216
4217         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4218                 n_blocks_count > ext4_blocks_count(es)) {
4219                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4220                         err = -EROFS;
4221                         goto restore_opts;
4222                 }
4223
4224                 if (*flags & MS_RDONLY) {
4225                         err = dquot_suspend(sb, -1);
4226                         if (err < 0)
4227                                 goto restore_opts;
4228
4229                         /*
4230                          * First of all, the unconditional stuff we have to do
4231                          * to disable replay of the journal when we next remount
4232                          */
4233                         sb->s_flags |= MS_RDONLY;
4234
4235                         /*
4236                          * OK, test if we are remounting a valid rw partition
4237                          * readonly, and if so set the rdonly flag and then
4238                          * mark the partition as valid again.
4239                          */
4240                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4241                             (sbi->s_mount_state & EXT4_VALID_FS))
4242                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4243
4244                         if (sbi->s_journal)
4245                                 ext4_mark_recovery_complete(sb, es);
4246                 } else {
4247                         /* Make sure we can mount this feature set readwrite */
4248                         if (!ext4_feature_set_ok(sb, 0)) {
4249                                 err = -EROFS;
4250                                 goto restore_opts;
4251                         }
4252                         /*
4253                          * Make sure the group descriptor checksums
4254                          * are sane.  If they aren't, refuse to remount r/w.
4255                          */
4256                         for (g = 0; g < sbi->s_groups_count; g++) {
4257                                 struct ext4_group_desc *gdp =
4258                                         ext4_get_group_desc(sb, g, NULL);
4259
4260                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4261                                         ext4_msg(sb, KERN_ERR,
4262                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4263                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4264                                                le16_to_cpu(gdp->bg_checksum));
4265                                         err = -EINVAL;
4266                                         goto restore_opts;
4267                                 }
4268                         }
4269
4270                         /*
4271                          * If we have an unprocessed orphan list hanging
4272                          * around from a previously readonly bdev mount,
4273                          * require a full umount/remount for now.
4274                          */
4275                         if (es->s_last_orphan) {
4276                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4277                                        "remount RDWR because of unprocessed "
4278                                        "orphan inode list.  Please "
4279                                        "umount/remount instead");
4280                                 err = -EINVAL;
4281                                 goto restore_opts;
4282                         }
4283
4284                         /*
4285                          * Mounting a RDONLY partition read-write, so reread
4286                          * and store the current valid flag.  (It may have
4287                          * been changed by e2fsck since we originally mounted
4288                          * the partition.)
4289                          */
4290                         if (sbi->s_journal)
4291                                 ext4_clear_journal_err(sb, es);
4292                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4293                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4294                                 goto restore_opts;
4295                         if (!ext4_setup_super(sb, es, 0))
4296                                 sb->s_flags &= ~MS_RDONLY;
4297                         enable_quota = 1;
4298                 }
4299         }
4300
4301         /*
4302          * Reinitialize lazy itable initialization thread based on
4303          * current settings
4304          */
4305         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4306                 ext4_unregister_li_request(sb);
4307         else {
4308                 ext4_group_t first_not_zeroed;
4309                 first_not_zeroed = ext4_has_uninit_itable(sb);
4310                 ext4_register_li_request(sb, first_not_zeroed);
4311         }
4312
4313         ext4_setup_system_zone(sb);
4314         if (sbi->s_journal == NULL)
4315                 ext4_commit_super(sb, 1);
4316
4317 #ifdef CONFIG_QUOTA
4318         /* Release old quota file names */
4319         for (i = 0; i < MAXQUOTAS; i++)
4320                 if (old_opts.s_qf_names[i] &&
4321                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4322                         kfree(old_opts.s_qf_names[i]);
4323 #endif
4324         unlock_super(sb);
4325         if (enable_quota)
4326                 dquot_resume(sb, -1);
4327
4328         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4329         kfree(orig_data);
4330         return 0;
4331
4332 restore_opts:
4333         sb->s_flags = old_sb_flags;
4334         sbi->s_mount_opt = old_opts.s_mount_opt;
4335         sbi->s_resuid = old_opts.s_resuid;
4336         sbi->s_resgid = old_opts.s_resgid;
4337         sbi->s_commit_interval = old_opts.s_commit_interval;
4338         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4339         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4340 #ifdef CONFIG_QUOTA
4341         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4342         for (i = 0; i < MAXQUOTAS; i++) {
4343                 if (sbi->s_qf_names[i] &&
4344                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4345                         kfree(sbi->s_qf_names[i]);
4346                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4347         }
4348 #endif
4349         unlock_super(sb);
4350         kfree(orig_data);
4351         return err;
4352 }
4353
4354 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4355 {
4356         struct super_block *sb = dentry->d_sb;
4357         struct ext4_sb_info *sbi = EXT4_SB(sb);
4358         struct ext4_super_block *es = sbi->s_es;
4359         u64 fsid;
4360
4361         if (test_opt(sb, MINIX_DF)) {
4362                 sbi->s_overhead_last = 0;
4363         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4364                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4365                 ext4_fsblk_t overhead = 0;
4366
4367                 /*
4368                  * Compute the overhead (FS structures).  This is constant
4369                  * for a given filesystem unless the number of block groups
4370                  * changes so we cache the previous value until it does.
4371                  */
4372
4373                 /*
4374                  * All of the blocks before first_data_block are
4375                  * overhead
4376                  */
4377                 overhead = le32_to_cpu(es->s_first_data_block);
4378
4379                 /*
4380                  * Add the overhead attributed to the superblock and
4381                  * block group descriptors.  If the sparse superblocks
4382                  * feature is turned on, then not all groups have this.
4383                  */
4384                 for (i = 0; i < ngroups; i++) {
4385                         overhead += ext4_bg_has_super(sb, i) +
4386                                 ext4_bg_num_gdb(sb, i);
4387                         cond_resched();
4388                 }
4389
4390                 /*
4391                  * Every block group has an inode bitmap, a block
4392                  * bitmap, and an inode table.
4393                  */
4394                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4395                 sbi->s_overhead_last = overhead;
4396                 smp_wmb();
4397                 sbi->s_blocks_last = ext4_blocks_count(es);
4398         }
4399
4400         buf->f_type = EXT4_SUPER_MAGIC;
4401         buf->f_bsize = sb->s_blocksize;
4402         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4403         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4404                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4405         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4406         if (buf->f_bfree < ext4_r_blocks_count(es))
4407                 buf->f_bavail = 0;
4408         buf->f_files = le32_to_cpu(es->s_inodes_count);
4409         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4410         buf->f_namelen = EXT4_NAME_LEN;
4411         fsid = le64_to_cpup((void *)es->s_uuid) ^
4412                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4413         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4414         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4415
4416         return 0;
4417 }
4418
4419 /* Helper function for writing quotas on sync - we need to start transaction
4420  * before quota file is locked for write. Otherwise the are possible deadlocks:
4421  * Process 1                         Process 2
4422  * ext4_create()                     quota_sync()
4423  *   jbd2_journal_start()                  write_dquot()
4424  *   dquot_initialize()                         down(dqio_mutex)
4425  *     down(dqio_mutex)                    jbd2_journal_start()
4426  *
4427  */
4428
4429 #ifdef CONFIG_QUOTA
4430
4431 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4432 {
4433         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4434 }
4435
4436 static int ext4_write_dquot(struct dquot *dquot)
4437 {
4438         int ret, err;
4439         handle_t *handle;
4440         struct inode *inode;
4441
4442         inode = dquot_to_inode(dquot);
4443         handle = ext4_journal_start(inode,
4444                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4445         if (IS_ERR(handle))
4446                 return PTR_ERR(handle);
4447         ret = dquot_commit(dquot);
4448         err = ext4_journal_stop(handle);
4449         if (!ret)
4450                 ret = err;
4451         return ret;
4452 }
4453
4454 static int ext4_acquire_dquot(struct dquot *dquot)
4455 {
4456         int ret, err;
4457         handle_t *handle;
4458
4459         handle = ext4_journal_start(dquot_to_inode(dquot),
4460                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4461         if (IS_ERR(handle))
4462                 return PTR_ERR(handle);
4463         ret = dquot_acquire(dquot);
4464         err = ext4_journal_stop(handle);
4465         if (!ret)
4466                 ret = err;
4467         return ret;
4468 }
4469
4470 static int ext4_release_dquot(struct dquot *dquot)
4471 {
4472         int ret, err;
4473         handle_t *handle;
4474
4475         handle = ext4_journal_start(dquot_to_inode(dquot),
4476                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4477         if (IS_ERR(handle)) {
4478                 /* Release dquot anyway to avoid endless cycle in dqput() */
4479                 dquot_release(dquot);
4480                 return PTR_ERR(handle);
4481         }
4482         ret = dquot_release(dquot);
4483         err = ext4_journal_stop(handle);
4484         if (!ret)
4485                 ret = err;
4486         return ret;
4487 }
4488
4489 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4490 {
4491         /* Are we journaling quotas? */
4492         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4493             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4494                 dquot_mark_dquot_dirty(dquot);
4495                 return ext4_write_dquot(dquot);
4496         } else {
4497                 return dquot_mark_dquot_dirty(dquot);
4498         }
4499 }
4500
4501 static int ext4_write_info(struct super_block *sb, int type)
4502 {
4503         int ret, err;
4504         handle_t *handle;
4505
4506         /* Data block + inode block */
4507         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4508         if (IS_ERR(handle))
4509                 return PTR_ERR(handle);
4510         ret = dquot_commit_info(sb, type);
4511         err = ext4_journal_stop(handle);
4512         if (!ret)
4513                 ret = err;
4514         return ret;
4515 }
4516
4517 /*
4518  * Turn on quotas during mount time - we need to find
4519  * the quota file and such...
4520  */
4521 static int ext4_quota_on_mount(struct super_block *sb, int type)
4522 {
4523         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4524                                         EXT4_SB(sb)->s_jquota_fmt, type);
4525 }
4526
4527 /*
4528  * Standard function to be called on quota_on
4529  */
4530 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4531                          char *name)
4532 {
4533         int err;
4534         struct path path;
4535
4536         if (!test_opt(sb, QUOTA))
4537                 return -EINVAL;
4538
4539         err = kern_path(name, LOOKUP_FOLLOW, &path);
4540         if (err)
4541                 return err;
4542
4543         /* Quotafile not on the same filesystem? */
4544         if (path.mnt->mnt_sb != sb) {
4545                 path_put(&path);
4546                 return -EXDEV;
4547         }
4548         /* Journaling quota? */
4549         if (EXT4_SB(sb)->s_qf_names[type]) {
4550                 /* Quotafile not in fs root? */
4551                 if (path.dentry->d_parent != sb->s_root)
4552                         ext4_msg(sb, KERN_WARNING,
4553                                 "Quota file not on filesystem root. "
4554                                 "Journaled quota will not work");
4555         }
4556
4557         /*
4558          * When we journal data on quota file, we have to flush journal to see
4559          * all updates to the file when we bypass pagecache...
4560          */
4561         if (EXT4_SB(sb)->s_journal &&
4562             ext4_should_journal_data(path.dentry->d_inode)) {
4563                 /*
4564                  * We don't need to lock updates but journal_flush() could
4565                  * otherwise be livelocked...
4566                  */
4567                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4568                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4569                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4570                 if (err) {
4571                         path_put(&path);
4572                         return err;
4573                 }
4574         }
4575
4576         err = dquot_quota_on_path(sb, type, format_id, &path);
4577         path_put(&path);
4578         return err;
4579 }
4580
4581 static int ext4_quota_off(struct super_block *sb, int type)
4582 {
4583         /* Force all delayed allocation blocks to be allocated.
4584          * Caller already holds s_umount sem */
4585         if (test_opt(sb, DELALLOC))
4586                 sync_filesystem(sb);
4587
4588         return dquot_quota_off(sb, type);
4589 }
4590
4591 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4592  * acquiring the locks... As quota files are never truncated and quota code
4593  * itself serializes the operations (and noone else should touch the files)
4594  * we don't have to be afraid of races */
4595 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4596                                size_t len, loff_t off)
4597 {
4598         struct inode *inode = sb_dqopt(sb)->files[type];
4599         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4600         int err = 0;
4601         int offset = off & (sb->s_blocksize - 1);
4602         int tocopy;
4603         size_t toread;
4604         struct buffer_head *bh;
4605         loff_t i_size = i_size_read(inode);
4606
4607         if (off > i_size)
4608                 return 0;
4609         if (off+len > i_size)
4610                 len = i_size-off;
4611         toread = len;
4612         while (toread > 0) {
4613                 tocopy = sb->s_blocksize - offset < toread ?
4614                                 sb->s_blocksize - offset : toread;
4615                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4616                 if (err)
4617                         return err;
4618                 if (!bh)        /* A hole? */
4619                         memset(data, 0, tocopy);
4620                 else
4621                         memcpy(data, bh->b_data+offset, tocopy);
4622                 brelse(bh);
4623                 offset = 0;
4624                 toread -= tocopy;
4625                 data += tocopy;
4626                 blk++;
4627         }
4628         return len;
4629 }
4630
4631 /* Write to quotafile (we know the transaction is already started and has
4632  * enough credits) */
4633 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4634                                 const char *data, size_t len, loff_t off)
4635 {
4636         struct inode *inode = sb_dqopt(sb)->files[type];
4637         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4638         int err = 0;
4639         int offset = off & (sb->s_blocksize - 1);
4640         struct buffer_head *bh;
4641         handle_t *handle = journal_current_handle();
4642
4643         if (EXT4_SB(sb)->s_journal && !handle) {
4644                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4645                         " cancelled because transaction is not started",
4646                         (unsigned long long)off, (unsigned long long)len);
4647                 return -EIO;
4648         }
4649         /*
4650          * Since we account only one data block in transaction credits,
4651          * then it is impossible to cross a block boundary.
4652          */
4653         if (sb->s_blocksize - offset < len) {
4654                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4655                         " cancelled because not block aligned",
4656                         (unsigned long long)off, (unsigned long long)len);
4657                 return -EIO;
4658         }
4659
4660         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4661         bh = ext4_bread(handle, inode, blk, 1, &err);
4662         if (!bh)
4663                 goto out;
4664         err = ext4_journal_get_write_access(handle, bh);
4665         if (err) {
4666                 brelse(bh);
4667                 goto out;
4668         }
4669         lock_buffer(bh);
4670         memcpy(bh->b_data+offset, data, len);
4671         flush_dcache_page(bh->b_page);
4672         unlock_buffer(bh);
4673         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4674         brelse(bh);
4675 out:
4676         if (err) {
4677                 mutex_unlock(&inode->i_mutex);
4678                 return err;
4679         }
4680         if (inode->i_size < off + len) {
4681                 i_size_write(inode, off + len);
4682                 EXT4_I(inode)->i_disksize = inode->i_size;
4683         }
4684         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4685         ext4_mark_inode_dirty(handle, inode);
4686         mutex_unlock(&inode->i_mutex);
4687         return len;
4688 }
4689
4690 #endif
4691
4692 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4693                        const char *dev_name, void *data)
4694 {
4695         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4696 }
4697
4698 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4699 static struct file_system_type ext2_fs_type = {
4700         .owner          = THIS_MODULE,
4701         .name           = "ext2",
4702         .mount          = ext4_mount,
4703         .kill_sb        = kill_block_super,
4704         .fs_flags       = FS_REQUIRES_DEV,
4705 };
4706
4707 static inline void register_as_ext2(void)
4708 {
4709         int err = register_filesystem(&ext2_fs_type);
4710         if (err)
4711                 printk(KERN_WARNING
4712                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4713 }
4714
4715 static inline void unregister_as_ext2(void)
4716 {
4717         unregister_filesystem(&ext2_fs_type);
4718 }
4719 MODULE_ALIAS("ext2");
4720 #else
4721 static inline void register_as_ext2(void) { }
4722 static inline void unregister_as_ext2(void) { }
4723 #endif
4724
4725 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4726 static inline void register_as_ext3(void)
4727 {
4728         int err = register_filesystem(&ext3_fs_type);
4729         if (err)
4730                 printk(KERN_WARNING
4731                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4732 }
4733
4734 static inline void unregister_as_ext3(void)
4735 {
4736         unregister_filesystem(&ext3_fs_type);
4737 }
4738 MODULE_ALIAS("ext3");
4739 #else
4740 static inline void register_as_ext3(void) { }
4741 static inline void unregister_as_ext3(void) { }
4742 #endif
4743
4744 static struct file_system_type ext4_fs_type = {
4745         .owner          = THIS_MODULE,
4746         .name           = "ext4",
4747         .mount          = ext4_mount,
4748         .kill_sb        = kill_block_super,
4749         .fs_flags       = FS_REQUIRES_DEV,
4750 };
4751
4752 int __init ext4_init_feat_adverts(void)
4753 {
4754         struct ext4_features *ef;
4755         int ret = -ENOMEM;
4756
4757         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4758         if (!ef)
4759                 goto out;
4760
4761         ef->f_kobj.kset = ext4_kset;
4762         init_completion(&ef->f_kobj_unregister);
4763         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4764                                    "features");
4765         if (ret) {
4766                 kfree(ef);
4767                 goto out;
4768         }
4769
4770         ext4_feat = ef;
4771         ret = 0;
4772 out:
4773         return ret;
4774 }
4775
4776 static int __init ext4_init_fs(void)
4777 {
4778         int err;
4779
4780         ext4_check_flag_values();
4781         err = ext4_init_pageio();
4782         if (err)
4783                 return err;
4784         err = ext4_init_system_zone();
4785         if (err)
4786                 goto out5;
4787         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4788         if (!ext4_kset)
4789                 goto out4;
4790         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4791
4792         err = ext4_init_feat_adverts();
4793
4794         err = ext4_init_mballoc();
4795         if (err)
4796                 goto out3;
4797
4798         err = ext4_init_xattr();
4799         if (err)
4800                 goto out2;
4801         err = init_inodecache();
4802         if (err)
4803                 goto out1;
4804         register_as_ext2();
4805         register_as_ext3();
4806         err = register_filesystem(&ext4_fs_type);
4807         if (err)
4808                 goto out;
4809
4810         ext4_li_info = NULL;
4811         mutex_init(&ext4_li_mtx);
4812         return 0;
4813 out:
4814         unregister_as_ext2();
4815         unregister_as_ext3();
4816         destroy_inodecache();
4817 out1:
4818         ext4_exit_xattr();
4819 out2:
4820         ext4_exit_mballoc();
4821 out3:
4822         kfree(ext4_feat);
4823         remove_proc_entry("fs/ext4", NULL);
4824         kset_unregister(ext4_kset);
4825 out4:
4826         ext4_exit_system_zone();
4827 out5:
4828         ext4_exit_pageio();
4829         return err;
4830 }
4831
4832 static void __exit ext4_exit_fs(void)
4833 {
4834         ext4_destroy_lazyinit_thread();
4835         unregister_as_ext2();
4836         unregister_as_ext3();
4837         unregister_filesystem(&ext4_fs_type);
4838         destroy_inodecache();
4839         ext4_exit_xattr();
4840         ext4_exit_mballoc();
4841         remove_proc_entry("fs/ext4", NULL);
4842         kset_unregister(ext4_kset);
4843         ext4_exit_system_zone();
4844         ext4_exit_pageio();
4845 }
4846
4847 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4848 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4849 MODULE_LICENSE("GPL");
4850 module_init(ext4_init_fs)
4851 module_exit(ext4_exit_fs)