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 static void ext4_clear_request_list(void);
  81
  82 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  83 static struct file_system_type ext3_fs_type = {
  84         .owner          = THIS_MODULE,
  85         .name           = "ext3",
  86         .mount          = ext4_mount,
  87         .kill_sb        = kill_block_super,
  88         .fs_flags       = FS_REQUIRES_DEV,
  89 };
  90 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  91 #else
  92 #define IS_EXT3_SB(sb) (0)
  93 #endif
  94
  95 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  96                                struct ext4_group_desc *bg)
  97 {
  98         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  99                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 100                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 101 }
 102
 103 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 104                                struct ext4_group_desc *bg)
 105 {
 106         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 107                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 108                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 109 }
 110
 111 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 112                               struct ext4_group_desc *bg)
 113 {
 114         return le32_to_cpu(bg->bg_inode_table_lo) |
 115                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 116                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 117 }
 118
 119 __u32 ext4_free_blks_count(struct super_block *sb,
 120                               struct ext4_group_desc *bg)
 121 {
 122         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 123                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 124                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 125 }
 126
 127 __u32 ext4_free_inodes_count(struct super_block *sb,
 128                               struct ext4_group_desc *bg)
 129 {
 130         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 131                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 132                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 133 }
 134
 135 __u32 ext4_used_dirs_count(struct super_block *sb,
 136                               struct ext4_group_desc *bg)
 137 {
 138         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 139                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 140                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 141 }
 142
 143 __u32 ext4_itable_unused_count(struct super_block *sb,
 144                               struct ext4_group_desc *bg)
 145 {
 146         return le16_to_cpu(bg->bg_itable_unused_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 149 }
 150
 151 void ext4_block_bitmap_set(struct super_block *sb,
 152                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 153 {
 154         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 155         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 156                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 157 }
 158
 159 void ext4_inode_bitmap_set(struct super_block *sb,
 160                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 161 {
 162         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 163         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 164                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 165 }
 166
 167 void ext4_inode_table_set(struct super_block *sb,
 168                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 169 {
 170         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 171         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 172                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 173 }
 174
 175 void ext4_free_blks_set(struct super_block *sb,
 176                           struct ext4_group_desc *bg, __u32 count)
 177 {
 178         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 179         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 180                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 181 }
 182
 183 void ext4_free_inodes_set(struct super_block *sb,
 184                           struct ext4_group_desc *bg, __u32 count)
 185 {
 186         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 187         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 188                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 189 }
 190
 191 void ext4_used_dirs_set(struct super_block *sb,
 192                           struct ext4_group_desc *bg, __u32 count)
 193 {
 194         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 195         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 196                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 197 }
 198
 199 void ext4_itable_unused_set(struct super_block *sb,
 200                           struct ext4_group_desc *bg, __u32 count)
 201 {
 202         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 205 }
 206
 207
 208 /* Just increment the non-pointer handle value */
 209 static handle_t *ext4_get_nojournal(void)
 210 {
 211         handle_t *handle = current->journal_info;
 212         unsigned long ref_cnt = (unsigned long)handle;
 213
 214         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 215
 216         ref_cnt++;
 217         handle = (handle_t *)ref_cnt;
 218
 219         current->journal_info = handle;
 220         return handle;
 221 }
 222
 223
 224 /* Decrement the non-pointer handle value */
 225 static void ext4_put_nojournal(handle_t *handle)
 226 {
 227         unsigned long ref_cnt = (unsigned long)handle;
 228
 229         BUG_ON(ref_cnt == 0);
 230
 231         ref_cnt--;
 232         handle = (handle_t *)ref_cnt;
 233
 234         current->journal_info = handle;
 235 }
 236
 237 /*
 238  * Wrappers for jbd2_journal_start/end.
 239  *
 240  * The only special thing we need to do here is to make sure that all
 241  * journal_end calls result in the superblock being marked dirty, so
 242  * that sync() will call the filesystem's write_super callback if
 243  * appropriate.
 244  */
 245 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 246 {
 247         journal_t *journal;
 248
 249         if (sb->s_flags & MS_RDONLY)
 250                 return ERR_PTR(-EROFS);
 251
 252         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 253         /* Special case here: if the journal has aborted behind our
 254          * backs (eg. EIO in the commit thread), then we still need to
 255          * take the FS itself readonly cleanly. */
 256         journal = EXT4_SB(sb)->s_journal;
 257         if (journal) {
 258                 if (is_journal_aborted(journal)) {
 259                         ext4_abort(sb, "Detected aborted journal");
 260                         return ERR_PTR(-EROFS);
 261                 }
 262                 return jbd2_journal_start(journal, nblocks);
 263         }
 264         return ext4_get_nojournal();
 265 }
 266
 267 /*
 268  * The only special thing we need to do here is to make sure that all
 269  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 270  * that sync() will call the filesystem's write_super callback if
 271  * appropriate.
 272  */
 273 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 274 {
 275         struct super_block *sb;
 276         int err;
 277         int rc;
 278
 279         if (!ext4_handle_valid(handle)) {
 280                 ext4_put_nojournal(handle);
 281                 return 0;
 282         }
 283         sb = handle->h_transaction->t_journal->j_private;
 284         err = handle->h_err;
 285         rc = jbd2_journal_stop(handle);
 286
 287         if (!err)
 288                 err = rc;
 289         if (err)
 290                 __ext4_std_error(sb, where, line, err);
 291         return err;
 292 }
 293
 294 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 295                                const char *err_fn, struct buffer_head *bh,
 296                                handle_t *handle, int err)
 297 {
 298         char nbuf[16];
 299         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 300
 301         BUG_ON(!ext4_handle_valid(handle));
 302
 303         if (bh)
 304                 BUFFER_TRACE(bh, "abort");
 305
 306         if (!handle->h_err)
 307                 handle->h_err = err;
 308
 309         if (is_handle_aborted(handle))
 310                 return;
 311
 312         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 313                caller, line, errstr, err_fn);
 314
 315         jbd2_journal_abort_handle(handle);
 316 }
 317
 318 static void __save_error_info(struct super_block *sb, const char *func,
 319                             unsigned int line)
 320 {
 321         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 322
 323         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 325         es->s_last_error_time = cpu_to_le32(get_seconds());
 326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 327         es->s_last_error_line = cpu_to_le32(line);
 328         if (!es->s_first_error_time) {
 329                 es->s_first_error_time = es->s_last_error_time;
 330                 strncpy(es->s_first_error_func, func,
 331                         sizeof(es->s_first_error_func));
 332                 es->s_first_error_line = cpu_to_le32(line);
 333                 es->s_first_error_ino = es->s_last_error_ino;
 334                 es->s_first_error_block = es->s_last_error_block;
 335         }
 336         /*
 337          * Start the daily error reporting function if it hasn't been
 338          * started already
 339          */
 340         if (!es->s_error_count)
 341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 342         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 343 }
 344
 345 static void save_error_info(struct super_block *sb, const char *func,
 346                             unsigned int line)
 347 {
 348         __save_error_info(sb, func, line);
 349         ext4_commit_super(sb, 1);
 350 }
 351
 352
 353 /* Deal with the reporting of failure conditions on a filesystem such as
 354  * inconsistencies detected or read IO failures.
 355  *
 356  * On ext2, we can store the error state of the filesystem in the
 357  * superblock.  That is not possible on ext4, because we may have other
 358  * write ordering constraints on the superblock which prevent us from
 359  * writing it out straight away; and given that the journal is about to
 360  * be aborted, we can't rely on the current, or future, transactions to
 361  * write out the superblock safely.
 362  *
 363  * We'll just use the jbd2_journal_abort() error code to record an error in
 364  * the journal instead.  On recovery, the journal will complain about
 365  * that error until we've noted it down and cleared it.
 366  */
 367
 368 static void ext4_handle_error(struct super_block *sb)
 369 {
 370         if (sb->s_flags & MS_RDONLY)
 371                 return;
 372
 373         if (!test_opt(sb, ERRORS_CONT)) {
 374                 journal_t *journal = EXT4_SB(sb)->s_journal;
 375
 376                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 377                 if (journal)
 378                         jbd2_journal_abort(journal, -EIO);
 379         }
 380         if (test_opt(sb, ERRORS_RO)) {
 381                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 382                 sb->s_flags |= MS_RDONLY;
 383         }
 384         if (test_opt(sb, ERRORS_PANIC))
 385                 panic("EXT4-fs (device %s): panic forced after error\n",
 386                         sb->s_id);
 387 }
 388
 389 void __ext4_error(struct super_block *sb, const char *function,
 390                   unsigned int line, const char *fmt, ...)
 391 {
 392         struct va_format vaf;
 393         va_list args;
 394
 395         va_start(args, fmt);
 396         vaf.fmt = fmt;
 397         vaf.va = &args;
 398         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 399                sb->s_id, function, line, current->comm, &vaf);
 400         va_end(args);
 401
 402         ext4_handle_error(sb);
 403 }
 404
 405 void ext4_error_inode(struct inode *inode, const char *function,
 406                       unsigned int line, ext4_fsblk_t block,
 407                       const char *fmt, ...)
 408 {
 409         va_list args;
 410         struct va_format vaf;
 411         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 412
 413         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 414         es->s_last_error_block = cpu_to_le64(block);
 415         save_error_info(inode->i_sb, function, line);
 416         va_start(args, fmt);
 417         vaf.fmt = fmt;
 418         vaf.va = &args;
 419         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 420                inode->i_sb->s_id, function, line, inode->i_ino);
 421         if (block)
 422                 printk(KERN_CONT "block %llu: ", block);
 423         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 424         va_end(args);
 425
 426         ext4_handle_error(inode->i_sb);
 427 }
 428
 429 void ext4_error_file(struct file *file, const char *function,
 430                      unsigned int line, ext4_fsblk_t block,
 431                      const char *fmt, ...)
 432 {
 433         va_list args;
 434         struct va_format vaf;
 435         struct ext4_super_block *es;
 436         struct inode *inode = file->f_dentry->d_inode;
 437         char pathname[80], *path;
 438
 439         es = EXT4_SB(inode->i_sb)->s_es;
 440         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 441         save_error_info(inode->i_sb, function, line);
 442         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 443         if (IS_ERR(path))
 444                 path = "(unknown)";
 445         printk(KERN_CRIT
 446                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 447                inode->i_sb->s_id, function, line, inode->i_ino);
 448         if (block)
 449                 printk(KERN_CONT "block %llu: ", block);
 450         va_start(args, fmt);
 451         vaf.fmt = fmt;
 452         vaf.va = &args;
 453         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 454         va_end(args);
 455
 456         ext4_handle_error(inode->i_sb);
 457 }
 458
 459 static const char *ext4_decode_error(struct super_block *sb, int errno,
 460                                      char nbuf[16])
 461 {
 462         char *errstr = NULL;
 463
 464         switch (errno) {
 465         case -EIO:
 466                 errstr = "IO failure";
 467                 break;
 468         case -ENOMEM:
 469                 errstr = "Out of memory";
 470                 break;
 471         case -EROFS:
 472                 if (!sb || (EXT4_SB(sb)->s_journal &&
 473                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 474                         errstr = "Journal has aborted";
 475                 else
 476                         errstr = "Readonly filesystem";
 477                 break;
 478         default:
 479                 /* If the caller passed in an extra buffer for unknown
 480                  * errors, textualise them now.  Else we just return
 481                  * NULL. */
 482                 if (nbuf) {
 483                         /* Check for truncated error codes... */
 484                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 485                                 errstr = nbuf;
 486                 }
 487                 break;
 488         }
 489
 490         return errstr;
 491 }
 492
 493 /* __ext4_std_error decodes expected errors from journaling functions
 494  * automatically and invokes the appropriate error response.  */
 495
 496 void __ext4_std_error(struct super_block *sb, const char *function,
 497                       unsigned int line, int errno)
 498 {
 499         char nbuf[16];
 500         const char *errstr;
 501
 502         /* Special case: if the error is EROFS, and we're not already
 503          * inside a transaction, then there's really no point in logging
 504          * an error. */
 505         if (errno == -EROFS && journal_current_handle() == NULL &&
 506             (sb->s_flags & MS_RDONLY))
 507                 return;
 508
 509         errstr = ext4_decode_error(sb, errno, nbuf);
 510         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 511                sb->s_id, function, line, errstr);
 512         save_error_info(sb, function, line);
 513
 514         ext4_handle_error(sb);
 515 }
 516
 517 /*
 518  * ext4_abort is a much stronger failure handler than ext4_error.  The
 519  * abort function may be used to deal with unrecoverable failures such
 520  * as journal IO errors or ENOMEM at a critical moment in log management.
 521  *
 522  * We unconditionally force the filesystem into an ABORT|READONLY state,
 523  * unless the error response on the fs has been set to panic in which
 524  * case we take the easy way out and panic immediately.
 525  */
 526
 527 void __ext4_abort(struct super_block *sb, const char *function,
 528                 unsigned int line, const char *fmt, ...)
 529 {
 530         va_list args;
 531
 532         save_error_info(sb, function, line);
 533         va_start(args, fmt);
 534         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 535                function, line);
 536         vprintk(fmt, args);
 537         printk("\n");
 538         va_end(args);
 539
 540         if ((sb->s_flags & MS_RDONLY) == 0) {
 541                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 542                 sb->s_flags |= MS_RDONLY;
 543                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 544                 if (EXT4_SB(sb)->s_journal)
 545                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 546                 save_error_info(sb, function, line);
 547         }
 548         if (test_opt(sb, ERRORS_PANIC))
 549                 panic("EXT4-fs panic from previous error\n");
 550 }
 551
 552 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 553 {
 554         struct va_format vaf;
 555         va_list args;
 556
 557         va_start(args, fmt);
 558         vaf.fmt = fmt;
 559         vaf.va = &args;
 560         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 561         va_end(args);
 562 }
 563
 564 void __ext4_warning(struct super_block *sb, const char *function,
 565                     unsigned int line, const char *fmt, ...)
 566 {
 567         struct va_format vaf;
 568         va_list args;
 569
 570         va_start(args, fmt);
 571         vaf.fmt = fmt;
 572         vaf.va = &args;
 573         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 574                sb->s_id, function, line, &vaf);
 575         va_end(args);
 576 }
 577
 578 void __ext4_grp_locked_error(const char *function, unsigned int line,
 579                              struct super_block *sb, ext4_group_t grp,
 580                              unsigned long ino, ext4_fsblk_t block,
 581                              const char *fmt, ...)
 582 __releases(bitlock)
 583 __acquires(bitlock)
 584 {
 585         struct va_format vaf;
 586         va_list args;
 587         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 588
 589         es->s_last_error_ino = cpu_to_le32(ino);
 590         es->s_last_error_block = cpu_to_le64(block);
 591         __save_error_info(sb, function, line);
 592
 593         va_start(args, fmt);
 594
 595         vaf.fmt = fmt;
 596         vaf.va = &args;
 597         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u",
 598                sb->s_id, function, line, grp);
 599         if (ino)
 600                 printk(KERN_CONT "inode %lu: ", ino);
 601         if (block)
 602                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 603         printk(KERN_CONT "%pV\n", &vaf);
 604         va_end(args);
 605
 606         if (test_opt(sb, ERRORS_CONT)) {
 607                 ext4_commit_super(sb, 0);
 608                 return;
 609         }
 610
 611         ext4_unlock_group(sb, grp);
 612         ext4_handle_error(sb);
 613         /*
 614          * We only get here in the ERRORS_RO case; relocking the group
 615          * may be dangerous, but nothing bad will happen since the
 616          * filesystem will have already been marked read/only and the
 617          * journal has been aborted.  We return 1 as a hint to callers
 618          * who might what to use the return value from
 619          * ext4_grp_locked_error() to distinguish beween the
 620          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 621          * aggressively from the ext4 function in question, with a
 622          * more appropriate error code.
 623          */
 624         ext4_lock_group(sb, grp);
 625         return;
 626 }
 627
 628 void ext4_update_dynamic_rev(struct super_block *sb)
 629 {
 630         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 631
 632         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 633                 return;
 634
 635         ext4_warning(sb,
 636                      "updating to rev %d because of new feature flag, "
 637                      "running e2fsck is recommended",
 638                      EXT4_DYNAMIC_REV);
 639
 640         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 641         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 642         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 643         /* leave es->s_feature_*compat flags alone */
 644         /* es->s_uuid will be set by e2fsck if empty */
 645
 646         /*
 647          * The rest of the superblock fields should be zero, and if not it
 648          * means they are likely already in use, so leave them alone.  We
 649          * can leave it up to e2fsck to clean up any inconsistencies there.
 650          */
 651 }
 652
 653 /*
 654  * Open the external journal device
 655  */
 656 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 657 {
 658         struct block_device *bdev;
 659         char b[BDEVNAME_SIZE];
 660
 661         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 662         if (IS_ERR(bdev))
 663                 goto fail;
 664         return bdev;
 665
 666 fail:
 667         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 668                         __bdevname(dev, b), PTR_ERR(bdev));
 669         return NULL;
 670 }
 671
 672 /*
 673  * Release the journal device
 674  */
 675 static int ext4_blkdev_put(struct block_device *bdev)
 676 {
 677         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 678 }
 679
 680 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 681 {
 682         struct block_device *bdev;
 683         int ret = -ENODEV;
 684
 685         bdev = sbi->journal_bdev;
 686         if (bdev) {
 687                 ret = ext4_blkdev_put(bdev);
 688                 sbi->journal_bdev = NULL;
 689         }
 690         return ret;
 691 }
 692
 693 static inline struct inode *orphan_list_entry(struct list_head *l)
 694 {
 695         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 696 }
 697
 698 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 699 {
 700         struct list_head *l;
 701
 702         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 703                  le32_to_cpu(sbi->s_es->s_last_orphan));
 704
 705         printk(KERN_ERR "sb_info orphan list:\n");
 706         list_for_each(l, &sbi->s_orphan) {
 707                 struct inode *inode = orphan_list_entry(l);
 708                 printk(KERN_ERR "  "
 709                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 710                        inode->i_sb->s_id, inode->i_ino, inode,
 711                        inode->i_mode, inode->i_nlink,
 712                        NEXT_ORPHAN(inode));
 713         }
 714 }
 715
 716 static void ext4_put_super(struct super_block *sb)
 717 {
 718         struct ext4_sb_info *sbi = EXT4_SB(sb);
 719         struct ext4_super_block *es = sbi->s_es;
 720         int i, err;
 721
 722         ext4_unregister_li_request(sb);
 723         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 724
 725         flush_workqueue(sbi->dio_unwritten_wq);
 726         destroy_workqueue(sbi->dio_unwritten_wq);
 727
 728         lock_super(sb);
 729         if (sb->s_dirt)
 730                 ext4_commit_super(sb, 1);
 731
 732         if (sbi->s_journal) {
 733                 err = jbd2_journal_destroy(sbi->s_journal);
 734                 sbi->s_journal = NULL;
 735                 if (err < 0)
 736                         ext4_abort(sb, "Couldn't clean up the journal");
 737         }
 738
 739         del_timer(&sbi->s_err_report);
 740         ext4_release_system_zone(sb);
 741         ext4_mb_release(sb);
 742         ext4_ext_release(sb);
 743         ext4_xattr_put_super(sb);
 744
 745         if (!(sb->s_flags & MS_RDONLY)) {
 746                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 747                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 748                 ext4_commit_super(sb, 1);
 749         }
 750         if (sbi->s_proc) {
 751                 remove_proc_entry(sb->s_id, ext4_proc_root);
 752         }
 753         kobject_del(&sbi->s_kobj);
 754
 755         for (i = 0; i < sbi->s_gdb_count; i++)
 756                 brelse(sbi->s_group_desc[i]);
 757         kfree(sbi->s_group_desc);
 758         if (is_vmalloc_addr(sbi->s_flex_groups))
 759                 vfree(sbi->s_flex_groups);
 760         else
 761                 kfree(sbi->s_flex_groups);
 762         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 763         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 764         percpu_counter_destroy(&sbi->s_dirs_counter);
 765         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 766         brelse(sbi->s_sbh);
 767 #ifdef CONFIG_QUOTA
 768         for (i = 0; i < MAXQUOTAS; i++)
 769                 kfree(sbi->s_qf_names[i]);
 770 #endif
 771
 772         /* Debugging code just in case the in-memory inode orphan list
 773          * isn't empty.  The on-disk one can be non-empty if we've
 774          * detected an error and taken the fs readonly, but the
 775          * in-memory list had better be clean by this point. */
 776         if (!list_empty(&sbi->s_orphan))
 777                 dump_orphan_list(sb, sbi);
 778         J_ASSERT(list_empty(&sbi->s_orphan));
 779
 780         invalidate_bdev(sb->s_bdev);
 781         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 782                 /*
 783                  * Invalidate the journal device's buffers.  We don't want them
 784                  * floating about in memory - the physical journal device may
 785                  * hotswapped, and it breaks the `ro-after' testing code.
 786                  */
 787                 sync_blockdev(sbi->journal_bdev);
 788                 invalidate_bdev(sbi->journal_bdev);
 789                 ext4_blkdev_remove(sbi);
 790         }
 791         sb->s_fs_info = NULL;
 792         /*
 793          * Now that we are completely done shutting down the
 794          * superblock, we need to actually destroy the kobject.
 795          */
 796         unlock_super(sb);
 797         kobject_put(&sbi->s_kobj);
 798         wait_for_completion(&sbi->s_kobj_unregister);
 799         kfree(sbi->s_blockgroup_lock);
 800         kfree(sbi);
 801 }
 802
 803 static struct kmem_cache *ext4_inode_cachep;
 804
 805 /*
 806  * Called inside transaction, so use GFP_NOFS
 807  */
 808 static struct inode *ext4_alloc_inode(struct super_block *sb)
 809 {
 810         struct ext4_inode_info *ei;
 811
 812         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 813         if (!ei)
 814                 return NULL;
 815
 816         ei->vfs_inode.i_version = 1;
 817         ei->vfs_inode.i_data.writeback_index = 0;
 818         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 819         INIT_LIST_HEAD(&ei->i_prealloc_list);
 820         spin_lock_init(&ei->i_prealloc_lock);
 821         ei->i_reserved_data_blocks = 0;
 822         ei->i_reserved_meta_blocks = 0;
 823         ei->i_allocated_meta_blocks = 0;
 824         ei->i_da_metadata_calc_len = 0;
 825         spin_lock_init(&(ei->i_block_reservation_lock));
 826 #ifdef CONFIG_QUOTA
 827         ei->i_reserved_quota = 0;
 828 #endif
 829         ei->jinode = NULL;
 830         INIT_LIST_HEAD(&ei->i_completed_io_list);
 831         spin_lock_init(&ei->i_completed_io_lock);
 832         ei->cur_aio_dio = NULL;
 833         ei->i_sync_tid = 0;
 834         ei->i_datasync_tid = 0;
 835         atomic_set(&ei->i_ioend_count, 0);
 836         atomic_set(&ei->i_aiodio_unwritten, 0);
 837
 838         return &ei->vfs_inode;
 839 }
 840
 841 static int ext4_drop_inode(struct inode *inode)
 842 {
 843         int drop = generic_drop_inode(inode);
 844
 845         trace_ext4_drop_inode(inode, drop);
 846         return drop;
 847 }
 848
 849 static void ext4_i_callback(struct rcu_head *head)
 850 {
 851         struct inode *inode = container_of(head, struct inode, i_rcu);
 852         INIT_LIST_HEAD(&inode->i_dentry);
 853         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 854 }
 855
 856 static void ext4_destroy_inode(struct inode *inode)
 857 {
 858         ext4_ioend_wait(inode);
 859         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 860                 ext4_msg(inode->i_sb, KERN_ERR,
 861                          "Inode %lu (%p): orphan list check failed!",
 862                          inode->i_ino, EXT4_I(inode));
 863                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 864                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 865                                 true);
 866                 dump_stack();
 867         }
 868         call_rcu(&inode->i_rcu, ext4_i_callback);
 869 }
 870
 871 static void init_once(void *foo)
 872 {
 873         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 874
 875         INIT_LIST_HEAD(&ei->i_orphan);
 876 #ifdef CONFIG_EXT4_FS_XATTR
 877         init_rwsem(&ei->xattr_sem);
 878 #endif
 879         init_rwsem(&ei->i_data_sem);
 880         inode_init_once(&ei->vfs_inode);
 881 }
 882
 883 static int init_inodecache(void)
 884 {
 885         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 886                                              sizeof(struct ext4_inode_info),
 887                                              0, (SLAB_RECLAIM_ACCOUNT|
 888                                                 SLAB_MEM_SPREAD),
 889                                              init_once);
 890         if (ext4_inode_cachep == NULL)
 891                 return -ENOMEM;
 892         return 0;
 893 }
 894
 895 static void destroy_inodecache(void)
 896 {
 897         kmem_cache_destroy(ext4_inode_cachep);
 898 }
 899
 900 void ext4_clear_inode(struct inode *inode)
 901 {
 902         invalidate_inode_buffers(inode);
 903         end_writeback(inode);
 904         dquot_drop(inode);
 905         ext4_discard_preallocations(inode);
 906         if (EXT4_I(inode)->jinode) {
 907                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 908                                                EXT4_I(inode)->jinode);
 909                 jbd2_free_inode(EXT4_I(inode)->jinode);
 910                 EXT4_I(inode)->jinode = NULL;
 911         }
 912 }
 913
 914 static inline void ext4_show_quota_options(struct seq_file *seq,
 915                                            struct super_block *sb)
 916 {
 917 #if defined(CONFIG_QUOTA)
 918         struct ext4_sb_info *sbi = EXT4_SB(sb);
 919
 920         if (sbi->s_jquota_fmt) {
 921                 char *fmtname = "";
 922
 923                 switch (sbi->s_jquota_fmt) {
 924                 case QFMT_VFS_OLD:
 925                         fmtname = "vfsold";
 926                         break;
 927                 case QFMT_VFS_V0:
 928                         fmtname = "vfsv0";
 929                         break;
 930                 case QFMT_VFS_V1:
 931                         fmtname = "vfsv1";
 932                         break;
 933                 }
 934                 seq_printf(seq, ",jqfmt=%s", fmtname);
 935         }
 936
 937         if (sbi->s_qf_names[USRQUOTA])
 938                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 939
 940         if (sbi->s_qf_names[GRPQUOTA])
 941                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 942
 943         if (test_opt(sb, USRQUOTA))
 944                 seq_puts(seq, ",usrquota");
 945
 946         if (test_opt(sb, GRPQUOTA))
 947                 seq_puts(seq, ",grpquota");
 948 #endif
 949 }
 950
 951 /*
 952  * Show an option if
 953  *  - it's set to a non-default value OR
 954  *  - if the per-sb default is different from the global default
 955  */
 956 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 957 {
 958         int def_errors;
 959         unsigned long def_mount_opts;
 960         struct super_block *sb = vfs->mnt_sb;
 961         struct ext4_sb_info *sbi = EXT4_SB(sb);
 962         struct ext4_super_block *es = sbi->s_es;
 963
 964         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 965         def_errors     = le16_to_cpu(es->s_errors);
 966
 967         if (sbi->s_sb_block != 1)
 968                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 969         if (test_opt(sb, MINIX_DF))
 970                 seq_puts(seq, ",minixdf");
 971         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 972                 seq_puts(seq, ",grpid");
 973         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 974                 seq_puts(seq, ",nogrpid");
 975         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 976             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 977                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 978         }
 979         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 980             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 981                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 982         }
 983         if (test_opt(sb, ERRORS_RO)) {
 984                 if (def_errors == EXT4_ERRORS_PANIC ||
 985                     def_errors == EXT4_ERRORS_CONTINUE) {
 986                         seq_puts(seq, ",errors=remount-ro");
 987                 }
 988         }
 989         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 990                 seq_puts(seq, ",errors=continue");
 991         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 992                 seq_puts(seq, ",errors=panic");
 993         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 994                 seq_puts(seq, ",nouid32");
 995         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 996                 seq_puts(seq, ",debug");
 997         if (test_opt(sb, OLDALLOC))
 998                 seq_puts(seq, ",oldalloc");
 999 #ifdef CONFIG_EXT4_FS_XATTR
1000         if (test_opt(sb, XATTR_USER) &&
1001                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
1002                 seq_puts(seq, ",user_xattr");
1003         if (!test_opt(sb, XATTR_USER) &&
1004             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
1005                 seq_puts(seq, ",nouser_xattr");
1006         }
1007 #endif
1008 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1009         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1010                 seq_puts(seq, ",acl");
1011         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1012                 seq_puts(seq, ",noacl");
1013 #endif
1014         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1015                 seq_printf(seq, ",commit=%u",
1016                            (unsigned) (sbi->s_commit_interval / HZ));
1017         }
1018         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1019                 seq_printf(seq, ",min_batch_time=%u",
1020                            (unsigned) sbi->s_min_batch_time);
1021         }
1022         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1023                 seq_printf(seq, ",max_batch_time=%u",
1024                            (unsigned) sbi->s_min_batch_time);
1025         }
1026
1027         /*
1028          * We're changing the default of barrier mount option, so
1029          * let's always display its mount state so it's clear what its
1030          * status is.
1031          */
1032         seq_puts(seq, ",barrier=");
1033         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1034         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1035                 seq_puts(seq, ",journal_async_commit");
1036         else if (test_opt(sb, JOURNAL_CHECKSUM))
1037                 seq_puts(seq, ",journal_checksum");
1038         if (test_opt(sb, I_VERSION))
1039                 seq_puts(seq, ",i_version");
1040         if (!test_opt(sb, DELALLOC) &&
1041             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1042                 seq_puts(seq, ",nodelalloc");
1043
1044         if (test_opt(sb, MBLK_IO_SUBMIT))
1045                 seq_puts(seq, ",mblk_io_submit");
1046         if (sbi->s_stripe)
1047                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1048         /*
1049          * journal mode get enabled in different ways
1050          * So just print the value even if we didn't specify it
1051          */
1052         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1053                 seq_puts(seq, ",data=journal");
1054         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1055                 seq_puts(seq, ",data=ordered");
1056         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1057                 seq_puts(seq, ",data=writeback");
1058
1059         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1060                 seq_printf(seq, ",inode_readahead_blks=%u",
1061                            sbi->s_inode_readahead_blks);
1062
1063         if (test_opt(sb, DATA_ERR_ABORT))
1064                 seq_puts(seq, ",data_err=abort");
1065
1066         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1067                 seq_puts(seq, ",noauto_da_alloc");
1068
1069         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1070                 seq_puts(seq, ",discard");
1071
1072         if (test_opt(sb, NOLOAD))
1073                 seq_puts(seq, ",norecovery");
1074
1075         if (test_opt(sb, DIOREAD_NOLOCK))
1076                 seq_puts(seq, ",dioread_nolock");
1077
1078         if (test_opt(sb, BLOCK_VALIDITY) &&
1079             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1080                 seq_puts(seq, ",block_validity");
1081
1082         if (!test_opt(sb, INIT_INODE_TABLE))
1083                 seq_puts(seq, ",noinit_inode_table");
1084         else if (sbi->s_li_wait_mult)
1085                 seq_printf(seq, ",init_inode_table=%u",
1086                            (unsigned) sbi->s_li_wait_mult);
1087
1088         ext4_show_quota_options(seq, sb);
1089
1090         return 0;
1091 }
1092
1093 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1094                                         u64 ino, u32 generation)
1095 {
1096         struct inode *inode;
1097
1098         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1099                 return ERR_PTR(-ESTALE);
1100         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1101                 return ERR_PTR(-ESTALE);
1102
1103         /* iget isn't really right if the inode is currently unallocated!!
1104          *
1105          * ext4_read_inode will return a bad_inode if the inode had been
1106          * deleted, so we should be safe.
1107          *
1108          * Currently we don't know the generation for parent directory, so
1109          * a generation of 0 means "accept any"
1110          */
1111         inode = ext4_iget(sb, ino);
1112         if (IS_ERR(inode))
1113                 return ERR_CAST(inode);
1114         if (generation && inode->i_generation != generation) {
1115                 iput(inode);
1116                 return ERR_PTR(-ESTALE);
1117         }
1118
1119         return inode;
1120 }
1121
1122 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1123                                         int fh_len, int fh_type)
1124 {
1125         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1126                                     ext4_nfs_get_inode);
1127 }
1128
1129 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1130                                         int fh_len, int fh_type)
1131 {
1132         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1133                                     ext4_nfs_get_inode);
1134 }
1135
1136 /*
1137  * Try to release metadata pages (indirect blocks, directories) which are
1138  * mapped via the block device.  Since these pages could have journal heads
1139  * which would prevent try_to_free_buffers() from freeing them, we must use
1140  * jbd2 layer's try_to_free_buffers() function to release them.
1141  */
1142 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1143                                  gfp_t wait)
1144 {
1145         journal_t *journal = EXT4_SB(sb)->s_journal;
1146
1147         WARN_ON(PageChecked(page));
1148         if (!page_has_buffers(page))
1149                 return 0;
1150         if (journal)
1151                 return jbd2_journal_try_to_free_buffers(journal, page,
1152                                                         wait & ~__GFP_WAIT);
1153         return try_to_free_buffers(page);
1154 }
1155
1156 #ifdef CONFIG_QUOTA
1157 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1158 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1159
1160 static int ext4_write_dquot(struct dquot *dquot);
1161 static int ext4_acquire_dquot(struct dquot *dquot);
1162 static int ext4_release_dquot(struct dquot *dquot);
1163 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1164 static int ext4_write_info(struct super_block *sb, int type);
1165 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1166                          struct path *path);
1167 static int ext4_quota_off(struct super_block *sb, int type);
1168 static int ext4_quota_on_mount(struct super_block *sb, int type);
1169 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1170                                size_t len, loff_t off);
1171 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1172                                 const char *data, size_t len, loff_t off);
1173
1174 static const struct dquot_operations ext4_quota_operations = {
1175 #ifdef CONFIG_QUOTA
1176         .get_reserved_space = ext4_get_reserved_space,
1177 #endif
1178         .write_dquot    = ext4_write_dquot,
1179         .acquire_dquot  = ext4_acquire_dquot,
1180         .release_dquot  = ext4_release_dquot,
1181         .mark_dirty     = ext4_mark_dquot_dirty,
1182         .write_info     = ext4_write_info,
1183         .alloc_dquot    = dquot_alloc,
1184         .destroy_dquot  = dquot_destroy,
1185 };
1186
1187 static const struct quotactl_ops ext4_qctl_operations = {
1188         .quota_on       = ext4_quota_on,
1189         .quota_off      = ext4_quota_off,
1190         .quota_sync     = dquot_quota_sync,
1191         .get_info       = dquot_get_dqinfo,
1192         .set_info       = dquot_set_dqinfo,
1193         .get_dqblk      = dquot_get_dqblk,
1194         .set_dqblk      = dquot_set_dqblk
1195 };
1196 #endif
1197
1198 static const struct super_operations ext4_sops = {
1199         .alloc_inode    = ext4_alloc_inode,
1200         .destroy_inode  = ext4_destroy_inode,
1201         .write_inode    = ext4_write_inode,
1202         .dirty_inode    = ext4_dirty_inode,
1203         .drop_inode     = ext4_drop_inode,
1204         .evict_inode    = ext4_evict_inode,
1205         .put_super      = ext4_put_super,
1206         .sync_fs        = ext4_sync_fs,
1207         .freeze_fs      = ext4_freeze,
1208         .unfreeze_fs    = ext4_unfreeze,
1209         .statfs         = ext4_statfs,
1210         .remount_fs     = ext4_remount,
1211         .show_options   = ext4_show_options,
1212 #ifdef CONFIG_QUOTA
1213         .quota_read     = ext4_quota_read,
1214         .quota_write    = ext4_quota_write,
1215 #endif
1216         .bdev_try_to_free_page = bdev_try_to_free_page,
1217 };
1218
1219 static const struct super_operations ext4_nojournal_sops = {
1220         .alloc_inode    = ext4_alloc_inode,
1221         .destroy_inode  = ext4_destroy_inode,
1222         .write_inode    = ext4_write_inode,
1223         .dirty_inode    = ext4_dirty_inode,
1224         .drop_inode     = ext4_drop_inode,
1225         .evict_inode    = ext4_evict_inode,
1226         .write_super    = ext4_write_super,
1227         .put_super      = ext4_put_super,
1228         .statfs         = ext4_statfs,
1229         .remount_fs     = ext4_remount,
1230         .show_options   = ext4_show_options,
1231 #ifdef CONFIG_QUOTA
1232         .quota_read     = ext4_quota_read,
1233         .quota_write    = ext4_quota_write,
1234 #endif
1235         .bdev_try_to_free_page = bdev_try_to_free_page,
1236 };
1237
1238 static const struct export_operations ext4_export_ops = {
1239         .fh_to_dentry = ext4_fh_to_dentry,
1240         .fh_to_parent = ext4_fh_to_parent,
1241         .get_parent = ext4_get_parent,
1242 };
1243
1244 enum {
1245         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1246         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1247         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1248         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1249         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1250         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1251         Opt_journal_update, Opt_journal_dev,
1252         Opt_journal_checksum, Opt_journal_async_commit,
1253         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1254         Opt_data_err_abort, Opt_data_err_ignore,
1255         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1256         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1257         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1258         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1259         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1260         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1261         Opt_inode_readahead_blks, Opt_journal_ioprio,
1262         Opt_dioread_nolock, Opt_dioread_lock,
1263         Opt_discard, Opt_nodiscard,
1264         Opt_init_inode_table, Opt_noinit_inode_table,
1265 };
1266
1267 static const match_table_t tokens = {
1268         {Opt_bsd_df, "bsddf"},
1269         {Opt_minix_df, "minixdf"},
1270         {Opt_grpid, "grpid"},
1271         {Opt_grpid, "bsdgroups"},
1272         {Opt_nogrpid, "nogrpid"},
1273         {Opt_nogrpid, "sysvgroups"},
1274         {Opt_resgid, "resgid=%u"},
1275         {Opt_resuid, "resuid=%u"},
1276         {Opt_sb, "sb=%u"},
1277         {Opt_err_cont, "errors=continue"},
1278         {Opt_err_panic, "errors=panic"},
1279         {Opt_err_ro, "errors=remount-ro"},
1280         {Opt_nouid32, "nouid32"},
1281         {Opt_debug, "debug"},
1282         {Opt_oldalloc, "oldalloc"},
1283         {Opt_orlov, "orlov"},
1284         {Opt_user_xattr, "user_xattr"},
1285         {Opt_nouser_xattr, "nouser_xattr"},
1286         {Opt_acl, "acl"},
1287         {Opt_noacl, "noacl"},
1288         {Opt_noload, "noload"},
1289         {Opt_noload, "norecovery"},
1290         {Opt_nobh, "nobh"},
1291         {Opt_bh, "bh"},
1292         {Opt_commit, "commit=%u"},
1293         {Opt_min_batch_time, "min_batch_time=%u"},
1294         {Opt_max_batch_time, "max_batch_time=%u"},
1295         {Opt_journal_update, "journal=update"},
1296         {Opt_journal_dev, "journal_dev=%u"},
1297         {Opt_journal_checksum, "journal_checksum"},
1298         {Opt_journal_async_commit, "journal_async_commit"},
1299         {Opt_abort, "abort"},
1300         {Opt_data_journal, "data=journal"},
1301         {Opt_data_ordered, "data=ordered"},
1302         {Opt_data_writeback, "data=writeback"},
1303         {Opt_data_err_abort, "data_err=abort"},
1304         {Opt_data_err_ignore, "data_err=ignore"},
1305         {Opt_offusrjquota, "usrjquota="},
1306         {Opt_usrjquota, "usrjquota=%s"},
1307         {Opt_offgrpjquota, "grpjquota="},
1308         {Opt_grpjquota, "grpjquota=%s"},
1309         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1310         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1311         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1312         {Opt_grpquota, "grpquota"},
1313         {Opt_noquota, "noquota"},
1314         {Opt_quota, "quota"},
1315         {Opt_usrquota, "usrquota"},
1316         {Opt_barrier, "barrier=%u"},
1317         {Opt_barrier, "barrier"},
1318         {Opt_nobarrier, "nobarrier"},
1319         {Opt_i_version, "i_version"},
1320         {Opt_stripe, "stripe=%u"},
1321         {Opt_resize, "resize"},
1322         {Opt_delalloc, "delalloc"},
1323         {Opt_nodelalloc, "nodelalloc"},
1324         {Opt_mblk_io_submit, "mblk_io_submit"},
1325         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1326         {Opt_block_validity, "block_validity"},
1327         {Opt_noblock_validity, "noblock_validity"},
1328         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1329         {Opt_journal_ioprio, "journal_ioprio=%u"},
1330         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1331         {Opt_auto_da_alloc, "auto_da_alloc"},
1332         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1333         {Opt_dioread_nolock, "dioread_nolock"},
1334         {Opt_dioread_lock, "dioread_lock"},
1335         {Opt_discard, "discard"},
1336         {Opt_nodiscard, "nodiscard"},
1337         {Opt_init_inode_table, "init_itable=%u"},
1338         {Opt_init_inode_table, "init_itable"},
1339         {Opt_noinit_inode_table, "noinit_itable"},
1340         {Opt_err, NULL},
1341 };
1342
1343 static ext4_fsblk_t get_sb_block(void **data)
1344 {
1345         ext4_fsblk_t    sb_block;
1346         char            *options = (char *) *data;
1347
1348         if (!options || strncmp(options, "sb=", 3) != 0)
1349                 return 1;       /* Default location */
1350
1351         options += 3;
1352         /* TODO: use simple_strtoll with >32bit ext4 */
1353         sb_block = simple_strtoul(options, &options, 0);
1354         if (*options && *options != ',') {
1355                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1356                        (char *) *data);
1357                 return 1;
1358         }
1359         if (*options == ',')
1360                 options++;
1361         *data = (void *) options;
1362
1363         return sb_block;
1364 }
1365
1366 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1367 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1368         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1369
1370 #ifdef CONFIG_QUOTA
1371 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1372 {
1373         struct ext4_sb_info *sbi = EXT4_SB(sb);
1374         char *qname;
1375
1376         if (sb_any_quota_loaded(sb) &&
1377                 !sbi->s_qf_names[qtype]) {
1378                 ext4_msg(sb, KERN_ERR,
1379                         "Cannot change journaled "
1380                         "quota options when quota turned on");
1381                 return 0;
1382         }
1383         qname = match_strdup(args);
1384         if (!qname) {
1385                 ext4_msg(sb, KERN_ERR,
1386                         "Not enough memory for storing quotafile name");
1387                 return 0;
1388         }
1389         if (sbi->s_qf_names[qtype] &&
1390                 strcmp(sbi->s_qf_names[qtype], qname)) {
1391                 ext4_msg(sb, KERN_ERR,
1392                         "%s quota file already specified", QTYPE2NAME(qtype));
1393                 kfree(qname);
1394                 return 0;
1395         }
1396         sbi->s_qf_names[qtype] = qname;
1397         if (strchr(sbi->s_qf_names[qtype], '/')) {
1398                 ext4_msg(sb, KERN_ERR,
1399                         "quotafile must be on filesystem root");
1400                 kfree(sbi->s_qf_names[qtype]);
1401                 sbi->s_qf_names[qtype] = NULL;
1402                 return 0;
1403         }
1404         set_opt(sb, QUOTA);
1405         return 1;
1406 }
1407
1408 static int clear_qf_name(struct super_block *sb, int qtype)
1409 {
1410
1411         struct ext4_sb_info *sbi = EXT4_SB(sb);
1412
1413         if (sb_any_quota_loaded(sb) &&
1414                 sbi->s_qf_names[qtype]) {
1415                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1416                         " when quota turned on");
1417                 return 0;
1418         }
1419         /*
1420          * The space will be released later when all options are confirmed
1421          * to be correct
1422          */
1423         sbi->s_qf_names[qtype] = NULL;
1424         return 1;
1425 }
1426 #endif
1427
1428 static int parse_options(char *options, struct super_block *sb,
1429                          unsigned long *journal_devnum,
1430                          unsigned int *journal_ioprio,
1431                          ext4_fsblk_t *n_blocks_count, int is_remount)
1432 {
1433         struct ext4_sb_info *sbi = EXT4_SB(sb);
1434         char *p;
1435         substring_t args[MAX_OPT_ARGS];
1436         int data_opt = 0;
1437         int option;
1438 #ifdef CONFIG_QUOTA
1439         int qfmt;
1440 #endif
1441
1442         if (!options)
1443                 return 1;
1444
1445         while ((p = strsep(&options, ",")) != NULL) {
1446                 int token;
1447                 if (!*p)
1448                         continue;
1449
1450                 /*
1451                  * Initialize args struct so we know whether arg was
1452                  * found; some options take optional arguments.
1453                  */
1454                 args[0].to = args[0].from = 0;
1455                 token = match_token(p, tokens, args);
1456                 switch (token) {
1457                 case Opt_bsd_df:
1458                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1459                         clear_opt(sb, MINIX_DF);
1460                         break;
1461                 case Opt_minix_df:
1462                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1463                         set_opt(sb, MINIX_DF);
1464
1465                         break;
1466                 case Opt_grpid:
1467                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1468                         set_opt(sb, GRPID);
1469
1470                         break;
1471                 case Opt_nogrpid:
1472                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1473                         clear_opt(sb, GRPID);
1474
1475                         break;
1476                 case Opt_resuid:
1477                         if (match_int(&args[0], &option))
1478                                 return 0;
1479                         sbi->s_resuid = option;
1480                         break;
1481                 case Opt_resgid:
1482                         if (match_int(&args[0], &option))
1483                                 return 0;
1484                         sbi->s_resgid = option;
1485                         break;
1486                 case Opt_sb:
1487                         /* handled by get_sb_block() instead of here */
1488                         /* *sb_block = match_int(&args[0]); */
1489                         break;
1490                 case Opt_err_panic:
1491                         clear_opt(sb, ERRORS_CONT);
1492                         clear_opt(sb, ERRORS_RO);
1493                         set_opt(sb, ERRORS_PANIC);
1494                         break;
1495                 case Opt_err_ro:
1496                         clear_opt(sb, ERRORS_CONT);
1497                         clear_opt(sb, ERRORS_PANIC);
1498                         set_opt(sb, ERRORS_RO);
1499                         break;
1500                 case Opt_err_cont:
1501                         clear_opt(sb, ERRORS_RO);
1502                         clear_opt(sb, ERRORS_PANIC);
1503                         set_opt(sb, ERRORS_CONT);
1504                         break;
1505                 case Opt_nouid32:
1506                         set_opt(sb, NO_UID32);
1507                         break;
1508                 case Opt_debug:
1509                         set_opt(sb, DEBUG);
1510                         break;
1511                 case Opt_oldalloc:
1512                         set_opt(sb, OLDALLOC);
1513                         break;
1514                 case Opt_orlov:
1515                         clear_opt(sb, OLDALLOC);
1516                         break;
1517 #ifdef CONFIG_EXT4_FS_XATTR
1518                 case Opt_user_xattr:
1519                         set_opt(sb, XATTR_USER);
1520                         break;
1521                 case Opt_nouser_xattr:
1522                         clear_opt(sb, XATTR_USER);
1523                         break;
1524 #else
1525                 case Opt_user_xattr:
1526                 case Opt_nouser_xattr:
1527                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1528                         break;
1529 #endif
1530 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1531                 case Opt_acl:
1532                         set_opt(sb, POSIX_ACL);
1533                         break;
1534                 case Opt_noacl:
1535                         clear_opt(sb, POSIX_ACL);
1536                         break;
1537 #else
1538                 case Opt_acl:
1539                 case Opt_noacl:
1540                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1541                         break;
1542 #endif
1543                 case Opt_journal_update:
1544                         /* @@@ FIXME */
1545                         /* Eventually we will want to be able to create
1546                            a journal file here.  For now, only allow the
1547                            user to specify an existing inode to be the
1548                            journal file. */
1549                         if (is_remount) {
1550                                 ext4_msg(sb, KERN_ERR,
1551                                          "Cannot specify journal on remount");
1552                                 return 0;
1553                         }
1554                         set_opt(sb, UPDATE_JOURNAL);
1555                         break;
1556                 case Opt_journal_dev:
1557                         if (is_remount) {
1558                                 ext4_msg(sb, KERN_ERR,
1559                                         "Cannot specify journal on remount");
1560                                 return 0;
1561                         }
1562                         if (match_int(&args[0], &option))
1563                                 return 0;
1564                         *journal_devnum = option;
1565                         break;
1566                 case Opt_journal_checksum:
1567                         set_opt(sb, JOURNAL_CHECKSUM);
1568                         break;
1569                 case Opt_journal_async_commit:
1570                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1571                         set_opt(sb, JOURNAL_CHECKSUM);
1572                         break;
1573                 case Opt_noload:
1574                         set_opt(sb, NOLOAD);
1575                         break;
1576                 case Opt_commit:
1577                         if (match_int(&args[0], &option))
1578                                 return 0;
1579                         if (option < 0)
1580                                 return 0;
1581                         if (option == 0)
1582                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1583                         sbi->s_commit_interval = HZ * option;
1584                         break;
1585                 case Opt_max_batch_time:
1586                         if (match_int(&args[0], &option))
1587                                 return 0;
1588                         if (option < 0)
1589                                 return 0;
1590                         if (option == 0)
1591                                 option = EXT4_DEF_MAX_BATCH_TIME;
1592                         sbi->s_max_batch_time = option;
1593                         break;
1594                 case Opt_min_batch_time:
1595                         if (match_int(&args[0], &option))
1596                                 return 0;
1597                         if (option < 0)
1598                                 return 0;
1599                         sbi->s_min_batch_time = option;
1600                         break;
1601                 case Opt_data_journal:
1602                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1603                         goto datacheck;
1604                 case Opt_data_ordered:
1605                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1606                         goto datacheck;
1607                 case Opt_data_writeback:
1608                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1609                 datacheck:
1610                         if (is_remount) {
1611                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1612                                         ext4_msg(sb, KERN_ERR,
1613                                                 "Cannot change data mode on remount");
1614                                         return 0;
1615                                 }
1616                         } else {
1617                                 clear_opt(sb, DATA_FLAGS);
1618                                 sbi->s_mount_opt |= data_opt;
1619                         }
1620                         break;
1621                 case Opt_data_err_abort:
1622                         set_opt(sb, DATA_ERR_ABORT);
1623                         break;
1624                 case Opt_data_err_ignore:
1625                         clear_opt(sb, DATA_ERR_ABORT);
1626                         break;
1627 #ifdef CONFIG_QUOTA
1628                 case Opt_usrjquota:
1629                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1630                                 return 0;
1631                         break;
1632                 case Opt_grpjquota:
1633                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1634                                 return 0;
1635                         break;
1636                 case Opt_offusrjquota:
1637                         if (!clear_qf_name(sb, USRQUOTA))
1638                                 return 0;
1639                         break;
1640                 case Opt_offgrpjquota:
1641                         if (!clear_qf_name(sb, GRPQUOTA))
1642                                 return 0;
1643                         break;
1644
1645                 case Opt_jqfmt_vfsold:
1646                         qfmt = QFMT_VFS_OLD;
1647                         goto set_qf_format;
1648                 case Opt_jqfmt_vfsv0:
1649                         qfmt = QFMT_VFS_V0;
1650                         goto set_qf_format;
1651                 case Opt_jqfmt_vfsv1:
1652                         qfmt = QFMT_VFS_V1;
1653 set_qf_format:
1654                         if (sb_any_quota_loaded(sb) &&
1655                             sbi->s_jquota_fmt != qfmt) {
1656                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1657                                         "journaled quota options when "
1658                                         "quota turned on");
1659                                 return 0;
1660                         }
1661                         sbi->s_jquota_fmt = qfmt;
1662                         break;
1663                 case Opt_quota:
1664                 case Opt_usrquota:
1665                         set_opt(sb, QUOTA);
1666                         set_opt(sb, USRQUOTA);
1667                         break;
1668                 case Opt_grpquota:
1669                         set_opt(sb, QUOTA);
1670                         set_opt(sb, GRPQUOTA);
1671                         break;
1672                 case Opt_noquota:
1673                         if (sb_any_quota_loaded(sb)) {
1674                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1675                                         "options when quota turned on");
1676                                 return 0;
1677                         }
1678                         clear_opt(sb, QUOTA);
1679                         clear_opt(sb, USRQUOTA);
1680                         clear_opt(sb, GRPQUOTA);
1681                         break;
1682 #else
1683                 case Opt_quota:
1684                 case Opt_usrquota:
1685                 case Opt_grpquota:
1686                         ext4_msg(sb, KERN_ERR,
1687                                 "quota options not supported");
1688                         break;
1689                 case Opt_usrjquota:
1690                 case Opt_grpjquota:
1691                 case Opt_offusrjquota:
1692                 case Opt_offgrpjquota:
1693                 case Opt_jqfmt_vfsold:
1694                 case Opt_jqfmt_vfsv0:
1695                 case Opt_jqfmt_vfsv1:
1696                         ext4_msg(sb, KERN_ERR,
1697                                 "journaled quota options not supported");
1698                         break;
1699                 case Opt_noquota:
1700                         break;
1701 #endif
1702                 case Opt_abort:
1703                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1704                         break;
1705                 case Opt_nobarrier:
1706                         clear_opt(sb, BARRIER);
1707                         break;
1708                 case Opt_barrier:
1709                         if (args[0].from) {
1710                                 if (match_int(&args[0], &option))
1711                                         return 0;
1712                         } else
1713                                 option = 1;     /* No argument, default to 1 */
1714                         if (option)
1715                                 set_opt(sb, BARRIER);
1716                         else
1717                                 clear_opt(sb, BARRIER);
1718                         break;
1719                 case Opt_ignore:
1720                         break;
1721                 case Opt_resize:
1722                         if (!is_remount) {
1723                                 ext4_msg(sb, KERN_ERR,
1724                                         "resize option only available "
1725                                         "for remount");
1726                                 return 0;
1727                         }
1728                         if (match_int(&args[0], &option) != 0)
1729                                 return 0;
1730                         *n_blocks_count = option;
1731                         break;
1732                 case Opt_nobh:
1733                         ext4_msg(sb, KERN_WARNING,
1734                                  "Ignoring deprecated nobh option");
1735                         break;
1736                 case Opt_bh:
1737                         ext4_msg(sb, KERN_WARNING,
1738                                  "Ignoring deprecated bh option");
1739                         break;
1740                 case Opt_i_version:
1741                         set_opt(sb, I_VERSION);
1742                         sb->s_flags |= MS_I_VERSION;
1743                         break;
1744                 case Opt_nodelalloc:
1745                         clear_opt(sb, DELALLOC);
1746                         break;
1747                 case Opt_mblk_io_submit:
1748                         set_opt(sb, MBLK_IO_SUBMIT);
1749                         break;
1750                 case Opt_nomblk_io_submit:
1751                         clear_opt(sb, MBLK_IO_SUBMIT);
1752                         break;
1753                 case Opt_stripe:
1754                         if (match_int(&args[0], &option))
1755                                 return 0;
1756                         if (option < 0)
1757                                 return 0;
1758                         sbi->s_stripe = option;
1759                         break;
1760                 case Opt_delalloc:
1761                         set_opt(sb, DELALLOC);
1762                         break;
1763                 case Opt_block_validity:
1764                         set_opt(sb, BLOCK_VALIDITY);
1765                         break;
1766                 case Opt_noblock_validity:
1767                         clear_opt(sb, BLOCK_VALIDITY);
1768                         break;
1769                 case Opt_inode_readahead_blks:
1770                         if (match_int(&args[0], &option))
1771                                 return 0;
1772                         if (option < 0 || option > (1 << 30))
1773                                 return 0;
1774                         if (!is_power_of_2(option)) {
1775                                 ext4_msg(sb, KERN_ERR,
1776                                          "EXT4-fs: inode_readahead_blks"
1777                                          " must be a power of 2");
1778                                 return 0;
1779                         }
1780                         sbi->s_inode_readahead_blks = option;
1781                         break;
1782                 case Opt_journal_ioprio:
1783                         if (match_int(&args[0], &option))
1784                                 return 0;
1785                         if (option < 0 || option > 7)
1786                                 break;
1787                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1788                                                             option);
1789                         break;
1790                 case Opt_noauto_da_alloc:
1791                         set_opt(sb, NO_AUTO_DA_ALLOC);
1792                         break;
1793                 case Opt_auto_da_alloc:
1794                         if (args[0].from) {
1795                                 if (match_int(&args[0], &option))
1796                                         return 0;
1797                         } else
1798                                 option = 1;     /* No argument, default to 1 */
1799                         if (option)
1800                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1801                         else
1802                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1803                         break;
1804                 case Opt_discard:
1805                         set_opt(sb, DISCARD);
1806                         break;
1807                 case Opt_nodiscard:
1808                         clear_opt(sb, DISCARD);
1809                         break;
1810                 case Opt_dioread_nolock:
1811                         set_opt(sb, DIOREAD_NOLOCK);
1812                         break;
1813                 case Opt_dioread_lock:
1814                         clear_opt(sb, DIOREAD_NOLOCK);
1815                         break;
1816                 case Opt_init_inode_table:
1817                         set_opt(sb, INIT_INODE_TABLE);
1818                         if (args[0].from) {
1819                                 if (match_int(&args[0], &option))
1820                                         return 0;
1821                         } else
1822                                 option = EXT4_DEF_LI_WAIT_MULT;
1823                         if (option < 0)
1824                                 return 0;
1825                         sbi->s_li_wait_mult = option;
1826                         break;
1827                 case Opt_noinit_inode_table:
1828                         clear_opt(sb, INIT_INODE_TABLE);
1829                         break;
1830                 default:
1831                         ext4_msg(sb, KERN_ERR,
1832                                "Unrecognized mount option \"%s\" "
1833                                "or missing value", p);
1834                         return 0;
1835                 }
1836         }
1837 #ifdef CONFIG_QUOTA
1838         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1839                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1840                         clear_opt(sb, USRQUOTA);
1841
1842                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1843                         clear_opt(sb, GRPQUOTA);
1844
1845                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1846                         ext4_msg(sb, KERN_ERR, "old and new quota "
1847                                         "format mixing");
1848                         return 0;
1849                 }
1850
1851                 if (!sbi->s_jquota_fmt) {
1852                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1853                                         "not specified");
1854                         return 0;
1855                 }
1856         } else {
1857                 if (sbi->s_jquota_fmt) {
1858                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1859                                         "specified with no journaling "
1860                                         "enabled");
1861                         return 0;
1862                 }
1863         }
1864 #endif
1865         return 1;
1866 }
1867
1868 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1869                             int read_only)
1870 {
1871         struct ext4_sb_info *sbi = EXT4_SB(sb);
1872         int res = 0;
1873
1874         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1875                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1876                          "forcing read-only mode");
1877                 res = MS_RDONLY;
1878         }
1879         if (read_only)
1880                 return res;
1881         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1882                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1883                          "running e2fsck is recommended");
1884         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1885                 ext4_msg(sb, KERN_WARNING,
1886                          "warning: mounting fs with errors, "
1887                          "running e2fsck is recommended");
1888         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1889                  le16_to_cpu(es->s_mnt_count) >=
1890                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "warning: maximal mount count reached, "
1893                          "running e2fsck is recommended");
1894         else if (le32_to_cpu(es->s_checkinterval) &&
1895                 (le32_to_cpu(es->s_lastcheck) +
1896                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1897                 ext4_msg(sb, KERN_WARNING,
1898                          "warning: checktime reached, "
1899                          "running e2fsck is recommended");
1900         if (!sbi->s_journal)
1901                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1902         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1903                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1904         le16_add_cpu(&es->s_mnt_count, 1);
1905         es->s_mtime = cpu_to_le32(get_seconds());
1906         ext4_update_dynamic_rev(sb);
1907         if (sbi->s_journal)
1908                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1909
1910         ext4_commit_super(sb, 1);
1911         if (test_opt(sb, DEBUG))
1912                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1913                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1914                         sb->s_blocksize,
1915                         sbi->s_groups_count,
1916                         EXT4_BLOCKS_PER_GROUP(sb),
1917                         EXT4_INODES_PER_GROUP(sb),
1918                         sbi->s_mount_opt, sbi->s_mount_opt2);
1919
1920         return res;
1921 }
1922
1923 static int ext4_fill_flex_info(struct super_block *sb)
1924 {
1925         struct ext4_sb_info *sbi = EXT4_SB(sb);
1926         struct ext4_group_desc *gdp = NULL;
1927         ext4_group_t flex_group_count;
1928         ext4_group_t flex_group;
1929         int groups_per_flex = 0;
1930         size_t size;
1931         int i;
1932
1933         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1934         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1935
1936         if (groups_per_flex < 2) {
1937                 sbi->s_log_groups_per_flex = 0;
1938                 return 1;
1939         }
1940
1941         /* We allocate both existing and potentially added groups */
1942         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1943                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1944                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1945         size = flex_group_count * sizeof(struct flex_groups);
1946         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1947         if (sbi->s_flex_groups == NULL) {
1948                 sbi->s_flex_groups = vzalloc(size);
1949                 if (sbi->s_flex_groups == NULL) {
1950                         ext4_msg(sb, KERN_ERR,
1951                                  "not enough memory for %u flex groups",
1952                                  flex_group_count);
1953                         goto failed;
1954                 }
1955         }
1956
1957         for (i = 0; i < sbi->s_groups_count; i++) {
1958                 gdp = ext4_get_group_desc(sb, i, NULL);
1959
1960                 flex_group = ext4_flex_group(sbi, i);
1961                 atomic_add(ext4_free_inodes_count(sb, gdp),
1962                            &sbi->s_flex_groups[flex_group].free_inodes);
1963                 atomic_add(ext4_free_blks_count(sb, gdp),
1964                            &sbi->s_flex_groups[flex_group].free_blocks);
1965                 atomic_add(ext4_used_dirs_count(sb, gdp),
1966                            &sbi->s_flex_groups[flex_group].used_dirs);
1967         }
1968
1969         return 1;
1970 failed:
1971         return 0;
1972 }
1973
1974 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1975                             struct ext4_group_desc *gdp)
1976 {
1977         __u16 crc = 0;
1978
1979         if (sbi->s_es->s_feature_ro_compat &
1980             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1981                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1982                 __le32 le_group = cpu_to_le32(block_group);
1983
1984                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1985                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1986                 crc = crc16(crc, (__u8 *)gdp, offset);
1987                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1988                 /* for checksum of struct ext4_group_desc do the rest...*/
1989                 if ((sbi->s_es->s_feature_incompat &
1990                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1991                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1992                         crc = crc16(crc, (__u8 *)gdp + offset,
1993                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1994                                         offset);
1995         }
1996
1997         return cpu_to_le16(crc);
1998 }
1999
2000 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2001                                 struct ext4_group_desc *gdp)
2002 {
2003         if ((sbi->s_es->s_feature_ro_compat &
2004              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2005             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2006                 return 0;
2007
2008         return 1;
2009 }
2010
2011 /* Called at mount-time, super-block is locked */
2012 static int ext4_check_descriptors(struct super_block *sb,
2013                                   ext4_group_t *first_not_zeroed)
2014 {
2015         struct ext4_sb_info *sbi = EXT4_SB(sb);
2016         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2017         ext4_fsblk_t last_block;
2018         ext4_fsblk_t block_bitmap;
2019         ext4_fsblk_t inode_bitmap;
2020         ext4_fsblk_t inode_table;
2021         int flexbg_flag = 0;
2022         ext4_group_t i, grp = sbi->s_groups_count;
2023
2024         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2025                 flexbg_flag = 1;
2026
2027         ext4_debug("Checking group descriptors");
2028
2029         for (i = 0; i < sbi->s_groups_count; i++) {
2030                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2031
2032                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2033                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2034                 else
2035                         last_block = first_block +
2036                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2037
2038                 if ((grp == sbi->s_groups_count) &&
2039                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2040                         grp = i;
2041
2042                 block_bitmap = ext4_block_bitmap(sb, gdp);
2043                 if (block_bitmap < first_block || block_bitmap > last_block) {
2044                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2045                                "Block bitmap for group %u not in group "
2046                                "(block %llu)!", i, block_bitmap);
2047                         return 0;
2048                 }
2049                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2050                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2051                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2052                                "Inode bitmap for group %u not in group "
2053                                "(block %llu)!", i, inode_bitmap);
2054                         return 0;
2055                 }
2056                 inode_table = ext4_inode_table(sb, gdp);
2057                 if (inode_table < first_block ||
2058                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2059                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2060                                "Inode table for group %u not in group "
2061                                "(block %llu)!", i, inode_table);
2062                         return 0;
2063                 }
2064                 ext4_lock_group(sb, i);
2065                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2066                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2067                                  "Checksum for group %u failed (%u!=%u)",
2068                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2069                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2070                         if (!(sb->s_flags & MS_RDONLY)) {
2071                                 ext4_unlock_group(sb, i);
2072                                 return 0;
2073                         }
2074                 }
2075                 ext4_unlock_group(sb, i);
2076                 if (!flexbg_flag)
2077                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2078         }
2079         if (NULL != first_not_zeroed)
2080                 *first_not_zeroed = grp;
2081
2082         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2083         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2084         return 1;
2085 }
2086
2087 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2088  * the superblock) which were deleted from all directories, but held open by
2089  * a process at the time of a crash.  We walk the list and try to delete these
2090  * inodes at recovery time (only with a read-write filesystem).
2091  *
2092  * In order to keep the orphan inode chain consistent during traversal (in
2093  * case of crash during recovery), we link each inode into the superblock
2094  * orphan list_head and handle it the same way as an inode deletion during
2095  * normal operation (which journals the operations for us).
2096  *
2097  * We only do an iget() and an iput() on each inode, which is very safe if we
2098  * accidentally point at an in-use or already deleted inode.  The worst that
2099  * can happen in this case is that we get a "bit already cleared" message from
2100  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2101  * e2fsck was run on this filesystem, and it must have already done the orphan
2102  * inode cleanup for us, so we can safely abort without any further action.
2103  */
2104 static void ext4_orphan_cleanup(struct super_block *sb,
2105                                 struct ext4_super_block *es)
2106 {
2107         unsigned int s_flags = sb->s_flags;
2108         int nr_orphans = 0, nr_truncates = 0;
2109 #ifdef CONFIG_QUOTA
2110         int i;
2111 #endif
2112         if (!es->s_last_orphan) {
2113                 jbd_debug(4, "no orphan inodes to clean up\n");
2114                 return;
2115         }
2116
2117         if (bdev_read_only(sb->s_bdev)) {
2118                 ext4_msg(sb, KERN_ERR, "write access "
2119                         "unavailable, skipping orphan cleanup");
2120                 return;
2121         }
2122
2123         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2124                 if (es->s_last_orphan)
2125                         jbd_debug(1, "Errors on filesystem, "
2126                                   "clearing orphan list.\n");
2127                 es->s_last_orphan = 0;
2128                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2129                 return;
2130         }
2131
2132         if (s_flags & MS_RDONLY) {
2133                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2134                 sb->s_flags &= ~MS_RDONLY;
2135         }
2136 #ifdef CONFIG_QUOTA
2137         /* Needed for iput() to work correctly and not trash data */
2138         sb->s_flags |= MS_ACTIVE;
2139         /* Turn on quotas so that they are updated correctly */
2140         for (i = 0; i < MAXQUOTAS; i++) {
2141                 if (EXT4_SB(sb)->s_qf_names[i]) {
2142                         int ret = ext4_quota_on_mount(sb, i);
2143                         if (ret < 0)
2144                                 ext4_msg(sb, KERN_ERR,
2145                                         "Cannot turn on journaled "
2146                                         "quota: error %d", ret);
2147                 }
2148         }
2149 #endif
2150
2151         while (es->s_last_orphan) {
2152                 struct inode *inode;
2153
2154                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2155                 if (IS_ERR(inode)) {
2156                         es->s_last_orphan = 0;
2157                         break;
2158                 }
2159
2160                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2161                 dquot_initialize(inode);
2162                 if (inode->i_nlink) {
2163                         ext4_msg(sb, KERN_DEBUG,
2164                                 "%s: truncating inode %lu to %lld bytes",
2165                                 __func__, inode->i_ino, inode->i_size);
2166                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2167                                   inode->i_ino, inode->i_size);
2168                         ext4_truncate(inode);
2169                         nr_truncates++;
2170                 } else {
2171                         ext4_msg(sb, KERN_DEBUG,
2172                                 "%s: deleting unreferenced inode %lu",
2173                                 __func__, inode->i_ino);
2174                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2175                                   inode->i_ino);
2176                         nr_orphans++;
2177                 }
2178                 iput(inode);  /* The delete magic happens here! */
2179         }
2180
2181 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2182
2183         if (nr_orphans)
2184                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2185                        PLURAL(nr_orphans));
2186         if (nr_truncates)
2187                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2188                        PLURAL(nr_truncates));
2189 #ifdef CONFIG_QUOTA
2190         /* Turn quotas off */
2191         for (i = 0; i < MAXQUOTAS; i++) {
2192                 if (sb_dqopt(sb)->files[i])
2193                         dquot_quota_off(sb, i);
2194         }
2195 #endif
2196         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2197 }
2198
2199 /*
2200  * Maximal extent format file size.
2201  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2202  * extent format containers, within a sector_t, and within i_blocks
2203  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2204  * so that won't be a limiting factor.
2205  *
2206  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2207  */
2208 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2209 {
2210         loff_t res;
2211         loff_t upper_limit = MAX_LFS_FILESIZE;
2212
2213         /* small i_blocks in vfs inode? */
2214         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2215                 /*
2216                  * CONFIG_LBDAF is not enabled implies the inode
2217                  * i_block represent total blocks in 512 bytes
2218                  * 32 == size of vfs inode i_blocks * 8
2219                  */
2220                 upper_limit = (1LL << 32) - 1;
2221
2222                 /* total blocks in file system block size */
2223                 upper_limit >>= (blkbits - 9);
2224                 upper_limit <<= blkbits;
2225         }
2226
2227         /* 32-bit extent-start container, ee_block */
2228         res = 1LL << 32;
2229         res <<= blkbits;
2230         res -= 1;
2231
2232         /* Sanity check against vm- & vfs- imposed limits */
2233         if (res > upper_limit)
2234                 res = upper_limit;
2235
2236         return res;
2237 }
2238
2239 /*
2240  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2241  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2242  * We need to be 1 filesystem block less than the 2^48 sector limit.
2243  */
2244 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2245 {
2246         loff_t res = EXT4_NDIR_BLOCKS;
2247         int meta_blocks;
2248         loff_t upper_limit;
2249         /* This is calculated to be the largest file size for a dense, block
2250          * mapped file such that the file's total number of 512-byte sectors,
2251          * including data and all indirect blocks, does not exceed (2^48 - 1).
2252          *
2253          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2254          * number of 512-byte sectors of the file.
2255          */
2256
2257         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2258                 /*
2259                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2260                  * the inode i_block field represents total file blocks in
2261                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2262                  */
2263                 upper_limit = (1LL << 32) - 1;
2264
2265                 /* total blocks in file system block size */
2266                 upper_limit >>= (bits - 9);
2267
2268         } else {
2269                 /*
2270                  * We use 48 bit ext4_inode i_blocks
2271                  * With EXT4_HUGE_FILE_FL set the i_blocks
2272                  * represent total number of blocks in
2273                  * file system block size
2274                  */
2275                 upper_limit = (1LL << 48) - 1;
2276
2277         }
2278
2279         /* indirect blocks */
2280         meta_blocks = 1;
2281         /* double indirect blocks */
2282         meta_blocks += 1 + (1LL << (bits-2));
2283         /* tripple indirect blocks */
2284         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2285
2286         upper_limit -= meta_blocks;
2287         upper_limit <<= bits;
2288
2289         res += 1LL << (bits-2);
2290         res += 1LL << (2*(bits-2));
2291         res += 1LL << (3*(bits-2));
2292         res <<= bits;
2293         if (res > upper_limit)
2294                 res = upper_limit;
2295
2296         if (res > MAX_LFS_FILESIZE)
2297                 res = MAX_LFS_FILESIZE;
2298
2299         return res;
2300 }
2301
2302 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2303                                    ext4_fsblk_t logical_sb_block, int nr)
2304 {
2305         struct ext4_sb_info *sbi = EXT4_SB(sb);
2306         ext4_group_t bg, first_meta_bg;
2307         int has_super = 0;
2308
2309         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2310
2311         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2312             nr < first_meta_bg)
2313                 return logical_sb_block + nr + 1;
2314         bg = sbi->s_desc_per_block * nr;
2315         if (ext4_bg_has_super(sb, bg))
2316                 has_super = 1;
2317
2318         return (has_super + ext4_group_first_block_no(sb, bg));
2319 }
2320
2321 /**
2322  * ext4_get_stripe_size: Get the stripe size.
2323  * @sbi: In memory super block info
2324  *
2325  * If we have specified it via mount option, then
2326  * use the mount option value. If the value specified at mount time is
2327  * greater than the blocks per group use the super block value.
2328  * If the super block value is greater than blocks per group return 0.
2329  * Allocator needs it be less than blocks per group.
2330  *
2331  */
2332 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2333 {
2334         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2335         unsigned long stripe_width =
2336                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2337
2338         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2339                 return sbi->s_stripe;
2340
2341         if (stripe_width <= sbi->s_blocks_per_group)
2342                 return stripe_width;
2343
2344         if (stride <= sbi->s_blocks_per_group)
2345                 return stride;
2346
2347         return 0;
2348 }
2349
2350 /* sysfs supprt */
2351
2352 struct ext4_attr {
2353         struct attribute attr;
2354         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2355         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2356                          const char *, size_t);
2357         int offset;
2358 };
2359
2360 static int parse_strtoul(const char *buf,
2361                 unsigned long max, unsigned long *value)
2362 {
2363         char *endp;
2364
2365         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2366         endp = skip_spaces(endp);
2367         if (*endp || *value > max)
2368                 return -EINVAL;
2369
2370         return 0;
2371 }
2372
2373 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2374                                               struct ext4_sb_info *sbi,
2375                                               char *buf)
2376 {
2377         return snprintf(buf, PAGE_SIZE, "%llu\n",
2378                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2379 }
2380
2381 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2382                                          struct ext4_sb_info *sbi, char *buf)
2383 {
2384         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2385
2386         if (!sb->s_bdev->bd_part)
2387                 return snprintf(buf, PAGE_SIZE, "0\n");
2388         return snprintf(buf, PAGE_SIZE, "%lu\n",
2389                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2390                          sbi->s_sectors_written_start) >> 1);
2391 }
2392
2393 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2394                                           struct ext4_sb_info *sbi, char *buf)
2395 {
2396         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2397
2398         if (!sb->s_bdev->bd_part)
2399                 return snprintf(buf, PAGE_SIZE, "0\n");
2400         return snprintf(buf, PAGE_SIZE, "%llu\n",
2401                         (unsigned long long)(sbi->s_kbytes_written +
2402                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2403                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2404 }
2405
2406 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2407                                           struct ext4_sb_info *sbi,
2408                                           const char *buf, size_t count)
2409 {
2410         unsigned long t;
2411
2412         if (parse_strtoul(buf, 0x40000000, &t))
2413                 return -EINVAL;
2414
2415         if (!is_power_of_2(t))
2416                 return -EINVAL;
2417
2418         sbi->s_inode_readahead_blks = t;
2419         return count;
2420 }
2421
2422 static ssize_t sbi_ui_show(struct ext4_attr *a,
2423                            struct ext4_sb_info *sbi, char *buf)
2424 {
2425         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2426
2427         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2428 }
2429
2430 static ssize_t sbi_ui_store(struct ext4_attr *a,
2431                             struct ext4_sb_info *sbi,
2432                             const char *buf, size_t count)
2433 {
2434         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2435         unsigned long t;
2436
2437         if (parse_strtoul(buf, 0xffffffff, &t))
2438                 return -EINVAL;
2439         *ui = t;
2440         return count;
2441 }
2442
2443 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2444 static struct ext4_attr ext4_attr_##_name = {                   \
2445         .attr = {.name = __stringify(_name), .mode = _mode },   \
2446         .show   = _show,                                        \
2447         .store  = _store,                                       \
2448         .offset = offsetof(struct ext4_sb_info, _elname),       \
2449 }
2450 #define EXT4_ATTR(name, mode, show, store) \
2451 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2452
2453 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2454 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2455 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2456 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2457         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2458 #define ATTR_LIST(name) &ext4_attr_##name.attr
2459
2460 EXT4_RO_ATTR(delayed_allocation_blocks);
2461 EXT4_RO_ATTR(session_write_kbytes);
2462 EXT4_RO_ATTR(lifetime_write_kbytes);
2463 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2464                  inode_readahead_blks_store, s_inode_readahead_blks);
2465 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2466 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2467 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2468 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2469 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2470 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2471 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2472 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2473
2474 static struct attribute *ext4_attrs[] = {
2475         ATTR_LIST(delayed_allocation_blocks),
2476         ATTR_LIST(session_write_kbytes),
2477         ATTR_LIST(lifetime_write_kbytes),
2478         ATTR_LIST(inode_readahead_blks),
2479         ATTR_LIST(inode_goal),
2480         ATTR_LIST(mb_stats),
2481         ATTR_LIST(mb_max_to_scan),
2482         ATTR_LIST(mb_min_to_scan),
2483         ATTR_LIST(mb_order2_req),
2484         ATTR_LIST(mb_stream_req),
2485         ATTR_LIST(mb_group_prealloc),
2486         ATTR_LIST(max_writeback_mb_bump),
2487         NULL,
2488 };
2489
2490 /* Features this copy of ext4 supports */
2491 EXT4_INFO_ATTR(lazy_itable_init);
2492 EXT4_INFO_ATTR(batched_discard);
2493
2494 static struct attribute *ext4_feat_attrs[] = {
2495         ATTR_LIST(lazy_itable_init),
2496         ATTR_LIST(batched_discard),
2497         NULL,
2498 };
2499
2500 static ssize_t ext4_attr_show(struct kobject *kobj,
2501                               struct attribute *attr, char *buf)
2502 {
2503         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2504                                                 s_kobj);
2505         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2506
2507         return a->show ? a->show(a, sbi, buf) : 0;
2508 }
2509
2510 static ssize_t ext4_attr_store(struct kobject *kobj,
2511                                struct attribute *attr,
2512                                const char *buf, size_t len)
2513 {
2514         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2515                                                 s_kobj);
2516         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2517
2518         return a->store ? a->store(a, sbi, buf, len) : 0;
2519 }
2520
2521 static void ext4_sb_release(struct kobject *kobj)
2522 {
2523         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2524                                                 s_kobj);
2525         complete(&sbi->s_kobj_unregister);
2526 }
2527
2528 static const struct sysfs_ops ext4_attr_ops = {
2529         .show   = ext4_attr_show,
2530         .store  = ext4_attr_store,
2531 };
2532
2533 static struct kobj_type ext4_ktype = {
2534         .default_attrs  = ext4_attrs,
2535         .sysfs_ops      = &ext4_attr_ops,
2536         .release        = ext4_sb_release,
2537 };
2538
2539 static void ext4_feat_release(struct kobject *kobj)
2540 {
2541         complete(&ext4_feat->f_kobj_unregister);
2542 }
2543
2544 static struct kobj_type ext4_feat_ktype = {
2545         .default_attrs  = ext4_feat_attrs,
2546         .sysfs_ops      = &ext4_attr_ops,
2547         .release        = ext4_feat_release,
2548 };
2549
2550 /*
2551  * Check whether this filesystem can be mounted based on
2552  * the features present and the RDONLY/RDWR mount requested.
2553  * Returns 1 if this filesystem can be mounted as requested,
2554  * 0 if it cannot be.
2555  */
2556 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2557 {
2558         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2559                 ext4_msg(sb, KERN_ERR,
2560                         "Couldn't mount because of "
2561                         "unsupported optional features (%x)",
2562                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2563                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2564                 return 0;
2565         }
2566
2567         if (readonly)
2568                 return 1;
2569
2570         /* Check that feature set is OK for a read-write mount */
2571         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2572                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2573                          "unsupported optional features (%x)",
2574                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2575                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2576                 return 0;
2577         }
2578         /*
2579          * Large file size enabled file system can only be mounted
2580          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2581          */
2582         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2583                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2584                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2585                                  "cannot be mounted RDWR without "
2586                                  "CONFIG_LBDAF");
2587                         return 0;
2588                 }
2589         }
2590         return 1;
2591 }
2592
2593 /*
2594  * This function is called once a day if we have errors logged
2595  * on the file system
2596  */
2597 static void print_daily_error_info(unsigned long arg)
2598 {
2599         struct super_block *sb = (struct super_block *) arg;
2600         struct ext4_sb_info *sbi;
2601         struct ext4_super_block *es;
2602
2603         sbi = EXT4_SB(sb);
2604         es = sbi->s_es;
2605
2606         if (es->s_error_count)
2607                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2608                          le32_to_cpu(es->s_error_count));
2609         if (es->s_first_error_time) {
2610                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2611                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2612                        (int) sizeof(es->s_first_error_func),
2613                        es->s_first_error_func,
2614                        le32_to_cpu(es->s_first_error_line));
2615                 if (es->s_first_error_ino)
2616                         printk(": inode %u",
2617                                le32_to_cpu(es->s_first_error_ino));
2618                 if (es->s_first_error_block)
2619                         printk(": block %llu", (unsigned long long)
2620                                le64_to_cpu(es->s_first_error_block));
2621                 printk("\n");
2622         }
2623         if (es->s_last_error_time) {
2624                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2625                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2626                        (int) sizeof(es->s_last_error_func),
2627                        es->s_last_error_func,
2628                        le32_to_cpu(es->s_last_error_line));
2629                 if (es->s_last_error_ino)
2630                         printk(": inode %u",
2631                                le32_to_cpu(es->s_last_error_ino));
2632                 if (es->s_last_error_block)
2633                         printk(": block %llu", (unsigned long long)
2634                                le64_to_cpu(es->s_last_error_block));
2635                 printk("\n");
2636         }
2637         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2638 }
2639
2640 static void ext4_lazyinode_timeout(unsigned long data)
2641 {
2642         struct task_struct *p = (struct task_struct *)data;
2643         wake_up_process(p);
2644 }
2645
2646 /* Find next suitable group and run ext4_init_inode_table */
2647 static int ext4_run_li_request(struct ext4_li_request *elr)
2648 {
2649         struct ext4_group_desc *gdp = NULL;
2650         ext4_group_t group, ngroups;
2651         struct super_block *sb;
2652         unsigned long timeout = 0;
2653         int ret = 0;
2654
2655         sb = elr->lr_super;
2656         ngroups = EXT4_SB(sb)->s_groups_count;
2657
2658         for (group = elr->lr_next_group; group < ngroups; group++) {
2659                 gdp = ext4_get_group_desc(sb, group, NULL);
2660                 if (!gdp) {
2661                         ret = 1;
2662                         break;
2663                 }
2664
2665                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2666                         break;
2667         }
2668
2669         if (group == ngroups)
2670                 ret = 1;
2671
2672         if (!ret) {
2673                 timeout = jiffies;
2674                 ret = ext4_init_inode_table(sb, group,
2675                                             elr->lr_timeout ? 0 : 1);
2676                 if (elr->lr_timeout == 0) {
2677                         timeout = jiffies - timeout;
2678                         if (elr->lr_sbi->s_li_wait_mult)
2679                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2680                         else
2681                                 timeout *= 20;
2682                         elr->lr_timeout = timeout;
2683                 }
2684                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2685                 elr->lr_next_group = group + 1;
2686         }
2687
2688         return ret;
2689 }
2690
2691 /*
2692  * Remove lr_request from the list_request and free the
2693  * request tructure. Should be called with li_list_mtx held
2694  */
2695 static void ext4_remove_li_request(struct ext4_li_request *elr)
2696 {
2697         struct ext4_sb_info *sbi;
2698
2699         if (!elr)
2700                 return;
2701
2702         sbi = elr->lr_sbi;
2703
2704         list_del(&elr->lr_request);
2705         sbi->s_li_request = NULL;
2706         kfree(elr);
2707 }
2708
2709 static void ext4_unregister_li_request(struct super_block *sb)
2710 {
2711         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2712
2713         if (!ext4_li_info)
2714                 return;
2715
2716         mutex_lock(&ext4_li_info->li_list_mtx);
2717         ext4_remove_li_request(elr);
2718         mutex_unlock(&ext4_li_info->li_list_mtx);
2719 }
2720
2721 static struct task_struct *ext4_lazyinit_task;
2722
2723 /*
2724  * This is the function where ext4lazyinit thread lives. It walks
2725  * through the request list searching for next scheduled filesystem.
2726  * When such a fs is found, run the lazy initialization request
2727  * (ext4_rn_li_request) and keep track of the time spend in this
2728  * function. Based on that time we compute next schedule time of
2729  * the request. When walking through the list is complete, compute
2730  * next waking time and put itself into sleep.
2731  */
2732 static int ext4_lazyinit_thread(void *arg)
2733 {
2734         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2735         struct list_head *pos, *n;
2736         struct ext4_li_request *elr;
2737         unsigned long next_wakeup;
2738         DEFINE_WAIT(wait);
2739
2740         BUG_ON(NULL == eli);
2741
2742         eli->li_timer.data = (unsigned long)current;
2743         eli->li_timer.function = ext4_lazyinode_timeout;
2744
2745         eli->li_task = current;
2746         wake_up(&eli->li_wait_task);
2747
2748 cont_thread:
2749         while (true) {
2750                 next_wakeup = MAX_JIFFY_OFFSET;
2751
2752                 mutex_lock(&eli->li_list_mtx);
2753                 if (list_empty(&eli->li_request_list)) {
2754                         mutex_unlock(&eli->li_list_mtx);
2755                         goto exit_thread;
2756                 }
2757
2758                 list_for_each_safe(pos, n, &eli->li_request_list) {
2759                         elr = list_entry(pos, struct ext4_li_request,
2760                                          lr_request);
2761
2762                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2763                                 if (ext4_run_li_request(elr) != 0) {
2764                                         /* error, remove the lazy_init job */
2765                                         ext4_remove_li_request(elr);
2766                                         continue;
2767                                 }
2768                         }
2769
2770                         if (time_before(elr->lr_next_sched, next_wakeup))
2771                                 next_wakeup = elr->lr_next_sched;
2772                 }
2773                 mutex_unlock(&eli->li_list_mtx);
2774
2775                 if (freezing(current))
2776                         refrigerator();
2777
2778                 if ((time_after_eq(jiffies, next_wakeup)) ||
2779                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2780                         cond_resched();
2781                         continue;
2782                 }
2783
2784                 eli->li_timer.expires = next_wakeup;
2785                 add_timer(&eli->li_timer);
2786                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2787                                 TASK_INTERRUPTIBLE);
2788                 if (time_before(jiffies, next_wakeup))
2789                         schedule();
2790                 finish_wait(&eli->li_wait_daemon, &wait);
2791                 if (kthread_should_stop()) {
2792                         ext4_clear_request_list();
2793                         goto exit_thread;
2794                 }
2795         }
2796
2797 exit_thread:
2798         /*
2799          * It looks like the request list is empty, but we need
2800          * to check it under the li_list_mtx lock, to prevent any
2801          * additions into it, and of course we should lock ext4_li_mtx
2802          * to atomically free the list and ext4_li_info, because at
2803          * this point another ext4 filesystem could be registering
2804          * new one.
2805          */
2806         mutex_lock(&ext4_li_mtx);
2807         mutex_lock(&eli->li_list_mtx);
2808         if (!list_empty(&eli->li_request_list)) {
2809                 mutex_unlock(&eli->li_list_mtx);
2810                 mutex_unlock(&ext4_li_mtx);
2811                 goto cont_thread;
2812         }
2813         mutex_unlock(&eli->li_list_mtx);
2814         del_timer_sync(&ext4_li_info->li_timer);
2815         eli->li_task = NULL;
2816         wake_up(&eli->li_wait_task);
2817
2818         kfree(ext4_li_info);
2819         ext4_lazyinit_task = NULL;
2820         ext4_li_info = NULL;
2821         mutex_unlock(&ext4_li_mtx);
2822
2823         return 0;
2824 }
2825
2826 static void ext4_clear_request_list(void)
2827 {
2828         struct list_head *pos, *n;
2829         struct ext4_li_request *elr;
2830
2831         mutex_lock(&ext4_li_info->li_list_mtx);
2832         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2833                 elr = list_entry(pos, struct ext4_li_request,
2834                                  lr_request);
2835                 ext4_remove_li_request(elr);
2836         }
2837         mutex_unlock(&ext4_li_info->li_list_mtx);
2838 }
2839
2840 static int ext4_run_lazyinit_thread(void)
2841 {
2842         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2843                                          ext4_li_info, "ext4lazyinit");
2844         if (IS_ERR(ext4_lazyinit_task)) {
2845                 int err = PTR_ERR(ext4_lazyinit_task);
2846                 ext4_clear_request_list();
2847                 del_timer_sync(&ext4_li_info->li_timer);
2848                 kfree(ext4_li_info);
2849                 ext4_li_info = NULL;
2850                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2851                                  "initialization thread\n",
2852                                  err);
2853                 return err;
2854         }
2855         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2856
2857         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2858         return 0;
2859 }
2860
2861 /*
2862  * Check whether it make sense to run itable init. thread or not.
2863  * If there is at least one uninitialized inode table, return
2864  * corresponding group number, else the loop goes through all
2865  * groups and return total number of groups.
2866  */
2867 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2868 {
2869         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2870         struct ext4_group_desc *gdp = NULL;
2871
2872         for (group = 0; group < ngroups; group++) {
2873                 gdp = ext4_get_group_desc(sb, group, NULL);
2874                 if (!gdp)
2875                         continue;
2876
2877                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2878                         break;
2879         }
2880
2881         return group;
2882 }
2883
2884 static int ext4_li_info_new(void)
2885 {
2886         struct ext4_lazy_init *eli = NULL;
2887
2888         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2889         if (!eli)
2890                 return -ENOMEM;
2891
2892         eli->li_task = NULL;
2893         INIT_LIST_HEAD(&eli->li_request_list);
2894         mutex_init(&eli->li_list_mtx);
2895
2896         init_waitqueue_head(&eli->li_wait_daemon);
2897         init_waitqueue_head(&eli->li_wait_task);
2898         init_timer(&eli->li_timer);
2899         eli->li_state |= EXT4_LAZYINIT_QUIT;
2900
2901         ext4_li_info = eli;
2902
2903         return 0;
2904 }
2905
2906 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2907                                             ext4_group_t start)
2908 {
2909         struct ext4_sb_info *sbi = EXT4_SB(sb);
2910         struct ext4_li_request *elr;
2911         unsigned long rnd;
2912
2913         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2914         if (!elr)
2915                 return NULL;
2916
2917         elr->lr_super = sb;
2918         elr->lr_sbi = sbi;
2919         elr->lr_next_group = start;
2920
2921         /*
2922          * Randomize first schedule time of the request to
2923          * spread the inode table initialization requests
2924          * better.
2925          */
2926         get_random_bytes(&rnd, sizeof(rnd));
2927         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2928                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2929
2930         return elr;
2931 }
2932
2933 static int ext4_register_li_request(struct super_block *sb,
2934                                     ext4_group_t first_not_zeroed)
2935 {
2936         struct ext4_sb_info *sbi = EXT4_SB(sb);
2937         struct ext4_li_request *elr;
2938         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2939         int ret = 0;
2940
2941         if (sbi->s_li_request != NULL)
2942                 return 0;
2943
2944         if (first_not_zeroed == ngroups ||
2945             (sb->s_flags & MS_RDONLY) ||
2946             !test_opt(sb, INIT_INODE_TABLE)) {
2947                 sbi->s_li_request = NULL;
2948                 return 0;
2949         }
2950
2951         if (first_not_zeroed == ngroups) {
2952                 sbi->s_li_request = NULL;
2953                 return 0;
2954         }
2955
2956         elr = ext4_li_request_new(sb, first_not_zeroed);
2957         if (!elr)
2958                 return -ENOMEM;
2959
2960         mutex_lock(&ext4_li_mtx);
2961
2962         if (NULL == ext4_li_info) {
2963                 ret = ext4_li_info_new();
2964                 if (ret)
2965                         goto out;
2966         }
2967
2968         mutex_lock(&ext4_li_info->li_list_mtx);
2969         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2970         mutex_unlock(&ext4_li_info->li_list_mtx);
2971
2972         sbi->s_li_request = elr;
2973
2974         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2975                 ret = ext4_run_lazyinit_thread();
2976                 if (ret)
2977                         goto out;
2978         }
2979 out:
2980         mutex_unlock(&ext4_li_mtx);
2981         if (ret)
2982                 kfree(elr);
2983         return ret;
2984 }
2985
2986 /*
2987  * We do not need to lock anything since this is called on
2988  * module unload.
2989  */
2990 static void ext4_destroy_lazyinit_thread(void)
2991 {
2992         /*
2993          * If thread exited earlier
2994          * there's nothing to be done.
2995          */
2996         if (!ext4_li_info || !ext4_lazyinit_task)
2997                 return;
2998
2999         kthread_stop(ext4_lazyinit_task);
3000 }
3001
3002 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3003                                 __releases(kernel_lock)
3004                                 __acquires(kernel_lock)
3005 {
3006         char *orig_data = kstrdup(data, GFP_KERNEL);
3007         struct buffer_head *bh;
3008         struct ext4_super_block *es = NULL;
3009         struct ext4_sb_info *sbi;
3010         ext4_fsblk_t block;
3011         ext4_fsblk_t sb_block = get_sb_block(&data);
3012         ext4_fsblk_t logical_sb_block;
3013         unsigned long offset = 0;
3014         unsigned long journal_devnum = 0;
3015         unsigned long def_mount_opts;
3016         struct inode *root;
3017         char *cp;
3018         const char *descr;
3019         int ret = -ENOMEM;
3020         int blocksize;
3021         unsigned int db_count;
3022         unsigned int i;
3023         int needs_recovery, has_huge_files;
3024         __u64 blocks_count;
3025         int err;
3026         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3027         ext4_group_t first_not_zeroed;
3028
3029         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3030         if (!sbi)
3031                 goto out_free_orig;
3032
3033         sbi->s_blockgroup_lock =
3034                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3035         if (!sbi->s_blockgroup_lock) {
3036                 kfree(sbi);
3037                 goto out_free_orig;
3038         }
3039         sb->s_fs_info = sbi;
3040         sbi->s_mount_opt = 0;
3041         sbi->s_resuid = EXT4_DEF_RESUID;
3042         sbi->s_resgid = EXT4_DEF_RESGID;
3043         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3044         sbi->s_sb_block = sb_block;
3045         if (sb->s_bdev->bd_part)
3046                 sbi->s_sectors_written_start =
3047                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3048
3049         /* Cleanup superblock name */
3050         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3051                 *cp = '!';
3052
3053         ret = -EINVAL;
3054         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3055         if (!blocksize) {
3056                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3057                 goto out_fail;
3058         }
3059
3060         /*
3061          * The ext4 superblock will not be buffer aligned for other than 1kB
3062          * block sizes.  We need to calculate the offset from buffer start.
3063          */
3064         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3065                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3066                 offset = do_div(logical_sb_block, blocksize);
3067         } else {
3068                 logical_sb_block = sb_block;
3069         }
3070
3071         if (!(bh = sb_bread(sb, logical_sb_block))) {
3072                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3073                 goto out_fail;
3074         }
3075         /*
3076          * Note: s_es must be initialized as soon as possible because
3077          *       some ext4 macro-instructions depend on its value
3078          */
3079         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3080         sbi->s_es = es;
3081         sb->s_magic = le16_to_cpu(es->s_magic);
3082         if (sb->s_magic != EXT4_SUPER_MAGIC)
3083                 goto cantfind_ext4;
3084         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3085
3086         /* Set defaults before we parse the mount options */
3087         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3088         set_opt(sb, INIT_INODE_TABLE);
3089         if (def_mount_opts & EXT4_DEFM_DEBUG)
3090                 set_opt(sb, DEBUG);
3091         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3092                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3093                         "2.6.38");
3094                 set_opt(sb, GRPID);
3095         }
3096         if (def_mount_opts & EXT4_DEFM_UID16)
3097                 set_opt(sb, NO_UID32);
3098 #ifdef CONFIG_EXT4_FS_XATTR
3099         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3100                 set_opt(sb, XATTR_USER);
3101 #endif
3102 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3103         if (def_mount_opts & EXT4_DEFM_ACL)
3104                 set_opt(sb, POSIX_ACL);
3105 #endif
3106         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3107                 set_opt(sb, JOURNAL_DATA);
3108         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3109                 set_opt(sb, ORDERED_DATA);
3110         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3111                 set_opt(sb, WRITEBACK_DATA);
3112
3113         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3114                 set_opt(sb, ERRORS_PANIC);
3115         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3116                 set_opt(sb, ERRORS_CONT);
3117         else
3118                 set_opt(sb, ERRORS_RO);
3119         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3120                 set_opt(sb, BLOCK_VALIDITY);
3121         if (def_mount_opts & EXT4_DEFM_DISCARD)
3122                 set_opt(sb, DISCARD);
3123
3124         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3125         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3126         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3127         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3128         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3129
3130         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3131                 set_opt(sb, BARRIER);
3132
3133         /*
3134          * enable delayed allocation by default
3135          * Use -o nodelalloc to turn it off
3136          */
3137         if (!IS_EXT3_SB(sb) &&
3138             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3139                 set_opt(sb, DELALLOC);
3140
3141         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3142                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3143                 ext4_msg(sb, KERN_WARNING,
3144                          "failed to parse options in superblock: %s",
3145                          sbi->s_es->s_mount_opts);
3146         }
3147         if (!parse_options((char *) data, sb, &journal_devnum,
3148                            &journal_ioprio, NULL, 0))
3149                 goto failed_mount;
3150
3151         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3152                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3153
3154         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3155             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3156              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3157              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3158                 ext4_msg(sb, KERN_WARNING,
3159                        "feature flags set on rev 0 fs, "
3160                        "running e2fsck is recommended");
3161
3162         /*
3163          * Check feature flags regardless of the revision level, since we
3164          * previously didn't change the revision level when setting the flags,
3165          * so there is a chance incompat flags are set on a rev 0 filesystem.
3166          */
3167         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3168                 goto failed_mount;
3169
3170         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3171
3172         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3173             blocksize > EXT4_MAX_BLOCK_SIZE) {
3174                 ext4_msg(sb, KERN_ERR,
3175                        "Unsupported filesystem blocksize %d", blocksize);
3176                 goto failed_mount;
3177         }
3178
3179         if (sb->s_blocksize != blocksize) {
3180                 /* Validate the filesystem blocksize */
3181                 if (!sb_set_blocksize(sb, blocksize)) {
3182                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3183                                         blocksize);
3184                         goto failed_mount;
3185                 }
3186
3187                 brelse(bh);
3188                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3189                 offset = do_div(logical_sb_block, blocksize);
3190                 bh = sb_bread(sb, logical_sb_block);
3191                 if (!bh) {
3192                         ext4_msg(sb, KERN_ERR,
3193                                "Can't read superblock on 2nd try");
3194                         goto failed_mount;
3195                 }
3196                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3197                 sbi->s_es = es;
3198                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3199                         ext4_msg(sb, KERN_ERR,
3200                                "Magic mismatch, very weird!");
3201                         goto failed_mount;
3202                 }
3203         }
3204
3205         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3206                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3207         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3208                                                       has_huge_files);
3209         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3210
3211         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3212                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3213                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3214         } else {
3215                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3216                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3217                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3218                     (!is_power_of_2(sbi->s_inode_size)) ||
3219                     (sbi->s_inode_size > blocksize)) {
3220                         ext4_msg(sb, KERN_ERR,
3221                                "unsupported inode size: %d",
3222                                sbi->s_inode_size);
3223                         goto failed_mount;
3224                 }
3225                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3226                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3227         }
3228
3229         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3230         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3231                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3232                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3233                     !is_power_of_2(sbi->s_desc_size)) {
3234                         ext4_msg(sb, KERN_ERR,
3235                                "unsupported descriptor size %lu",
3236                                sbi->s_desc_size);
3237                         goto failed_mount;
3238                 }
3239         } else
3240                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3241
3242         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3243         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3244         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3245                 goto cantfind_ext4;
3246
3247         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3248         if (sbi->s_inodes_per_block == 0)
3249                 goto cantfind_ext4;
3250         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3251                                         sbi->s_inodes_per_block;
3252         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3253         sbi->s_sbh = bh;
3254         sbi->s_mount_state = le16_to_cpu(es->s_state);
3255         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3256         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3257
3258         for (i = 0; i < 4; i++)
3259                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3260         sbi->s_def_hash_version = es->s_def_hash_version;
3261         i = le32_to_cpu(es->s_flags);
3262         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3263                 sbi->s_hash_unsigned = 3;
3264         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3265 #ifdef __CHAR_UNSIGNED__
3266                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3267                 sbi->s_hash_unsigned = 3;
3268 #else
3269                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3270 #endif
3271                 sb->s_dirt = 1;
3272         }
3273
3274         if (sbi->s_blocks_per_group > blocksize * 8) {
3275                 ext4_msg(sb, KERN_ERR,
3276                        "#blocks per group too big: %lu",
3277                        sbi->s_blocks_per_group);
3278                 goto failed_mount;
3279         }
3280         if (sbi->s_inodes_per_group > blocksize * 8) {
3281                 ext4_msg(sb, KERN_ERR,
3282                        "#inodes per group too big: %lu",
3283                        sbi->s_inodes_per_group);
3284                 goto failed_mount;
3285         }
3286
3287         /*
3288          * Test whether we have more sectors than will fit in sector_t,
3289          * and whether the max offset is addressable by the page cache.
3290          */
3291         err = generic_check_addressable(sb->s_blocksize_bits,
3292                                         ext4_blocks_count(es));
3293         if (err) {
3294                 ext4_msg(sb, KERN_ERR, "filesystem"
3295                          " too large to mount safely on this system");
3296                 if (sizeof(sector_t) < 8)
3297                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3298                 ret = err;
3299                 goto failed_mount;
3300         }
3301
3302         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3303                 goto cantfind_ext4;
3304
3305         /* check blocks count against device size */
3306         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3307         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3308                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3309                        "exceeds size of device (%llu blocks)",
3310                        ext4_blocks_count(es), blocks_count);
3311                 goto failed_mount;
3312         }
3313
3314         /*
3315          * It makes no sense for the first data block to be beyond the end
3316          * of the filesystem.
3317          */
3318         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3319                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3320                          "block %u is beyond end of filesystem (%llu)",
3321                          le32_to_cpu(es->s_first_data_block),
3322                          ext4_blocks_count(es));
3323                 goto failed_mount;
3324         }
3325         blocks_count = (ext4_blocks_count(es) -
3326                         le32_to_cpu(es->s_first_data_block) +
3327                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3328         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3329         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3330                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3331                        "(block count %llu, first data block %u, "
3332                        "blocks per group %lu)", sbi->s_groups_count,
3333                        ext4_blocks_count(es),
3334                        le32_to_cpu(es->s_first_data_block),
3335                        EXT4_BLOCKS_PER_GROUP(sb));
3336                 goto failed_mount;
3337         }
3338         sbi->s_groups_count = blocks_count;
3339         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3340                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3341         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3342                    EXT4_DESC_PER_BLOCK(sb);
3343         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3344                                     GFP_KERNEL);
3345         if (sbi->s_group_desc == NULL) {
3346                 ext4_msg(sb, KERN_ERR, "not enough memory");
3347                 goto failed_mount;
3348         }
3349
3350 #ifdef CONFIG_PROC_FS
3351         if (ext4_proc_root)
3352                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3353 #endif
3354
3355         bgl_lock_init(sbi->s_blockgroup_lock);
3356
3357         for (i = 0; i < db_count; i++) {
3358                 block = descriptor_loc(sb, logical_sb_block, i);
3359                 sbi->s_group_desc[i] = sb_bread(sb, block);
3360                 if (!sbi->s_group_desc[i]) {
3361                         ext4_msg(sb, KERN_ERR,
3362                                "can't read group descriptor %d", i);
3363                         db_count = i;
3364                         goto failed_mount2;
3365                 }
3366         }
3367         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3368                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3369                 goto failed_mount2;
3370         }
3371         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3372                 if (!ext4_fill_flex_info(sb)) {
3373                         ext4_msg(sb, KERN_ERR,
3374                                "unable to initialize "
3375                                "flex_bg meta info!");
3376                         goto failed_mount2;
3377                 }
3378
3379         sbi->s_gdb_count = db_count;
3380         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3381         spin_lock_init(&sbi->s_next_gen_lock);
3382
3383         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3384                         ext4_count_free_blocks(sb));
3385         if (!err) {
3386                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3387                                 ext4_count_free_inodes(sb));
3388         }
3389         if (!err) {
3390                 err = percpu_counter_init(&sbi->s_dirs_counter,
3391                                 ext4_count_dirs(sb));
3392         }
3393         if (!err) {
3394                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3395         }
3396         if (err) {
3397                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3398                 goto failed_mount3;
3399         }
3400
3401         sbi->s_stripe = ext4_get_stripe_size(sbi);
3402         sbi->s_max_writeback_mb_bump = 128;
3403
3404         /*
3405          * set up enough so that it can read an inode
3406          */
3407         if (!test_opt(sb, NOLOAD) &&
3408             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3409                 sb->s_op = &ext4_sops;
3410         else
3411                 sb->s_op = &ext4_nojournal_sops;
3412         sb->s_export_op = &ext4_export_ops;
3413         sb->s_xattr = ext4_xattr_handlers;
3414 #ifdef CONFIG_QUOTA
3415         sb->s_qcop = &ext4_qctl_operations;
3416         sb->dq_op = &ext4_quota_operations;
3417 #endif
3418         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3419
3420         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3421         mutex_init(&sbi->s_orphan_lock);
3422         mutex_init(&sbi->s_resize_lock);
3423
3424         sb->s_root = NULL;
3425
3426         needs_recovery = (es->s_last_orphan != 0 ||
3427                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3428                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3429
3430         /*
3431          * The first inode we look at is the journal inode.  Don't try
3432          * root first: it may be modified in the journal!
3433          */
3434         if (!test_opt(sb, NOLOAD) &&
3435             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3436                 if (ext4_load_journal(sb, es, journal_devnum))
3437                         goto failed_mount3;
3438         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3439               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3440                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3441                        "suppressed and not mounted read-only");
3442                 goto failed_mount_wq;
3443         } else {
3444                 clear_opt(sb, DATA_FLAGS);
3445                 set_opt(sb, WRITEBACK_DATA);
3446                 sbi->s_journal = NULL;
3447                 needs_recovery = 0;
3448                 goto no_journal;
3449         }
3450
3451         if (ext4_blocks_count(es) > 0xffffffffULL &&
3452             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3453                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3454                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3455                 goto failed_mount_wq;
3456         }
3457
3458         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3459                 jbd2_journal_set_features(sbi->s_journal,
3460                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3461                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3462         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3463                 jbd2_journal_set_features(sbi->s_journal,
3464                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3465                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3466                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3467         } else {
3468                 jbd2_journal_clear_features(sbi->s_journal,
3469                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3470                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3471         }
3472
3473         /* We have now updated the journal if required, so we can
3474          * validate the data journaling mode. */
3475         switch (test_opt(sb, DATA_FLAGS)) {
3476         case 0:
3477                 /* No mode set, assume a default based on the journal
3478                  * capabilities: ORDERED_DATA if the journal can
3479                  * cope, else JOURNAL_DATA
3480                  */
3481                 if (jbd2_journal_check_available_features
3482                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3483                         set_opt(sb, ORDERED_DATA);
3484                 else
3485                         set_opt(sb, JOURNAL_DATA);
3486                 break;
3487
3488         case EXT4_MOUNT_ORDERED_DATA:
3489         case EXT4_MOUNT_WRITEBACK_DATA:
3490                 if (!jbd2_journal_check_available_features
3491                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3492                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3493                                "requested data journaling mode");
3494                         goto failed_mount_wq;
3495                 }
3496         default:
3497                 break;
3498         }
3499         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3500
3501         /*
3502          * The journal may have updated the bg summary counts, so we
3503          * need to update the global counters.
3504          */
3505         percpu_counter_set(&sbi->s_freeblocks_counter,
3506                            ext4_count_free_blocks(sb));
3507         percpu_counter_set(&sbi->s_freeinodes_counter,
3508                            ext4_count_free_inodes(sb));
3509         percpu_counter_set(&sbi->s_dirs_counter,
3510                            ext4_count_dirs(sb));
3511         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3512
3513 no_journal:
3514         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3515         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3516                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3517                 goto failed_mount_wq;
3518         }
3519
3520         /*
3521          * The jbd2_journal_load will have done any necessary log recovery,
3522          * so we can safely mount the rest of the filesystem now.
3523          */
3524
3525         root = ext4_iget(sb, EXT4_ROOT_INO);
3526         if (IS_ERR(root)) {
3527                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3528                 ret = PTR_ERR(root);
3529                 goto failed_mount4;
3530         }
3531         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3532                 iput(root);
3533                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3534                 goto failed_mount4;
3535         }
3536         sb->s_root = d_alloc_root(root);
3537         if (!sb->s_root) {
3538                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3539                 iput(root);
3540                 ret = -ENOMEM;
3541                 goto failed_mount4;
3542         }
3543
3544         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3545
3546         /* determine the minimum size of new large inodes, if present */
3547         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3548                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3549                                                      EXT4_GOOD_OLD_INODE_SIZE;
3550                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3551                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3552                         if (sbi->s_want_extra_isize <
3553                             le16_to_cpu(es->s_want_extra_isize))
3554                                 sbi->s_want_extra_isize =
3555                                         le16_to_cpu(es->s_want_extra_isize);
3556                         if (sbi->s_want_extra_isize <
3557                             le16_to_cpu(es->s_min_extra_isize))
3558                                 sbi->s_want_extra_isize =
3559                                         le16_to_cpu(es->s_min_extra_isize);
3560                 }
3561         }
3562         /* Check if enough inode space is available */
3563         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3564                                                         sbi->s_inode_size) {
3565                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3566                                                        EXT4_GOOD_OLD_INODE_SIZE;
3567                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3568                          "available");
3569         }
3570
3571         if (test_opt(sb, DELALLOC) &&
3572             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3573                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3574                          "requested data journaling mode");
3575                 clear_opt(sb, DELALLOC);
3576         }
3577         if (test_opt(sb, DIOREAD_NOLOCK)) {
3578                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3579                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3580                                 "option - requested data journaling mode");
3581                         clear_opt(sb, DIOREAD_NOLOCK);
3582                 }
3583                 if (sb->s_blocksize < PAGE_SIZE) {
3584                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3585                                 "option - block size is too small");
3586                         clear_opt(sb, DIOREAD_NOLOCK);
3587                 }
3588         }
3589
3590         err = ext4_setup_system_zone(sb);
3591         if (err) {
3592                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3593                          "zone (%d)", err);
3594                 goto failed_mount4;
3595         }
3596
3597         ext4_ext_init(sb);
3598         err = ext4_mb_init(sb, needs_recovery);
3599         if (err) {
3600                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3601                          err);
3602                 goto failed_mount4;
3603         }
3604
3605         err = ext4_register_li_request(sb, first_not_zeroed);
3606         if (err)
3607                 goto failed_mount4;
3608
3609         sbi->s_kobj.kset = ext4_kset;
3610         init_completion(&sbi->s_kobj_unregister);
3611         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3612                                    "%s", sb->s_id);
3613         if (err) {
3614                 ext4_mb_release(sb);
3615                 ext4_ext_release(sb);
3616                 goto failed_mount4;
3617         };
3618
3619         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3620         ext4_orphan_cleanup(sb, es);
3621         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3622         if (needs_recovery) {
3623                 ext4_msg(sb, KERN_INFO, "recovery complete");
3624                 ext4_mark_recovery_complete(sb, es);
3625         }
3626         if (EXT4_SB(sb)->s_journal) {
3627                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3628                         descr = " journalled data mode";
3629                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3630                         descr = " ordered data mode";
3631                 else
3632                         descr = " writeback data mode";
3633         } else
3634                 descr = "out journal";
3635
3636         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3637                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3638                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3639
3640         init_timer(&sbi->s_err_report);
3641         sbi->s_err_report.function = print_daily_error_info;
3642         sbi->s_err_report.data = (unsigned long) sb;
3643         if (es->s_error_count)
3644                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3645
3646         kfree(orig_data);
3647         return 0;
3648
3649 cantfind_ext4:
3650         if (!silent)
3651                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3652         goto failed_mount;
3653
3654 failed_mount4:
3655         ext4_msg(sb, KERN_ERR, "mount failed");
3656         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3657 failed_mount_wq:
3658         ext4_release_system_zone(sb);
3659         if (sbi->s_journal) {
3660                 jbd2_journal_destroy(sbi->s_journal);
3661                 sbi->s_journal = NULL;
3662         }
3663 failed_mount3:
3664         if (sbi->s_flex_groups) {
3665                 if (is_vmalloc_addr(sbi->s_flex_groups))
3666                         vfree(sbi->s_flex_groups);
3667                 else
3668                         kfree(sbi->s_flex_groups);
3669         }
3670         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3671         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3672         percpu_counter_destroy(&sbi->s_dirs_counter);
3673         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3674 failed_mount2:
3675         for (i = 0; i < db_count; i++)
3676                 brelse(sbi->s_group_desc[i]);
3677         kfree(sbi->s_group_desc);
3678 failed_mount:
3679         if (sbi->s_proc) {
3680                 remove_proc_entry(sb->s_id, ext4_proc_root);
3681         }
3682 #ifdef CONFIG_QUOTA
3683         for (i = 0; i < MAXQUOTAS; i++)
3684                 kfree(sbi->s_qf_names[i]);
3685 #endif
3686         ext4_blkdev_remove(sbi);
3687         brelse(bh);
3688 out_fail:
3689         sb->s_fs_info = NULL;
3690         kfree(sbi->s_blockgroup_lock);
3691         kfree(sbi);
3692 out_free_orig:
3693         kfree(orig_data);
3694         return ret;
3695 }
3696
3697 /*
3698  * Setup any per-fs journal parameters now.  We'll do this both on
3699  * initial mount, once the journal has been initialised but before we've
3700  * done any recovery; and again on any subsequent remount.
3701  */
3702 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3703 {
3704         struct ext4_sb_info *sbi = EXT4_SB(sb);
3705
3706         journal->j_commit_interval = sbi->s_commit_interval;
3707         journal->j_min_batch_time = sbi->s_min_batch_time;
3708         journal->j_max_batch_time = sbi->s_max_batch_time;
3709
3710         write_lock(&journal->j_state_lock);
3711         if (test_opt(sb, BARRIER))
3712                 journal->j_flags |= JBD2_BARRIER;
3713         else
3714                 journal->j_flags &= ~JBD2_BARRIER;
3715         if (test_opt(sb, DATA_ERR_ABORT))
3716                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3717         else
3718                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3719         write_unlock(&journal->j_state_lock);
3720 }
3721
3722 static journal_t *ext4_get_journal(struct super_block *sb,
3723                                    unsigned int journal_inum)
3724 {
3725         struct inode *journal_inode;
3726         journal_t *journal;
3727
3728         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3729
3730         /* First, test for the existence of a valid inode on disk.  Bad
3731          * things happen if we iget() an unused inode, as the subsequent
3732          * iput() will try to delete it. */
3733
3734         journal_inode = ext4_iget(sb, journal_inum);
3735         if (IS_ERR(journal_inode)) {
3736                 ext4_msg(sb, KERN_ERR, "no journal found");
3737                 return NULL;
3738         }
3739         if (!journal_inode->i_nlink) {
3740                 make_bad_inode(journal_inode);
3741                 iput(journal_inode);
3742                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3743                 return NULL;
3744         }
3745
3746         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3747                   journal_inode, journal_inode->i_size);
3748         if (!S_ISREG(journal_inode->i_mode)) {
3749                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3750                 iput(journal_inode);
3751                 return NULL;
3752         }
3753
3754         journal = jbd2_journal_init_inode(journal_inode);
3755         if (!journal) {
3756                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3757                 iput(journal_inode);
3758                 return NULL;
3759         }
3760         journal->j_private = sb;
3761         ext4_init_journal_params(sb, journal);
3762         return journal;
3763 }
3764
3765 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3766                                        dev_t j_dev)
3767 {
3768         struct buffer_head *bh;
3769         journal_t *journal;
3770         ext4_fsblk_t start;
3771         ext4_fsblk_t len;
3772         int hblock, blocksize;
3773         ext4_fsblk_t sb_block;
3774         unsigned long offset;
3775         struct ext4_super_block *es;
3776         struct block_device *bdev;
3777
3778         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3779
3780         bdev = ext4_blkdev_get(j_dev, sb);
3781         if (bdev == NULL)
3782                 return NULL;
3783
3784         blocksize = sb->s_blocksize;
3785         hblock = bdev_logical_block_size(bdev);
3786         if (blocksize < hblock) {
3787                 ext4_msg(sb, KERN_ERR,
3788                         "blocksize too small for journal device");
3789                 goto out_bdev;
3790         }
3791
3792         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3793         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3794         set_blocksize(bdev, blocksize);
3795         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3796                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3797                        "external journal");
3798                 goto out_bdev;
3799         }
3800
3801         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3802         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3803             !(le32_to_cpu(es->s_feature_incompat) &
3804               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3805                 ext4_msg(sb, KERN_ERR, "external journal has "
3806                                         "bad superblock");
3807                 brelse(bh);
3808                 goto out_bdev;
3809         }
3810
3811         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3812                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3813                 brelse(bh);
3814                 goto out_bdev;
3815         }
3816
3817         len = ext4_blocks_count(es);
3818         start = sb_block + 1;
3819         brelse(bh);     /* we're done with the superblock */
3820
3821         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3822                                         start, len, blocksize);
3823         if (!journal) {
3824                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3825                 goto out_bdev;
3826         }
3827         journal->j_private = sb;
3828         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3829         wait_on_buffer(journal->j_sb_buffer);
3830         if (!buffer_uptodate(journal->j_sb_buffer)) {
3831                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3832                 goto out_journal;
3833         }
3834         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3835                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3836                                         "user (unsupported) - %d",
3837                         be32_to_cpu(journal->j_superblock->s_nr_users));
3838                 goto out_journal;
3839         }
3840         EXT4_SB(sb)->journal_bdev = bdev;
3841         ext4_init_journal_params(sb, journal);
3842         return journal;
3843
3844 out_journal:
3845         jbd2_journal_destroy(journal);
3846 out_bdev:
3847         ext4_blkdev_put(bdev);
3848         return NULL;
3849 }
3850
3851 static int ext4_load_journal(struct super_block *sb,
3852                              struct ext4_super_block *es,
3853                              unsigned long journal_devnum)
3854 {
3855         journal_t *journal;
3856         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3857         dev_t journal_dev;
3858         int err = 0;
3859         int really_read_only;
3860
3861         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3862
3863         if (journal_devnum &&
3864             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3865                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3866                         "numbers have changed");
3867                 journal_dev = new_decode_dev(journal_devnum);
3868         } else
3869                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3870
3871         really_read_only = bdev_read_only(sb->s_bdev);
3872
3873         /*
3874          * Are we loading a blank journal or performing recovery after a
3875          * crash?  For recovery, we need to check in advance whether we
3876          * can get read-write access to the device.
3877          */
3878         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3879                 if (sb->s_flags & MS_RDONLY) {
3880                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3881                                         "required on readonly filesystem");
3882                         if (really_read_only) {
3883                                 ext4_msg(sb, KERN_ERR, "write access "
3884                                         "unavailable, cannot proceed");
3885                                 return -EROFS;
3886                         }
3887                         ext4_msg(sb, KERN_INFO, "write access will "
3888                                "be enabled during recovery");
3889                 }
3890         }
3891
3892         if (journal_inum && journal_dev) {
3893                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3894                        "and inode journals!");
3895                 return -EINVAL;
3896         }
3897
3898         if (journal_inum) {
3899                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3900                         return -EINVAL;
3901         } else {
3902                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3903                         return -EINVAL;
3904         }
3905
3906         if (!(journal->j_flags & JBD2_BARRIER))
3907                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3908
3909         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3910                 err = jbd2_journal_update_format(journal);
3911                 if (err)  {
3912                         ext4_msg(sb, KERN_ERR, "error updating journal");
3913                         jbd2_journal_destroy(journal);
3914                         return err;
3915                 }
3916         }
3917
3918         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3919                 err = jbd2_journal_wipe(journal, !really_read_only);
3920         if (!err) {
3921                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3922                 if (save)
3923                         memcpy(save, ((char *) es) +
3924                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3925                 err = jbd2_journal_load(journal);
3926                 if (save)
3927                         memcpy(((char *) es) + EXT4_S_ERR_START,
3928                                save, EXT4_S_ERR_LEN);
3929                 kfree(save);
3930         }
3931
3932         if (err) {
3933                 ext4_msg(sb, KERN_ERR, "error loading journal");
3934                 jbd2_journal_destroy(journal);
3935                 return err;
3936         }
3937
3938         EXT4_SB(sb)->s_journal = journal;
3939         ext4_clear_journal_err(sb, es);
3940
3941         if (!really_read_only && journal_devnum &&
3942             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3943                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3944
3945                 /* Make sure we flush the recovery flag to disk. */
3946                 ext4_commit_super(sb, 1);
3947         }
3948
3949         return 0;
3950 }
3951
3952 static int ext4_commit_super(struct super_block *sb, int sync)
3953 {
3954         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3955         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3956         int error = 0;
3957
3958         if (!sbh)
3959                 return error;
3960         if (buffer_write_io_error(sbh)) {
3961                 /*
3962                  * Oh, dear.  A previous attempt to write the
3963                  * superblock failed.  This could happen because the
3964                  * USB device was yanked out.  Or it could happen to
3965                  * be a transient write error and maybe the block will
3966                  * be remapped.  Nothing we can do but to retry the
3967                  * write and hope for the best.
3968                  */
3969                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3970                        "superblock detected");
3971                 clear_buffer_write_io_error(sbh);
3972                 set_buffer_uptodate(sbh);
3973         }
3974         /*
3975          * If the file system is mounted read-only, don't update the
3976          * superblock write time.  This avoids updating the superblock
3977          * write time when we are mounting the root file system
3978          * read/only but we need to replay the journal; at that point,
3979          * for people who are east of GMT and who make their clock
3980          * tick in localtime for Windows bug-for-bug compatibility,
3981          * the clock is set in the future, and this will cause e2fsck
3982          * to complain and force a full file system check.
3983          */
3984         if (!(sb->s_flags & MS_RDONLY))
3985                 es->s_wtime = cpu_to_le32(get_seconds());
3986         if (sb->s_bdev->bd_part)
3987                 es->s_kbytes_written =
3988                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3989                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3990                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3991         else
3992                 es->s_kbytes_written =
3993                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3994         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3995                                            &EXT4_SB(sb)->s_freeblocks_counter));
3996         es->s_free_inodes_count =
3997                 cpu_to_le32(percpu_counter_sum_positive(
3998                                 &EXT4_SB(sb)->s_freeinodes_counter));
3999         sb->s_dirt = 0;
4000         BUFFER_TRACE(sbh, "marking dirty");
4001         mark_buffer_dirty(sbh);
4002         if (sync) {
4003                 error = sync_dirty_buffer(sbh);
4004                 if (error)
4005                         return error;
4006
4007                 error = buffer_write_io_error(sbh);
4008                 if (error) {
4009                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4010                                "superblock");
4011                         clear_buffer_write_io_error(sbh);
4012                         set_buffer_uptodate(sbh);
4013                 }
4014         }
4015         return error;
4016 }
4017
4018 /*
4019  * Have we just finished recovery?  If so, and if we are mounting (or
4020  * remounting) the filesystem readonly, then we will end up with a
4021  * consistent fs on disk.  Record that fact.
4022  */
4023 static void ext4_mark_recovery_complete(struct super_block *sb,
4024                                         struct ext4_super_block *es)
4025 {
4026         journal_t *journal = EXT4_SB(sb)->s_journal;
4027
4028         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4029                 BUG_ON(journal != NULL);
4030                 return;
4031         }
4032         jbd2_journal_lock_updates(journal);
4033         if (jbd2_journal_flush(journal) < 0)
4034                 goto out;
4035
4036         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4037             sb->s_flags & MS_RDONLY) {
4038                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4039                 ext4_commit_super(sb, 1);
4040         }
4041
4042 out:
4043         jbd2_journal_unlock_updates(journal);
4044 }
4045
4046 /*
4047  * If we are mounting (or read-write remounting) a filesystem whose journal
4048  * has recorded an error from a previous lifetime, move that error to the
4049  * main filesystem now.
4050  */
4051 static void ext4_clear_journal_err(struct super_block *sb,
4052                                    struct ext4_super_block *es)
4053 {
4054         journal_t *journal;
4055         int j_errno;
4056         const char *errstr;
4057
4058         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4059
4060         journal = EXT4_SB(sb)->s_journal;
4061
4062         /*
4063          * Now check for any error status which may have been recorded in the
4064          * journal by a prior ext4_error() or ext4_abort()
4065          */
4066
4067         j_errno = jbd2_journal_errno(journal);
4068         if (j_errno) {
4069                 char nbuf[16];
4070
4071                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4072                 ext4_warning(sb, "Filesystem error recorded "
4073                              "from previous mount: %s", errstr);
4074                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4075
4076                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4077                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4078                 ext4_commit_super(sb, 1);
4079
4080                 jbd2_journal_clear_err(journal);
4081         }
4082 }
4083
4084 /*
4085  * Force the running and committing transactions to commit,
4086  * and wait on the commit.
4087  */
4088 int ext4_force_commit(struct super_block *sb)
4089 {
4090         journal_t *journal;
4091         int ret = 0;
4092
4093         if (sb->s_flags & MS_RDONLY)
4094                 return 0;
4095
4096         journal = EXT4_SB(sb)->s_journal;
4097         if (journal) {
4098                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4099                 ret = ext4_journal_force_commit(journal);
4100         }
4101
4102         return ret;
4103 }
4104
4105 static void ext4_write_super(struct super_block *sb)
4106 {
4107         lock_super(sb);
4108         ext4_commit_super(sb, 1);
4109         unlock_super(sb);
4110 }
4111
4112 static int ext4_sync_fs(struct super_block *sb, int wait)
4113 {
4114         int ret = 0;
4115         tid_t target;
4116         struct ext4_sb_info *sbi = EXT4_SB(sb);
4117
4118         trace_ext4_sync_fs(sb, wait);
4119         flush_workqueue(sbi->dio_unwritten_wq);
4120         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4121                 if (wait)
4122                         jbd2_log_wait_commit(sbi->s_journal, target);
4123         }
4124         return ret;
4125 }
4126
4127 /*
4128  * LVM calls this function before a (read-only) snapshot is created.  This
4129  * gives us a chance to flush the journal completely and mark the fs clean.
4130  */
4131 static int ext4_freeze(struct super_block *sb)
4132 {
4133         int error = 0;
4134         journal_t *journal;
4135
4136         if (sb->s_flags & MS_RDONLY)
4137                 return 0;
4138
4139         journal = EXT4_SB(sb)->s_journal;
4140
4141         /* Now we set up the journal barrier. */
4142         jbd2_journal_lock_updates(journal);
4143
4144         /*
4145          * Don't clear the needs_recovery flag if we failed to flush
4146          * the journal.
4147          */
4148         error = jbd2_journal_flush(journal);
4149         if (error < 0)
4150                 goto out;
4151
4152         /* Journal blocked and flushed, clear needs_recovery flag. */
4153         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4154         error = ext4_commit_super(sb, 1);
4155 out:
4156         /* we rely on s_frozen to stop further updates */
4157         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4158         return error;
4159 }
4160
4161 /*
4162  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4163  * flag here, even though the filesystem is not technically dirty yet.
4164  */
4165 static int ext4_unfreeze(struct super_block *sb)
4166 {
4167         if (sb->s_flags & MS_RDONLY)
4168                 return 0;
4169
4170         lock_super(sb);
4171         /* Reset the needs_recovery flag before the fs is unlocked. */
4172         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4173         ext4_commit_super(sb, 1);
4174         unlock_super(sb);
4175         return 0;
4176 }
4177
4178 /*
4179  * Structure to save mount options for ext4_remount's benefit
4180  */
4181 struct ext4_mount_options {
4182         unsigned long s_mount_opt;
4183         unsigned long s_mount_opt2;
4184         uid_t s_resuid;
4185         gid_t s_resgid;
4186         unsigned long s_commit_interval;
4187         u32 s_min_batch_time, s_max_batch_time;
4188 #ifdef CONFIG_QUOTA
4189         int s_jquota_fmt;
4190         char *s_qf_names[MAXQUOTAS];
4191 #endif
4192 };
4193
4194 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4195 {
4196         struct ext4_super_block *es;
4197         struct ext4_sb_info *sbi = EXT4_SB(sb);
4198         ext4_fsblk_t n_blocks_count = 0;
4199         unsigned long old_sb_flags;
4200         struct ext4_mount_options old_opts;
4201         int enable_quota = 0;
4202         ext4_group_t g;
4203         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4204         int err;
4205 #ifdef CONFIG_QUOTA
4206         int i;
4207 #endif
4208         char *orig_data = kstrdup(data, GFP_KERNEL);
4209
4210         /* Store the original options */
4211         lock_super(sb);
4212         old_sb_flags = sb->s_flags;
4213         old_opts.s_mount_opt = sbi->s_mount_opt;
4214         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4215         old_opts.s_resuid = sbi->s_resuid;
4216         old_opts.s_resgid = sbi->s_resgid;
4217         old_opts.s_commit_interval = sbi->s_commit_interval;
4218         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4219         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4220 #ifdef CONFIG_QUOTA
4221         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4222         for (i = 0; i < MAXQUOTAS; i++)
4223                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4224 #endif
4225         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4226                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4227
4228         /*
4229          * Allow the "check" option to be passed as a remount option.
4230          */
4231         if (!parse_options(data, sb, NULL, &journal_ioprio,
4232                            &n_blocks_count, 1)) {
4233                 err = -EINVAL;
4234                 goto restore_opts;
4235         }
4236
4237         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4238                 ext4_abort(sb, "Abort forced by user");
4239
4240         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4241                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4242
4243         es = sbi->s_es;
4244
4245         if (sbi->s_journal) {
4246                 ext4_init_journal_params(sb, sbi->s_journal);
4247                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4248         }
4249
4250         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4251                 n_blocks_count > ext4_blocks_count(es)) {
4252                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4253                         err = -EROFS;
4254                         goto restore_opts;
4255                 }
4256
4257                 if (*flags & MS_RDONLY) {
4258                         err = dquot_suspend(sb, -1);
4259                         if (err < 0)
4260                                 goto restore_opts;
4261
4262                         /*
4263                          * First of all, the unconditional stuff we have to do
4264                          * to disable replay of the journal when we next remount
4265                          */
4266                         sb->s_flags |= MS_RDONLY;
4267
4268                         /*
4269                          * OK, test if we are remounting a valid rw partition
4270                          * readonly, and if so set the rdonly flag and then
4271                          * mark the partition as valid again.
4272                          */
4273                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4274                             (sbi->s_mount_state & EXT4_VALID_FS))
4275                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4276
4277                         if (sbi->s_journal)
4278                                 ext4_mark_recovery_complete(sb, es);
4279                 } else {
4280                         /* Make sure we can mount this feature set readwrite */
4281                         if (!ext4_feature_set_ok(sb, 0)) {
4282                                 err = -EROFS;
4283                                 goto restore_opts;
4284                         }
4285                         /*
4286                          * Make sure the group descriptor checksums
4287                          * are sane.  If they aren't, refuse to remount r/w.
4288                          */
4289                         for (g = 0; g < sbi->s_groups_count; g++) {
4290                                 struct ext4_group_desc *gdp =
4291                                         ext4_get_group_desc(sb, g, NULL);
4292
4293                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4294                                         ext4_msg(sb, KERN_ERR,
4295                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4296                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4297                                                le16_to_cpu(gdp->bg_checksum));
4298                                         err = -EINVAL;
4299                                         goto restore_opts;
4300                                 }
4301                         }
4302
4303                         /*
4304                          * If we have an unprocessed orphan list hanging
4305                          * around from a previously readonly bdev mount,
4306                          * require a full umount/remount for now.
4307                          */
4308                         if (es->s_last_orphan) {
4309                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4310                                        "remount RDWR because of unprocessed "
4311                                        "orphan inode list.  Please "
4312                                        "umount/remount instead");
4313                                 err = -EINVAL;
4314                                 goto restore_opts;
4315                         }
4316
4317                         /*
4318                          * Mounting a RDONLY partition read-write, so reread
4319                          * and store the current valid flag.  (It may have
4320                          * been changed by e2fsck since we originally mounted
4321                          * the partition.)
4322                          */
4323                         if (sbi->s_journal)
4324                                 ext4_clear_journal_err(sb, es);
4325                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4326                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4327                                 goto restore_opts;
4328                         if (!ext4_setup_super(sb, es, 0))
4329                                 sb->s_flags &= ~MS_RDONLY;
4330                         enable_quota = 1;
4331                 }
4332         }
4333
4334         /*
4335          * Reinitialize lazy itable initialization thread based on
4336          * current settings
4337          */
4338         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4339                 ext4_unregister_li_request(sb);
4340         else {
4341                 ext4_group_t first_not_zeroed;
4342                 first_not_zeroed = ext4_has_uninit_itable(sb);
4343                 ext4_register_li_request(sb, first_not_zeroed);
4344         }
4345
4346         ext4_setup_system_zone(sb);
4347         if (sbi->s_journal == NULL)
4348                 ext4_commit_super(sb, 1);
4349
4350 #ifdef CONFIG_QUOTA
4351         /* Release old quota file names */
4352         for (i = 0; i < MAXQUOTAS; i++)
4353                 if (old_opts.s_qf_names[i] &&
4354                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4355                         kfree(old_opts.s_qf_names[i]);
4356 #endif
4357         unlock_super(sb);
4358         if (enable_quota)
4359                 dquot_resume(sb, -1);
4360
4361         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4362         kfree(orig_data);
4363         return 0;
4364
4365 restore_opts:
4366         sb->s_flags = old_sb_flags;
4367         sbi->s_mount_opt = old_opts.s_mount_opt;
4368         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4369         sbi->s_resuid = old_opts.s_resuid;
4370         sbi->s_resgid = old_opts.s_resgid;
4371         sbi->s_commit_interval = old_opts.s_commit_interval;
4372         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4373         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4374 #ifdef CONFIG_QUOTA
4375         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4376         for (i = 0; i < MAXQUOTAS; i++) {
4377                 if (sbi->s_qf_names[i] &&
4378                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4379                         kfree(sbi->s_qf_names[i]);
4380                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4381         }
4382 #endif
4383         unlock_super(sb);
4384         kfree(orig_data);
4385         return err;
4386 }
4387
4388 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4389 {
4390         struct super_block *sb = dentry->d_sb;
4391         struct ext4_sb_info *sbi = EXT4_SB(sb);
4392         struct ext4_super_block *es = sbi->s_es;
4393         u64 fsid;
4394
4395         if (test_opt(sb, MINIX_DF)) {
4396                 sbi->s_overhead_last = 0;
4397         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4398                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4399                 ext4_fsblk_t overhead = 0;
4400
4401                 /*
4402                  * Compute the overhead (FS structures).  This is constant
4403                  * for a given filesystem unless the number of block groups
4404                  * changes so we cache the previous value until it does.
4405                  */
4406
4407                 /*
4408                  * All of the blocks before first_data_block are
4409                  * overhead
4410                  */
4411                 overhead = le32_to_cpu(es->s_first_data_block);
4412
4413                 /*
4414                  * Add the overhead attributed to the superblock and
4415                  * block group descriptors.  If the sparse superblocks
4416                  * feature is turned on, then not all groups have this.
4417                  */
4418                 for (i = 0; i < ngroups; i++) {
4419                         overhead += ext4_bg_has_super(sb, i) +
4420                                 ext4_bg_num_gdb(sb, i);
4421                         cond_resched();
4422                 }
4423
4424                 /*
4425                  * Every block group has an inode bitmap, a block
4426                  * bitmap, and an inode table.
4427                  */
4428                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4429                 sbi->s_overhead_last = overhead;
4430                 smp_wmb();
4431                 sbi->s_blocks_last = ext4_blocks_count(es);
4432         }
4433
4434         buf->f_type = EXT4_SUPER_MAGIC;
4435         buf->f_bsize = sb->s_blocksize;
4436         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4437         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4438                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4439         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4440         if (buf->f_bfree < ext4_r_blocks_count(es))
4441                 buf->f_bavail = 0;
4442         buf->f_files = le32_to_cpu(es->s_inodes_count);
4443         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4444         buf->f_namelen = EXT4_NAME_LEN;
4445         fsid = le64_to_cpup((void *)es->s_uuid) ^
4446                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4447         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4448         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4449
4450         return 0;
4451 }
4452
4453 /* Helper function for writing quotas on sync - we need to start transaction
4454  * before quota file is locked for write. Otherwise the are possible deadlocks:
4455  * Process 1                         Process 2
4456  * ext4_create()                     quota_sync()
4457  *   jbd2_journal_start()                  write_dquot()
4458  *   dquot_initialize()                         down(dqio_mutex)
4459  *     down(dqio_mutex)                    jbd2_journal_start()
4460  *
4461  */
4462
4463 #ifdef CONFIG_QUOTA
4464
4465 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4466 {
4467         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4468 }
4469
4470 static int ext4_write_dquot(struct dquot *dquot)
4471 {
4472         int ret, err;
4473         handle_t *handle;
4474         struct inode *inode;
4475
4476         inode = dquot_to_inode(dquot);
4477         handle = ext4_journal_start(inode,
4478                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4479         if (IS_ERR(handle))
4480                 return PTR_ERR(handle);
4481         ret = dquot_commit(dquot);
4482         err = ext4_journal_stop(handle);
4483         if (!ret)
4484                 ret = err;
4485         return ret;
4486 }
4487
4488 static int ext4_acquire_dquot(struct dquot *dquot)
4489 {
4490         int ret, err;
4491         handle_t *handle;
4492
4493         handle = ext4_journal_start(dquot_to_inode(dquot),
4494                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4495         if (IS_ERR(handle))
4496                 return PTR_ERR(handle);
4497         ret = dquot_acquire(dquot);
4498         err = ext4_journal_stop(handle);
4499         if (!ret)
4500                 ret = err;
4501         return ret;
4502 }
4503
4504 static int ext4_release_dquot(struct dquot *dquot)
4505 {
4506         int ret, err;
4507         handle_t *handle;
4508
4509         handle = ext4_journal_start(dquot_to_inode(dquot),
4510                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4511         if (IS_ERR(handle)) {
4512                 /* Release dquot anyway to avoid endless cycle in dqput() */
4513                 dquot_release(dquot);
4514                 return PTR_ERR(handle);
4515         }
4516         ret = dquot_release(dquot);
4517         err = ext4_journal_stop(handle);
4518         if (!ret)
4519                 ret = err;
4520         return ret;
4521 }
4522
4523 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4524 {
4525         /* Are we journaling quotas? */
4526         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4527             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4528                 dquot_mark_dquot_dirty(dquot);
4529                 return ext4_write_dquot(dquot);
4530         } else {
4531                 return dquot_mark_dquot_dirty(dquot);
4532         }
4533 }
4534
4535 static int ext4_write_info(struct super_block *sb, int type)
4536 {
4537         int ret, err;
4538         handle_t *handle;
4539
4540         /* Data block + inode block */
4541         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4542         if (IS_ERR(handle))
4543                 return PTR_ERR(handle);
4544         ret = dquot_commit_info(sb, type);
4545         err = ext4_journal_stop(handle);
4546         if (!ret)
4547                 ret = err;
4548         return ret;
4549 }
4550
4551 /*
4552  * Turn on quotas during mount time - we need to find
4553  * the quota file and such...
4554  */
4555 static int ext4_quota_on_mount(struct super_block *sb, int type)
4556 {
4557         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4558                                         EXT4_SB(sb)->s_jquota_fmt, type);
4559 }
4560
4561 /*
4562  * Standard function to be called on quota_on
4563  */
4564 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4565                          struct path *path)
4566 {
4567         int err;
4568
4569         if (!test_opt(sb, QUOTA))
4570                 return -EINVAL;
4571
4572         /* Quotafile not on the same filesystem? */
4573         if (path->mnt->mnt_sb != sb)
4574                 return -EXDEV;
4575         /* Journaling quota? */
4576         if (EXT4_SB(sb)->s_qf_names[type]) {
4577                 /* Quotafile not in fs root? */
4578                 if (path->dentry->d_parent != sb->s_root)
4579                         ext4_msg(sb, KERN_WARNING,
4580                                 "Quota file not on filesystem root. "
4581                                 "Journaled quota will not work");
4582         }
4583
4584         /*
4585          * When we journal data on quota file, we have to flush journal to see
4586          * all updates to the file when we bypass pagecache...
4587          */
4588         if (EXT4_SB(sb)->s_journal &&
4589             ext4_should_journal_data(path->dentry->d_inode)) {
4590                 /*
4591                  * We don't need to lock updates but journal_flush() could
4592                  * otherwise be livelocked...
4593                  */
4594                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4595                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4596                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4597                 if (err)
4598                         return err;
4599         }
4600
4601         return dquot_quota_on(sb, type, format_id, path);
4602 }
4603
4604 static int ext4_quota_off(struct super_block *sb, int type)
4605 {
4606         /* Force all delayed allocation blocks to be allocated.
4607          * Caller already holds s_umount sem */
4608         if (test_opt(sb, DELALLOC))
4609                 sync_filesystem(sb);
4610
4611         return dquot_quota_off(sb, type);
4612 }
4613
4614 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4615  * acquiring the locks... As quota files are never truncated and quota code
4616  * itself serializes the operations (and noone else should touch the files)
4617  * we don't have to be afraid of races */
4618 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4619                                size_t len, loff_t off)
4620 {
4621         struct inode *inode = sb_dqopt(sb)->files[type];
4622         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4623         int err = 0;
4624         int offset = off & (sb->s_blocksize - 1);
4625         int tocopy;
4626         size_t toread;
4627         struct buffer_head *bh;
4628         loff_t i_size = i_size_read(inode);
4629
4630         if (off > i_size)
4631                 return 0;
4632         if (off+len > i_size)
4633                 len = i_size-off;
4634         toread = len;
4635         while (toread > 0) {
4636                 tocopy = sb->s_blocksize - offset < toread ?
4637                                 sb->s_blocksize - offset : toread;
4638                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4639                 if (err)
4640                         return err;
4641                 if (!bh)        /* A hole? */
4642                         memset(data, 0, tocopy);
4643                 else
4644                         memcpy(data, bh->b_data+offset, tocopy);
4645                 brelse(bh);
4646                 offset = 0;
4647                 toread -= tocopy;
4648                 data += tocopy;
4649                 blk++;
4650         }
4651         return len;
4652 }
4653
4654 /* Write to quotafile (we know the transaction is already started and has
4655  * enough credits) */
4656 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4657                                 const char *data, size_t len, loff_t off)
4658 {
4659         struct inode *inode = sb_dqopt(sb)->files[type];
4660         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4661         int err = 0;
4662         int offset = off & (sb->s_blocksize - 1);
4663         struct buffer_head *bh;
4664         handle_t *handle = journal_current_handle();
4665
4666         if (EXT4_SB(sb)->s_journal && !handle) {
4667                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4668                         " cancelled because transaction is not started",
4669                         (unsigned long long)off, (unsigned long long)len);
4670                 return -EIO;
4671         }
4672         /*
4673          * Since we account only one data block in transaction credits,
4674          * then it is impossible to cross a block boundary.
4675          */
4676         if (sb->s_blocksize - offset < len) {
4677                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4678                         " cancelled because not block aligned",
4679                         (unsigned long long)off, (unsigned long long)len);
4680                 return -EIO;
4681         }
4682
4683         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4684         bh = ext4_bread(handle, inode, blk, 1, &err);
4685         if (!bh)
4686                 goto out;
4687         err = ext4_journal_get_write_access(handle, bh);
4688         if (err) {
4689                 brelse(bh);
4690                 goto out;
4691         }
4692         lock_buffer(bh);
4693         memcpy(bh->b_data+offset, data, len);
4694         flush_dcache_page(bh->b_page);
4695         unlock_buffer(bh);
4696         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4697         brelse(bh);
4698 out:
4699         if (err) {
4700                 mutex_unlock(&inode->i_mutex);
4701                 return err;
4702         }
4703         if (inode->i_size < off + len) {
4704                 i_size_write(inode, off + len);
4705                 EXT4_I(inode)->i_disksize = inode->i_size;
4706         }
4707         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4708         ext4_mark_inode_dirty(handle, inode);
4709         mutex_unlock(&inode->i_mutex);
4710         return len;
4711 }
4712
4713 #endif
4714
4715 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4716                        const char *dev_name, void *data)
4717 {
4718         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4719 }
4720
4721 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4722 static struct file_system_type ext2_fs_type = {
4723         .owner          = THIS_MODULE,
4724         .name           = "ext2",
4725         .mount          = ext4_mount,
4726         .kill_sb        = kill_block_super,
4727         .fs_flags       = FS_REQUIRES_DEV,
4728 };
4729
4730 static inline void register_as_ext2(void)
4731 {
4732         int err = register_filesystem(&ext2_fs_type);
4733         if (err)
4734                 printk(KERN_WARNING
4735                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4736 }
4737
4738 static inline void unregister_as_ext2(void)
4739 {
4740         unregister_filesystem(&ext2_fs_type);
4741 }
4742 MODULE_ALIAS("ext2");
4743 #else
4744 static inline void register_as_ext2(void) { }
4745 static inline void unregister_as_ext2(void) { }
4746 #endif
4747
4748 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4749 static inline void register_as_ext3(void)
4750 {
4751         int err = register_filesystem(&ext3_fs_type);
4752         if (err)
4753                 printk(KERN_WARNING
4754                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4755 }
4756
4757 static inline void unregister_as_ext3(void)
4758 {
4759         unregister_filesystem(&ext3_fs_type);
4760 }
4761 MODULE_ALIAS("ext3");
4762 #else
4763 static inline void register_as_ext3(void) { }
4764 static inline void unregister_as_ext3(void) { }
4765 #endif
4766
4767 static struct file_system_type ext4_fs_type = {
4768         .owner          = THIS_MODULE,
4769         .name           = "ext4",
4770         .mount          = ext4_mount,
4771         .kill_sb        = kill_block_super,
4772         .fs_flags       = FS_REQUIRES_DEV,
4773 };
4774
4775 static int __init ext4_init_feat_adverts(void)
4776 {
4777         struct ext4_features *ef;
4778         int ret = -ENOMEM;
4779
4780         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4781         if (!ef)
4782                 goto out;
4783
4784         ef->f_kobj.kset = ext4_kset;
4785         init_completion(&ef->f_kobj_unregister);
4786         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4787                                    "features");
4788         if (ret) {
4789                 kfree(ef);
4790                 goto out;
4791         }
4792
4793         ext4_feat = ef;
4794         ret = 0;
4795 out:
4796         return ret;
4797 }
4798
4799 static void ext4_exit_feat_adverts(void)
4800 {
4801         kobject_put(&ext4_feat->f_kobj);
4802         wait_for_completion(&ext4_feat->f_kobj_unregister);
4803         kfree(ext4_feat);
4804 }
4805
4806 /* Shared across all ext4 file systems */
4807 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4808 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4809
4810 static int __init ext4_init_fs(void)
4811 {
4812         int i, err;
4813
4814         ext4_check_flag_values();
4815
4816         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4817                 mutex_init(&ext4__aio_mutex[i]);
4818                 init_waitqueue_head(&ext4__ioend_wq[i]);
4819         }
4820
4821         err = ext4_init_pageio();
4822         if (err)
4823                 return err;
4824         err = ext4_init_system_zone();
4825         if (err)
4826                 goto out7;
4827         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4828         if (!ext4_kset)
4829                 goto out6;
4830         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4831         if (!ext4_proc_root)
4832                 goto out5;
4833
4834         err = ext4_init_feat_adverts();
4835         if (err)
4836                 goto out4;
4837
4838         err = ext4_init_mballoc();
4839         if (err)
4840                 goto out3;
4841
4842         err = ext4_init_xattr();
4843         if (err)
4844                 goto out2;
4845         err = init_inodecache();
4846         if (err)
4847                 goto out1;
4848         register_as_ext2();
4849         register_as_ext3();
4850         err = register_filesystem(&ext4_fs_type);
4851         if (err)
4852                 goto out;
4853
4854         ext4_li_info = NULL;
4855         mutex_init(&ext4_li_mtx);
4856         return 0;
4857 out:
4858         unregister_as_ext2();
4859         unregister_as_ext3();
4860         destroy_inodecache();
4861 out1:
4862         ext4_exit_xattr();
4863 out2:
4864         ext4_exit_mballoc();
4865 out3:
4866         ext4_exit_feat_adverts();
4867 out4:
4868         remove_proc_entry("fs/ext4", NULL);
4869 out5:
4870         kset_unregister(ext4_kset);
4871 out6:
4872         ext4_exit_system_zone();
4873 out7:
4874         ext4_exit_pageio();
4875         return err;
4876 }
4877
4878 static void __exit ext4_exit_fs(void)
4879 {
4880         ext4_destroy_lazyinit_thread();
4881         unregister_as_ext2();
4882         unregister_as_ext3();
4883         unregister_filesystem(&ext4_fs_type);
4884         destroy_inodecache();
4885         ext4_exit_xattr();
4886         ext4_exit_mballoc();
4887         ext4_exit_feat_adverts();
4888         remove_proc_entry("fs/ext4", NULL);
4889         kset_unregister(ext4_kset);
4890         ext4_exit_system_zone();
4891         ext4_exit_pageio();
4892 }
4893
4894 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4895 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4896 MODULE_LICENSE("GPL");
4897 module_init(ext4_init_fs)
4898 module_exit(ext4_exit_fs)