fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_jbd2.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "mballoc.h"
  51
  52 #define CREATE_TRACE_POINTS
  53 #include <trace/events/ext4.h>
  54
  55 static struct proc_dir_entry *ext4_proc_root;
  56 static struct kset *ext4_kset;
  57 struct ext4_lazy_init *ext4_li_info;
  58 struct mutex ext4_li_mtx;
  59 struct ext4_features *ext4_feat;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_commit_super(struct super_block *sb, int sync);
  64 static void ext4_mark_recovery_complete(struct super_block *sb,
  65                                         struct ext4_super_block *es);
  66 static void ext4_clear_journal_err(struct super_block *sb,
  67                                    struct ext4_super_block *es);
  68 static int ext4_sync_fs(struct super_block *sb, int wait);
  69 static const char *ext4_decode_error(struct super_block *sb, int errno,
  70                                      char nbuf[16]);
  71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  73 static int ext4_unfreeze(struct super_block *sb);
  74 static void ext4_write_super(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static void ext4_destroy_lazyinit_thread(void);
  79 static void ext4_unregister_li_request(struct super_block *sb);
  80
  81 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  82 static struct file_system_type ext3_fs_type = {
  83         .owner          = THIS_MODULE,
  84         .name           = "ext3",
  85         .mount          = ext4_mount,
  86         .kill_sb        = kill_block_super,
  87         .fs_flags       = FS_REQUIRES_DEV,
  88 };
  89 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  90 #else
  91 #define IS_EXT3_SB(sb) (0)
  92 #endif
  93
  94 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  95                                struct ext4_group_desc *bg)
  96 {
  97         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  98                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  99                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 100 }
 101
 102 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 103                                struct ext4_group_desc *bg)
 104 {
 105         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 106                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 107                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 108 }
 109
 110 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 111                               struct ext4_group_desc *bg)
 112 {
 113         return le32_to_cpu(bg->bg_inode_table_lo) |
 114                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 115                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 116 }
 117
 118 __u32 ext4_free_blks_count(struct super_block *sb,
 119                               struct ext4_group_desc *bg)
 120 {
 121         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 122                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 123                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 124 }
 125
 126 __u32 ext4_free_inodes_count(struct super_block *sb,
 127                               struct ext4_group_desc *bg)
 128 {
 129         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 130                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 131                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 132 }
 133
 134 __u32 ext4_used_dirs_count(struct super_block *sb,
 135                               struct ext4_group_desc *bg)
 136 {
 137         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 138                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 139                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 140 }
 141
 142 __u32 ext4_itable_unused_count(struct super_block *sb,
 143                               struct ext4_group_desc *bg)
 144 {
 145         return le16_to_cpu(bg->bg_itable_unused_lo) |
 146                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 147                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 148 }
 149
 150 void ext4_block_bitmap_set(struct super_block *sb,
 151                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 152 {
 153         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 154         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 155                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 156 }
 157
 158 void ext4_inode_bitmap_set(struct super_block *sb,
 159                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 160 {
 161         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 162         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 163                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 164 }
 165
 166 void ext4_inode_table_set(struct super_block *sb,
 167                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 168 {
 169         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 170         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 171                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 172 }
 173
 174 void ext4_free_blks_set(struct super_block *sb,
 175                           struct ext4_group_desc *bg, __u32 count)
 176 {
 177         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 178         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 179                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 180 }
 181
 182 void ext4_free_inodes_set(struct super_block *sb,
 183                           struct ext4_group_desc *bg, __u32 count)
 184 {
 185         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 186         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 187                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 188 }
 189
 190 void ext4_used_dirs_set(struct super_block *sb,
 191                           struct ext4_group_desc *bg, __u32 count)
 192 {
 193         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 194         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 195                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 196 }
 197
 198 void ext4_itable_unused_set(struct super_block *sb,
 199                           struct ext4_group_desc *bg, __u32 count)
 200 {
 201         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 202         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 203                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 204 }
 205
 206
 207 /* Just increment the non-pointer handle value */
 208 static handle_t *ext4_get_nojournal(void)
 209 {
 210         handle_t *handle = current->journal_info;
 211         unsigned long ref_cnt = (unsigned long)handle;
 212
 213         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 214
 215         ref_cnt++;
 216         handle = (handle_t *)ref_cnt;
 217
 218         current->journal_info = handle;
 219         return handle;
 220 }
 221
 222
 223 /* Decrement the non-pointer handle value */
 224 static void ext4_put_nojournal(handle_t *handle)
 225 {
 226         unsigned long ref_cnt = (unsigned long)handle;
 227
 228         BUG_ON(ref_cnt == 0);
 229
 230         ref_cnt--;
 231         handle = (handle_t *)ref_cnt;
 232
 233         current->journal_info = handle;
 234 }
 235
 236 /*
 237  * Wrappers for jbd2_journal_start/end.
 238  *
 239  * The only special thing we need to do here is to make sure that all
 240  * journal_end calls result in the superblock being marked dirty, so
 241  * that sync() will call the filesystem's write_super callback if
 242  * appropriate.
 243  */
 244 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 245 {
 246         journal_t *journal;
 247
 248         if (sb->s_flags & MS_RDONLY)
 249                 return ERR_PTR(-EROFS);
 250
 251         vfs_check_frozen(sb, SB_FREEZE_TRANS);
 252         /* Special case here: if the journal has aborted behind our
 253          * backs (eg. EIO in the commit thread), then we still need to
 254          * take the FS itself readonly cleanly. */
 255         journal = EXT4_SB(sb)->s_journal;
 256         if (journal) {
 257                 if (is_journal_aborted(journal)) {
 258                         ext4_abort(sb, "Detected aborted journal");
 259                         return ERR_PTR(-EROFS);
 260                 }
 261                 return jbd2_journal_start(journal, nblocks);
 262         }
 263         return ext4_get_nojournal();
 264 }
 265
 266 /*
 267  * The only special thing we need to do here is to make sure that all
 268  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 269  * that sync() will call the filesystem's write_super callback if
 270  * appropriate.
 271  */
 272 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 273 {
 274         struct super_block *sb;
 275         int err;
 276         int rc;
 277
 278         if (!ext4_handle_valid(handle)) {
 279                 ext4_put_nojournal(handle);
 280                 return 0;
 281         }
 282         sb = handle->h_transaction->t_journal->j_private;
 283         err = handle->h_err;
 284         rc = jbd2_journal_stop(handle);
 285
 286         if (!err)
 287                 err = rc;
 288         if (err)
 289                 __ext4_std_error(sb, where, line, err);
 290         return err;
 291 }
 292
 293 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 294                                const char *err_fn, struct buffer_head *bh,
 295                                handle_t *handle, int err)
 296 {
 297         char nbuf[16];
 298         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 299
 300         BUG_ON(!ext4_handle_valid(handle));
 301
 302         if (bh)
 303                 BUFFER_TRACE(bh, "abort");
 304
 305         if (!handle->h_err)
 306                 handle->h_err = err;
 307
 308         if (is_handle_aborted(handle))
 309                 return;
 310
 311         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 312                caller, line, errstr, err_fn);
 313
 314         jbd2_journal_abort_handle(handle);
 315 }
 316
 317 static void __save_error_info(struct super_block *sb, const char *func,
 318                             unsigned int line)
 319 {
 320         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 321
 322         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 324         es->s_last_error_time = cpu_to_le32(get_seconds());
 325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 326         es->s_last_error_line = cpu_to_le32(line);
 327         if (!es->s_first_error_time) {
 328                 es->s_first_error_time = es->s_last_error_time;
 329                 strncpy(es->s_first_error_func, func,
 330                         sizeof(es->s_first_error_func));
 331                 es->s_first_error_line = cpu_to_le32(line);
 332                 es->s_first_error_ino = es->s_last_error_ino;
 333                 es->s_first_error_block = es->s_last_error_block;
 334         }
 335         /*
 336          * Start the daily error reporting function if it hasn't been
 337          * started already
 338          */
 339         if (!es->s_error_count)
 340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 341         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 342 }
 343
 344 static void save_error_info(struct super_block *sb, const char *func,
 345                             unsigned int line)
 346 {
 347         __save_error_info(sb, func, line);
 348         ext4_commit_super(sb, 1);
 349 }
 350
 351
 352 /* Deal with the reporting of failure conditions on a filesystem such as
 353  * inconsistencies detected or read IO failures.
 354  *
 355  * On ext2, we can store the error state of the filesystem in the
 356  * superblock.  That is not possible on ext4, because we may have other
 357  * write ordering constraints on the superblock which prevent us from
 358  * writing it out straight away; and given that the journal is about to
 359  * be aborted, we can't rely on the current, or future, transactions to
 360  * write out the superblock safely.
 361  *
 362  * We'll just use the jbd2_journal_abort() error code to record an error in
 363  * the journal instead.  On recovery, the journal will complain about
 364  * that error until we've noted it down and cleared it.
 365  */
 366
 367 static void ext4_handle_error(struct super_block *sb)
 368 {
 369         if (sb->s_flags & MS_RDONLY)
 370                 return;
 371
 372         if (!test_opt(sb, ERRORS_CONT)) {
 373                 journal_t *journal = EXT4_SB(sb)->s_journal;
 374
 375                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 376                 if (journal)
 377                         jbd2_journal_abort(journal, -EIO);
 378         }
 379         if (test_opt(sb, ERRORS_RO)) {
 380                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 381                 sb->s_flags |= MS_RDONLY;
 382         }
 383         if (test_opt(sb, ERRORS_PANIC))
 384                 panic("EXT4-fs (device %s): panic forced after error\n",
 385                         sb->s_id);
 386 }
 387
 388 void __ext4_error(struct super_block *sb, const char *function,
 389                   unsigned int line, const char *fmt, ...)
 390 {
 391         struct va_format vaf;
 392         va_list args;
 393
 394         va_start(args, fmt);
 395         vaf.fmt = fmt;
 396         vaf.va = &args;
 397         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 398                sb->s_id, function, line, current->comm, &vaf);
 399         va_end(args);
 400
 401         ext4_handle_error(sb);
 402 }
 403
 404 void ext4_error_inode(struct inode *inode, const char *function,
 405                       unsigned int line, ext4_fsblk_t block,
 406                       const char *fmt, ...)
 407 {
 408         va_list args;
 409         struct va_format vaf;
 410         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 411
 412         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 413         es->s_last_error_block = cpu_to_le64(block);
 414         save_error_info(inode->i_sb, function, line);
 415         va_start(args, fmt);
 416         vaf.fmt = fmt;
 417         vaf.va = &args;
 418         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 419                inode->i_sb->s_id, function, line, inode->i_ino);
 420         if (block)
 421                 printk(KERN_CONT "block %llu: ", block);
 422         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 423         va_end(args);
 424
 425         ext4_handle_error(inode->i_sb);
 426 }
 427
 428 void ext4_error_file(struct file *file, const char *function,
 429                      unsigned int line, ext4_fsblk_t block,
 430                      const char *fmt, ...)
 431 {
 432         va_list args;
 433         struct va_format vaf;
 434         struct ext4_super_block *es;
 435         struct inode *inode = file->f_dentry->d_inode;
 436         char pathname[80], *path;
 437
 438         es = EXT4_SB(inode->i_sb)->s_es;
 439         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 440         save_error_info(inode->i_sb, function, line);
 441         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 442         if (IS_ERR(path))
 443                 path = "(unknown)";
 444         printk(KERN_CRIT
 445                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 446                inode->i_sb->s_id, function, line, inode->i_ino);
 447         if (block)
 448                 printk(KERN_CONT "block %llu: ", block);
 449         va_start(args, fmt);
 450         vaf.fmt = fmt;
 451         vaf.va = &args;
 452         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 453         va_end(args);
 454
 455         ext4_handle_error(inode->i_sb);
 456 }
 457
 458 static const char *ext4_decode_error(struct super_block *sb, int errno,
 459                                      char nbuf[16])
 460 {
 461         char *errstr = NULL;
 462
 463         switch (errno) {
 464         case -EIO:
 465                 errstr = "IO failure";
 466                 break;
 467         case -ENOMEM:
 468                 errstr = "Out of memory";
 469                 break;
 470         case -EROFS:
 471                 if (!sb || (EXT4_SB(sb)->s_journal &&
 472                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 473                         errstr = "Journal has aborted";
 474                 else
 475                         errstr = "Readonly filesystem";
 476                 break;
 477         default:
 478                 /* If the caller passed in an extra buffer for unknown
 479                  * errors, textualise them now.  Else we just return
 480                  * NULL. */
 481                 if (nbuf) {
 482                         /* Check for truncated error codes... */
 483                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 484                                 errstr = nbuf;
 485                 }
 486                 break;
 487         }
 488
 489         return errstr;
 490 }
 491
 492 /* __ext4_std_error decodes expected errors from journaling functions
 493  * automatically and invokes the appropriate error response.  */
 494
 495 void __ext4_std_error(struct super_block *sb, const char *function,
 496                       unsigned int line, int errno)
 497 {
 498         char nbuf[16];
 499         const char *errstr;
 500
 501         /* Special case: if the error is EROFS, and we're not already
 502          * inside a transaction, then there's really no point in logging
 503          * an error. */
 504         if (errno == -EROFS && journal_current_handle() == NULL &&
 505             (sb->s_flags & MS_RDONLY))
 506                 return;
 507
 508         errstr = ext4_decode_error(sb, errno, nbuf);
 509         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 510                sb->s_id, function, line, errstr);
 511         save_error_info(sb, function, line);
 512
 513         ext4_handle_error(sb);
 514 }
 515
 516 /*
 517  * ext4_abort is a much stronger failure handler than ext4_error.  The
 518  * abort function may be used to deal with unrecoverable failures such
 519  * as journal IO errors or ENOMEM at a critical moment in log management.
 520  *
 521  * We unconditionally force the filesystem into an ABORT|READONLY state,
 522  * unless the error response on the fs has been set to panic in which
 523  * case we take the easy way out and panic immediately.
 524  */
 525
 526 void __ext4_abort(struct super_block *sb, const char *function,
 527                 unsigned int line, const char *fmt, ...)
 528 {
 529         va_list args;
 530
 531         save_error_info(sb, function, line);
 532         va_start(args, fmt);
 533         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 534                function, line);
 535         vprintk(fmt, args);
 536         printk("\n");
 537         va_end(args);
 538
 539         if ((sb->s_flags & MS_RDONLY) == 0) {
 540                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 541                 sb->s_flags |= MS_RDONLY;
 542                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 543                 if (EXT4_SB(sb)->s_journal)
 544                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 545                 save_error_info(sb, function, line);
 546         }
 547         if (test_opt(sb, ERRORS_PANIC))
 548                 panic("EXT4-fs panic from previous error\n");
 549 }
 550
 551 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 552 {
 553         struct va_format vaf;
 554         va_list args;
 555
 556         va_start(args, fmt);
 557         vaf.fmt = fmt;
 558         vaf.va = &args;
 559         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 560         va_end(args);
 561 }
 562
 563 void __ext4_warning(struct super_block *sb, const char *function,
 564                     unsigned int line, const char *fmt, ...)
 565 {
 566         struct va_format vaf;
 567         va_list args;
 568
 569         va_start(args, fmt);
 570         vaf.fmt = fmt;
 571         vaf.va = &args;
 572         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 573                sb->s_id, function, line, &vaf);
 574         va_end(args);
 575 }
 576
 577 void __ext4_grp_locked_error(const char *function, unsigned int line,
 578                              struct super_block *sb, ext4_group_t grp,
 579                              unsigned long ino, ext4_fsblk_t block,
 580                              const char *fmt, ...)
 581 __releases(bitlock)
 582 __acquires(bitlock)
 583 {
 584         struct va_format vaf;
 585         va_list args;
 586         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 587
 588         es->s_last_error_ino = cpu_to_le32(ino);
 589         es->s_last_error_block = cpu_to_le64(block);
 590         __save_error_info(sb, function, line);
 591
 592         va_start(args, fmt);
 593
 594         vaf.fmt = fmt;
 595         vaf.va = &args;
 596         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u",
 597                sb->s_id, function, line, grp);
 598         if (ino)
 599                 printk(KERN_CONT "inode %lu: ", ino);
 600         if (block)
 601                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 602         printk(KERN_CONT "%pV\n", &vaf);
 603         va_end(args);
 604
 605         if (test_opt(sb, ERRORS_CONT)) {
 606                 ext4_commit_super(sb, 0);
 607                 return;
 608         }
 609
 610         ext4_unlock_group(sb, grp);
 611         ext4_handle_error(sb);
 612         /*
 613          * We only get here in the ERRORS_RO case; relocking the group
 614          * may be dangerous, but nothing bad will happen since the
 615          * filesystem will have already been marked read/only and the
 616          * journal has been aborted.  We return 1 as a hint to callers
 617          * who might what to use the return value from
 618          * ext4_grp_locked_error() to distinguish beween the
 619          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 620          * aggressively from the ext4 function in question, with a
 621          * more appropriate error code.
 622          */
 623         ext4_lock_group(sb, grp);
 624         return;
 625 }
 626
 627 void ext4_update_dynamic_rev(struct super_block *sb)
 628 {
 629         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 630
 631         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 632                 return;
 633
 634         ext4_warning(sb,
 635                      "updating to rev %d because of new feature flag, "
 636                      "running e2fsck is recommended",
 637                      EXT4_DYNAMIC_REV);
 638
 639         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 640         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 641         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 642         /* leave es->s_feature_*compat flags alone */
 643         /* es->s_uuid will be set by e2fsck if empty */
 644
 645         /*
 646          * The rest of the superblock fields should be zero, and if not it
 647          * means they are likely already in use, so leave them alone.  We
 648          * can leave it up to e2fsck to clean up any inconsistencies there.
 649          */
 650 }
 651
 652 /*
 653  * Open the external journal device
 654  */
 655 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 656 {
 657         struct block_device *bdev;
 658         char b[BDEVNAME_SIZE];
 659
 660         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 661         if (IS_ERR(bdev))
 662                 goto fail;
 663         return bdev;
 664
 665 fail:
 666         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 667                         __bdevname(dev, b), PTR_ERR(bdev));
 668         return NULL;
 669 }
 670
 671 /*
 672  * Release the journal device
 673  */
 674 static int ext4_blkdev_put(struct block_device *bdev)
 675 {
 676         bd_release(bdev);
 677         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 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         /*
 822          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 823          * therefore it can be null here.  Don't check it, just initialize
 824          * jinode.
 825          */
 826         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 827         ei->i_reserved_data_blocks = 0;
 828         ei->i_reserved_meta_blocks = 0;
 829         ei->i_allocated_meta_blocks = 0;
 830         ei->i_da_metadata_calc_len = 0;
 831         spin_lock_init(&(ei->i_block_reservation_lock));
 832 #ifdef CONFIG_QUOTA
 833         ei->i_reserved_quota = 0;
 834 #endif
 835         INIT_LIST_HEAD(&ei->i_completed_io_list);
 836         spin_lock_init(&ei->i_completed_io_lock);
 837         ei->cur_aio_dio = NULL;
 838         ei->i_sync_tid = 0;
 839         ei->i_datasync_tid = 0;
 840         atomic_set(&ei->i_ioend_count, 0);
 841
 842         return &ei->vfs_inode;
 843 }
 844
 845 static int ext4_drop_inode(struct inode *inode)
 846 {
 847         int drop = generic_drop_inode(inode);
 848
 849         trace_ext4_drop_inode(inode, drop);
 850         return drop;
 851 }
 852
 853 static void ext4_destroy_inode(struct inode *inode)
 854 {
 855         ext4_ioend_wait(inode);
 856         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 857                 ext4_msg(inode->i_sb, KERN_ERR,
 858                          "Inode %lu (%p): orphan list check failed!",
 859                          inode->i_ino, EXT4_I(inode));
 860                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 861                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 862                                 true);
 863                 dump_stack();
 864         }
 865         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 866 }
 867
 868 static void init_once(void *foo)
 869 {
 870         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 871
 872         INIT_LIST_HEAD(&ei->i_orphan);
 873 #ifdef CONFIG_EXT4_FS_XATTR
 874         init_rwsem(&ei->xattr_sem);
 875 #endif
 876         init_rwsem(&ei->i_data_sem);
 877         inode_init_once(&ei->vfs_inode);
 878 }
 879
 880 static int init_inodecache(void)
 881 {
 882         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 883                                              sizeof(struct ext4_inode_info),
 884                                              0, (SLAB_RECLAIM_ACCOUNT|
 885                                                 SLAB_MEM_SPREAD),
 886                                              init_once);
 887         if (ext4_inode_cachep == NULL)
 888                 return -ENOMEM;
 889         return 0;
 890 }
 891
 892 static void destroy_inodecache(void)
 893 {
 894         kmem_cache_destroy(ext4_inode_cachep);
 895 }
 896
 897 void ext4_clear_inode(struct inode *inode)
 898 {
 899         invalidate_inode_buffers(inode);
 900         end_writeback(inode);
 901         dquot_drop(inode);
 902         ext4_discard_preallocations(inode);
 903         if (EXT4_JOURNAL(inode))
 904                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 905                                        &EXT4_I(inode)->jinode);
 906 }
 907
 908 static inline void ext4_show_quota_options(struct seq_file *seq,
 909                                            struct super_block *sb)
 910 {
 911 #if defined(CONFIG_QUOTA)
 912         struct ext4_sb_info *sbi = EXT4_SB(sb);
 913
 914         if (sbi->s_jquota_fmt) {
 915                 char *fmtname = "";
 916
 917                 switch (sbi->s_jquota_fmt) {
 918                 case QFMT_VFS_OLD:
 919                         fmtname = "vfsold";
 920                         break;
 921                 case QFMT_VFS_V0:
 922                         fmtname = "vfsv0";
 923                         break;
 924                 case QFMT_VFS_V1:
 925                         fmtname = "vfsv1";
 926                         break;
 927                 }
 928                 seq_printf(seq, ",jqfmt=%s", fmtname);
 929         }
 930
 931         if (sbi->s_qf_names[USRQUOTA])
 932                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 933
 934         if (sbi->s_qf_names[GRPQUOTA])
 935                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 936
 937         if (test_opt(sb, USRQUOTA))
 938                 seq_puts(seq, ",usrquota");
 939
 940         if (test_opt(sb, GRPQUOTA))
 941                 seq_puts(seq, ",grpquota");
 942 #endif
 943 }
 944
 945 /*
 946  * Show an option if
 947  *  - it's set to a non-default value OR
 948  *  - if the per-sb default is different from the global default
 949  */
 950 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 951 {
 952         int def_errors;
 953         unsigned long def_mount_opts;
 954         struct super_block *sb = vfs->mnt_sb;
 955         struct ext4_sb_info *sbi = EXT4_SB(sb);
 956         struct ext4_super_block *es = sbi->s_es;
 957
 958         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 959         def_errors     = le16_to_cpu(es->s_errors);
 960
 961         if (sbi->s_sb_block != 1)
 962                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 963         if (test_opt(sb, MINIX_DF))
 964                 seq_puts(seq, ",minixdf");
 965         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 966                 seq_puts(seq, ",grpid");
 967         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 968                 seq_puts(seq, ",nogrpid");
 969         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 970             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 971                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 972         }
 973         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 974             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 975                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 976         }
 977         if (test_opt(sb, ERRORS_RO)) {
 978                 if (def_errors == EXT4_ERRORS_PANIC ||
 979                     def_errors == EXT4_ERRORS_CONTINUE) {
 980                         seq_puts(seq, ",errors=remount-ro");
 981                 }
 982         }
 983         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 984                 seq_puts(seq, ",errors=continue");
 985         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 986                 seq_puts(seq, ",errors=panic");
 987         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 988                 seq_puts(seq, ",nouid32");
 989         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 990                 seq_puts(seq, ",debug");
 991         if (test_opt(sb, OLDALLOC))
 992                 seq_puts(seq, ",oldalloc");
 993 #ifdef CONFIG_EXT4_FS_XATTR
 994         if (test_opt(sb, XATTR_USER) &&
 995                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 996                 seq_puts(seq, ",user_xattr");
 997         if (!test_opt(sb, XATTR_USER) &&
 998             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 999                 seq_puts(seq, ",nouser_xattr");
1000         }
1001 #endif
1002 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1003         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1004                 seq_puts(seq, ",acl");
1005         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1006                 seq_puts(seq, ",noacl");
1007 #endif
1008         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1009                 seq_printf(seq, ",commit=%u",
1010                            (unsigned) (sbi->s_commit_interval / HZ));
1011         }
1012         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1013                 seq_printf(seq, ",min_batch_time=%u",
1014                            (unsigned) sbi->s_min_batch_time);
1015         }
1016         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1017                 seq_printf(seq, ",max_batch_time=%u",
1018                            (unsigned) sbi->s_min_batch_time);
1019         }
1020
1021         /*
1022          * We're changing the default of barrier mount option, so
1023          * let's always display its mount state so it's clear what its
1024          * status is.
1025          */
1026         seq_puts(seq, ",barrier=");
1027         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1028         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1029                 seq_puts(seq, ",journal_async_commit");
1030         else if (test_opt(sb, JOURNAL_CHECKSUM))
1031                 seq_puts(seq, ",journal_checksum");
1032         if (test_opt(sb, I_VERSION))
1033                 seq_puts(seq, ",i_version");
1034         if (!test_opt(sb, DELALLOC) &&
1035             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1036                 seq_puts(seq, ",nodelalloc");
1037
1038         if (test_opt(sb, MBLK_IO_SUBMIT))
1039                 seq_puts(seq, ",mblk_io_submit");
1040         if (sbi->s_stripe)
1041                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1042         /*
1043          * journal mode get enabled in different ways
1044          * So just print the value even if we didn't specify it
1045          */
1046         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1047                 seq_puts(seq, ",data=journal");
1048         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1049                 seq_puts(seq, ",data=ordered");
1050         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1051                 seq_puts(seq, ",data=writeback");
1052
1053         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1054                 seq_printf(seq, ",inode_readahead_blks=%u",
1055                            sbi->s_inode_readahead_blks);
1056
1057         if (test_opt(sb, DATA_ERR_ABORT))
1058                 seq_puts(seq, ",data_err=abort");
1059
1060         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1061                 seq_puts(seq, ",noauto_da_alloc");
1062
1063         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1064                 seq_puts(seq, ",discard");
1065
1066         if (test_opt(sb, NOLOAD))
1067                 seq_puts(seq, ",norecovery");
1068
1069         if (test_opt(sb, DIOREAD_NOLOCK))
1070                 seq_puts(seq, ",dioread_nolock");
1071
1072         if (test_opt(sb, BLOCK_VALIDITY) &&
1073             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1074                 seq_puts(seq, ",block_validity");
1075
1076         if (!test_opt(sb, INIT_INODE_TABLE))
1077                 seq_puts(seq, ",noinit_inode_table");
1078         else if (sbi->s_li_wait_mult)
1079                 seq_printf(seq, ",init_inode_table=%u",
1080                            (unsigned) sbi->s_li_wait_mult);
1081
1082         ext4_show_quota_options(seq, sb);
1083
1084         return 0;
1085 }
1086
1087 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1088                                         u64 ino, u32 generation)
1089 {
1090         struct inode *inode;
1091
1092         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1093                 return ERR_PTR(-ESTALE);
1094         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1095                 return ERR_PTR(-ESTALE);
1096
1097         /* iget isn't really right if the inode is currently unallocated!!
1098          *
1099          * ext4_read_inode will return a bad_inode if the inode had been
1100          * deleted, so we should be safe.
1101          *
1102          * Currently we don't know the generation for parent directory, so
1103          * a generation of 0 means "accept any"
1104          */
1105         inode = ext4_iget(sb, ino);
1106         if (IS_ERR(inode))
1107                 return ERR_CAST(inode);
1108         if (generation && inode->i_generation != generation) {
1109                 iput(inode);
1110                 return ERR_PTR(-ESTALE);
1111         }
1112
1113         return inode;
1114 }
1115
1116 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1117                                         int fh_len, int fh_type)
1118 {
1119         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1120                                     ext4_nfs_get_inode);
1121 }
1122
1123 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1124                                         int fh_len, int fh_type)
1125 {
1126         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1127                                     ext4_nfs_get_inode);
1128 }
1129
1130 /*
1131  * Try to release metadata pages (indirect blocks, directories) which are
1132  * mapped via the block device.  Since these pages could have journal heads
1133  * which would prevent try_to_free_buffers() from freeing them, we must use
1134  * jbd2 layer's try_to_free_buffers() function to release them.
1135  */
1136 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1137                                  gfp_t wait)
1138 {
1139         journal_t *journal = EXT4_SB(sb)->s_journal;
1140
1141         WARN_ON(PageChecked(page));
1142         if (!page_has_buffers(page))
1143                 return 0;
1144         if (journal)
1145                 return jbd2_journal_try_to_free_buffers(journal, page,
1146                                                         wait & ~__GFP_WAIT);
1147         return try_to_free_buffers(page);
1148 }
1149
1150 #ifdef CONFIG_QUOTA
1151 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1152 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1153
1154 static int ext4_write_dquot(struct dquot *dquot);
1155 static int ext4_acquire_dquot(struct dquot *dquot);
1156 static int ext4_release_dquot(struct dquot *dquot);
1157 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1158 static int ext4_write_info(struct super_block *sb, int type);
1159 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1160                                 char *path);
1161 static int ext4_quota_off(struct super_block *sb, int type);
1162 static int ext4_quota_on_mount(struct super_block *sb, int type);
1163 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1164                                size_t len, loff_t off);
1165 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1166                                 const char *data, size_t len, loff_t off);
1167
1168 static const struct dquot_operations ext4_quota_operations = {
1169 #ifdef CONFIG_QUOTA
1170         .get_reserved_space = ext4_get_reserved_space,
1171 #endif
1172         .write_dquot    = ext4_write_dquot,
1173         .acquire_dquot  = ext4_acquire_dquot,
1174         .release_dquot  = ext4_release_dquot,
1175         .mark_dirty     = ext4_mark_dquot_dirty,
1176         .write_info     = ext4_write_info,
1177         .alloc_dquot    = dquot_alloc,
1178         .destroy_dquot  = dquot_destroy,
1179 };
1180
1181 static const struct quotactl_ops ext4_qctl_operations = {
1182         .quota_on       = ext4_quota_on,
1183         .quota_off      = ext4_quota_off,
1184         .quota_sync     = dquot_quota_sync,
1185         .get_info       = dquot_get_dqinfo,
1186         .set_info       = dquot_set_dqinfo,
1187         .get_dqblk      = dquot_get_dqblk,
1188         .set_dqblk      = dquot_set_dqblk
1189 };
1190 #endif
1191
1192 static const struct super_operations ext4_sops = {
1193         .alloc_inode    = ext4_alloc_inode,
1194         .destroy_inode  = ext4_destroy_inode,
1195         .write_inode    = ext4_write_inode,
1196         .dirty_inode    = ext4_dirty_inode,
1197         .drop_inode     = ext4_drop_inode,
1198         .evict_inode    = ext4_evict_inode,
1199         .put_super      = ext4_put_super,
1200         .sync_fs        = ext4_sync_fs,
1201         .freeze_fs      = ext4_freeze,
1202         .unfreeze_fs    = ext4_unfreeze,
1203         .statfs         = ext4_statfs,
1204         .remount_fs     = ext4_remount,
1205         .show_options   = ext4_show_options,
1206 #ifdef CONFIG_QUOTA
1207         .quota_read     = ext4_quota_read,
1208         .quota_write    = ext4_quota_write,
1209 #endif
1210         .bdev_try_to_free_page = bdev_try_to_free_page,
1211 };
1212
1213 static const struct super_operations ext4_nojournal_sops = {
1214         .alloc_inode    = ext4_alloc_inode,
1215         .destroy_inode  = ext4_destroy_inode,
1216         .write_inode    = ext4_write_inode,
1217         .dirty_inode    = ext4_dirty_inode,
1218         .drop_inode     = ext4_drop_inode,
1219         .evict_inode    = ext4_evict_inode,
1220         .write_super    = ext4_write_super,
1221         .put_super      = ext4_put_super,
1222         .statfs         = ext4_statfs,
1223         .remount_fs     = ext4_remount,
1224         .show_options   = ext4_show_options,
1225 #ifdef CONFIG_QUOTA
1226         .quota_read     = ext4_quota_read,
1227         .quota_write    = ext4_quota_write,
1228 #endif
1229         .bdev_try_to_free_page = bdev_try_to_free_page,
1230 };
1231
1232 static const struct export_operations ext4_export_ops = {
1233         .fh_to_dentry = ext4_fh_to_dentry,
1234         .fh_to_parent = ext4_fh_to_parent,
1235         .get_parent = ext4_get_parent,
1236 };
1237
1238 enum {
1239         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1240         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1241         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1242         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1243         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1244         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1245         Opt_journal_update, Opt_journal_dev,
1246         Opt_journal_checksum, Opt_journal_async_commit,
1247         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1248         Opt_data_err_abort, Opt_data_err_ignore,
1249         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1250         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1251         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1252         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1253         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1254         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1255         Opt_inode_readahead_blks, Opt_journal_ioprio,
1256         Opt_dioread_nolock, Opt_dioread_lock,
1257         Opt_discard, Opt_nodiscard,
1258         Opt_init_inode_table, Opt_noinit_inode_table,
1259 };
1260
1261 static const match_table_t tokens = {
1262         {Opt_bsd_df, "bsddf"},
1263         {Opt_minix_df, "minixdf"},
1264         {Opt_grpid, "grpid"},
1265         {Opt_grpid, "bsdgroups"},
1266         {Opt_nogrpid, "nogrpid"},
1267         {Opt_nogrpid, "sysvgroups"},
1268         {Opt_resgid, "resgid=%u"},
1269         {Opt_resuid, "resuid=%u"},
1270         {Opt_sb, "sb=%u"},
1271         {Opt_err_cont, "errors=continue"},
1272         {Opt_err_panic, "errors=panic"},
1273         {Opt_err_ro, "errors=remount-ro"},
1274         {Opt_nouid32, "nouid32"},
1275         {Opt_debug, "debug"},
1276         {Opt_oldalloc, "oldalloc"},
1277         {Opt_orlov, "orlov"},
1278         {Opt_user_xattr, "user_xattr"},
1279         {Opt_nouser_xattr, "nouser_xattr"},
1280         {Opt_acl, "acl"},
1281         {Opt_noacl, "noacl"},
1282         {Opt_noload, "noload"},
1283         {Opt_noload, "norecovery"},
1284         {Opt_nobh, "nobh"},
1285         {Opt_bh, "bh"},
1286         {Opt_commit, "commit=%u"},
1287         {Opt_min_batch_time, "min_batch_time=%u"},
1288         {Opt_max_batch_time, "max_batch_time=%u"},
1289         {Opt_journal_update, "journal=update"},
1290         {Opt_journal_dev, "journal_dev=%u"},
1291         {Opt_journal_checksum, "journal_checksum"},
1292         {Opt_journal_async_commit, "journal_async_commit"},
1293         {Opt_abort, "abort"},
1294         {Opt_data_journal, "data=journal"},
1295         {Opt_data_ordered, "data=ordered"},
1296         {Opt_data_writeback, "data=writeback"},
1297         {Opt_data_err_abort, "data_err=abort"},
1298         {Opt_data_err_ignore, "data_err=ignore"},
1299         {Opt_offusrjquota, "usrjquota="},
1300         {Opt_usrjquota, "usrjquota=%s"},
1301         {Opt_offgrpjquota, "grpjquota="},
1302         {Opt_grpjquota, "grpjquota=%s"},
1303         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1304         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1305         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1306         {Opt_grpquota, "grpquota"},
1307         {Opt_noquota, "noquota"},
1308         {Opt_quota, "quota"},
1309         {Opt_usrquota, "usrquota"},
1310         {Opt_barrier, "barrier=%u"},
1311         {Opt_barrier, "barrier"},
1312         {Opt_nobarrier, "nobarrier"},
1313         {Opt_i_version, "i_version"},
1314         {Opt_stripe, "stripe=%u"},
1315         {Opt_resize, "resize"},
1316         {Opt_delalloc, "delalloc"},
1317         {Opt_nodelalloc, "nodelalloc"},
1318         {Opt_mblk_io_submit, "mblk_io_submit"},
1319         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1320         {Opt_block_validity, "block_validity"},
1321         {Opt_noblock_validity, "noblock_validity"},
1322         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1323         {Opt_journal_ioprio, "journal_ioprio=%u"},
1324         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1325         {Opt_auto_da_alloc, "auto_da_alloc"},
1326         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1327         {Opt_dioread_nolock, "dioread_nolock"},
1328         {Opt_dioread_lock, "dioread_lock"},
1329         {Opt_discard, "discard"},
1330         {Opt_nodiscard, "nodiscard"},
1331         {Opt_init_inode_table, "init_itable=%u"},
1332         {Opt_init_inode_table, "init_itable"},
1333         {Opt_noinit_inode_table, "noinit_itable"},
1334         {Opt_err, NULL},
1335 };
1336
1337 static ext4_fsblk_t get_sb_block(void **data)
1338 {
1339         ext4_fsblk_t    sb_block;
1340         char            *options = (char *) *data;
1341
1342         if (!options || strncmp(options, "sb=", 3) != 0)
1343                 return 1;       /* Default location */
1344
1345         options += 3;
1346         /* TODO: use simple_strtoll with >32bit ext4 */
1347         sb_block = simple_strtoul(options, &options, 0);
1348         if (*options && *options != ',') {
1349                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1350                        (char *) *data);
1351                 return 1;
1352         }
1353         if (*options == ',')
1354                 options++;
1355         *data = (void *) options;
1356
1357         return sb_block;
1358 }
1359
1360 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1361 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1362         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1363
1364 #ifdef CONFIG_QUOTA
1365 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1366 {
1367         struct ext4_sb_info *sbi = EXT4_SB(sb);
1368         char *qname;
1369
1370         if (sb_any_quota_loaded(sb) &&
1371                 !sbi->s_qf_names[qtype]) {
1372                 ext4_msg(sb, KERN_ERR,
1373                         "Cannot change journaled "
1374                         "quota options when quota turned on");
1375                 return 0;
1376         }
1377         qname = match_strdup(args);
1378         if (!qname) {
1379                 ext4_msg(sb, KERN_ERR,
1380                         "Not enough memory for storing quotafile name");
1381                 return 0;
1382         }
1383         if (sbi->s_qf_names[qtype] &&
1384                 strcmp(sbi->s_qf_names[qtype], qname)) {
1385                 ext4_msg(sb, KERN_ERR,
1386                         "%s quota file already specified", QTYPE2NAME(qtype));
1387                 kfree(qname);
1388                 return 0;
1389         }
1390         sbi->s_qf_names[qtype] = qname;
1391         if (strchr(sbi->s_qf_names[qtype], '/')) {
1392                 ext4_msg(sb, KERN_ERR,
1393                         "quotafile must be on filesystem root");
1394                 kfree(sbi->s_qf_names[qtype]);
1395                 sbi->s_qf_names[qtype] = NULL;
1396                 return 0;
1397         }
1398         set_opt(sb, QUOTA);
1399         return 1;
1400 }
1401
1402 static int clear_qf_name(struct super_block *sb, int qtype)
1403 {
1404
1405         struct ext4_sb_info *sbi = EXT4_SB(sb);
1406
1407         if (sb_any_quota_loaded(sb) &&
1408                 sbi->s_qf_names[qtype]) {
1409                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1410                         " when quota turned on");
1411                 return 0;
1412         }
1413         /*
1414          * The space will be released later when all options are confirmed
1415          * to be correct
1416          */
1417         sbi->s_qf_names[qtype] = NULL;
1418         return 1;
1419 }
1420 #endif
1421
1422 static int parse_options(char *options, struct super_block *sb,
1423                          unsigned long *journal_devnum,
1424                          unsigned int *journal_ioprio,
1425                          ext4_fsblk_t *n_blocks_count, int is_remount)
1426 {
1427         struct ext4_sb_info *sbi = EXT4_SB(sb);
1428         char *p;
1429         substring_t args[MAX_OPT_ARGS];
1430         int data_opt = 0;
1431         int option;
1432 #ifdef CONFIG_QUOTA
1433         int qfmt;
1434 #endif
1435
1436         if (!options)
1437                 return 1;
1438
1439         while ((p = strsep(&options, ",")) != NULL) {
1440                 int token;
1441                 if (!*p)
1442                         continue;
1443
1444                 /*
1445                  * Initialize args struct so we know whether arg was
1446                  * found; some options take optional arguments.
1447                  */
1448                 args[0].to = args[0].from = 0;
1449                 token = match_token(p, tokens, args);
1450                 switch (token) {
1451                 case Opt_bsd_df:
1452                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1453                         clear_opt(sb, MINIX_DF);
1454                         break;
1455                 case Opt_minix_df:
1456                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1457                         set_opt(sb, MINIX_DF);
1458
1459                         break;
1460                 case Opt_grpid:
1461                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1462                         set_opt(sb, GRPID);
1463
1464                         break;
1465                 case Opt_nogrpid:
1466                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1467                         clear_opt(sb, GRPID);
1468
1469                         break;
1470                 case Opt_resuid:
1471                         if (match_int(&args[0], &option))
1472                                 return 0;
1473                         sbi->s_resuid = option;
1474                         break;
1475                 case Opt_resgid:
1476                         if (match_int(&args[0], &option))
1477                                 return 0;
1478                         sbi->s_resgid = option;
1479                         break;
1480                 case Opt_sb:
1481                         /* handled by get_sb_block() instead of here */
1482                         /* *sb_block = match_int(&args[0]); */
1483                         break;
1484                 case Opt_err_panic:
1485                         clear_opt(sb, ERRORS_CONT);
1486                         clear_opt(sb, ERRORS_RO);
1487                         set_opt(sb, ERRORS_PANIC);
1488                         break;
1489                 case Opt_err_ro:
1490                         clear_opt(sb, ERRORS_CONT);
1491                         clear_opt(sb, ERRORS_PANIC);
1492                         set_opt(sb, ERRORS_RO);
1493                         break;
1494                 case Opt_err_cont:
1495                         clear_opt(sb, ERRORS_RO);
1496                         clear_opt(sb, ERRORS_PANIC);
1497                         set_opt(sb, ERRORS_CONT);
1498                         break;
1499                 case Opt_nouid32:
1500                         set_opt(sb, NO_UID32);
1501                         break;
1502                 case Opt_debug:
1503                         set_opt(sb, DEBUG);
1504                         break;
1505                 case Opt_oldalloc:
1506                         set_opt(sb, OLDALLOC);
1507                         break;
1508                 case Opt_orlov:
1509                         clear_opt(sb, OLDALLOC);
1510                         break;
1511 #ifdef CONFIG_EXT4_FS_XATTR
1512                 case Opt_user_xattr:
1513                         set_opt(sb, XATTR_USER);
1514                         break;
1515                 case Opt_nouser_xattr:
1516                         clear_opt(sb, XATTR_USER);
1517                         break;
1518 #else
1519                 case Opt_user_xattr:
1520                 case Opt_nouser_xattr:
1521                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1522                         break;
1523 #endif
1524 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1525                 case Opt_acl:
1526                         set_opt(sb, POSIX_ACL);
1527                         break;
1528                 case Opt_noacl:
1529                         clear_opt(sb, POSIX_ACL);
1530                         break;
1531 #else
1532                 case Opt_acl:
1533                 case Opt_noacl:
1534                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1535                         break;
1536 #endif
1537                 case Opt_journal_update:
1538                         /* @@@ FIXME */
1539                         /* Eventually we will want to be able to create
1540                            a journal file here.  For now, only allow the
1541                            user to specify an existing inode to be the
1542                            journal file. */
1543                         if (is_remount) {
1544                                 ext4_msg(sb, KERN_ERR,
1545                                          "Cannot specify journal on remount");
1546                                 return 0;
1547                         }
1548                         set_opt(sb, UPDATE_JOURNAL);
1549                         break;
1550                 case Opt_journal_dev:
1551                         if (is_remount) {
1552                                 ext4_msg(sb, KERN_ERR,
1553                                         "Cannot specify journal on remount");
1554                                 return 0;
1555                         }
1556                         if (match_int(&args[0], &option))
1557                                 return 0;
1558                         *journal_devnum = option;
1559                         break;
1560                 case Opt_journal_checksum:
1561                         set_opt(sb, JOURNAL_CHECKSUM);
1562                         break;
1563                 case Opt_journal_async_commit:
1564                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1565                         set_opt(sb, JOURNAL_CHECKSUM);
1566                         break;
1567                 case Opt_noload:
1568                         set_opt(sb, NOLOAD);
1569                         break;
1570                 case Opt_commit:
1571                         if (match_int(&args[0], &option))
1572                                 return 0;
1573                         if (option < 0)
1574                                 return 0;
1575                         if (option == 0)
1576                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1577                         sbi->s_commit_interval = HZ * option;
1578                         break;
1579                 case Opt_max_batch_time:
1580                         if (match_int(&args[0], &option))
1581                                 return 0;
1582                         if (option < 0)
1583                                 return 0;
1584                         if (option == 0)
1585                                 option = EXT4_DEF_MAX_BATCH_TIME;
1586                         sbi->s_max_batch_time = option;
1587                         break;
1588                 case Opt_min_batch_time:
1589                         if (match_int(&args[0], &option))
1590                                 return 0;
1591                         if (option < 0)
1592                                 return 0;
1593                         sbi->s_min_batch_time = option;
1594                         break;
1595                 case Opt_data_journal:
1596                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1597                         goto datacheck;
1598                 case Opt_data_ordered:
1599                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1600                         goto datacheck;
1601                 case Opt_data_writeback:
1602                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1603                 datacheck:
1604                         if (is_remount) {
1605                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1606                                         ext4_msg(sb, KERN_ERR,
1607                                                 "Cannot change data mode on remount");
1608                                         return 0;
1609                                 }
1610                         } else {
1611                                 clear_opt(sb, DATA_FLAGS);
1612                                 sbi->s_mount_opt |= data_opt;
1613                         }
1614                         break;
1615                 case Opt_data_err_abort:
1616                         set_opt(sb, DATA_ERR_ABORT);
1617                         break;
1618                 case Opt_data_err_ignore:
1619                         clear_opt(sb, DATA_ERR_ABORT);
1620                         break;
1621 #ifdef CONFIG_QUOTA
1622                 case Opt_usrjquota:
1623                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1624                                 return 0;
1625                         break;
1626                 case Opt_grpjquota:
1627                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1628                                 return 0;
1629                         break;
1630                 case Opt_offusrjquota:
1631                         if (!clear_qf_name(sb, USRQUOTA))
1632                                 return 0;
1633                         break;
1634                 case Opt_offgrpjquota:
1635                         if (!clear_qf_name(sb, GRPQUOTA))
1636                                 return 0;
1637                         break;
1638
1639                 case Opt_jqfmt_vfsold:
1640                         qfmt = QFMT_VFS_OLD;
1641                         goto set_qf_format;
1642                 case Opt_jqfmt_vfsv0:
1643                         qfmt = QFMT_VFS_V0;
1644                         goto set_qf_format;
1645                 case Opt_jqfmt_vfsv1:
1646                         qfmt = QFMT_VFS_V1;
1647 set_qf_format:
1648                         if (sb_any_quota_loaded(sb) &&
1649                             sbi->s_jquota_fmt != qfmt) {
1650                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1651                                         "journaled quota options when "
1652                                         "quota turned on");
1653                                 return 0;
1654                         }
1655                         sbi->s_jquota_fmt = qfmt;
1656                         break;
1657                 case Opt_quota:
1658                 case Opt_usrquota:
1659                         set_opt(sb, QUOTA);
1660                         set_opt(sb, USRQUOTA);
1661                         break;
1662                 case Opt_grpquota:
1663                         set_opt(sb, QUOTA);
1664                         set_opt(sb, GRPQUOTA);
1665                         break;
1666                 case Opt_noquota:
1667                         if (sb_any_quota_loaded(sb)) {
1668                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1669                                         "options when quota turned on");
1670                                 return 0;
1671                         }
1672                         clear_opt(sb, QUOTA);
1673                         clear_opt(sb, USRQUOTA);
1674                         clear_opt(sb, GRPQUOTA);
1675                         break;
1676 #else
1677                 case Opt_quota:
1678                 case Opt_usrquota:
1679                 case Opt_grpquota:
1680                         ext4_msg(sb, KERN_ERR,
1681                                 "quota options not supported");
1682                         break;
1683                 case Opt_usrjquota:
1684                 case Opt_grpjquota:
1685                 case Opt_offusrjquota:
1686                 case Opt_offgrpjquota:
1687                 case Opt_jqfmt_vfsold:
1688                 case Opt_jqfmt_vfsv0:
1689                 case Opt_jqfmt_vfsv1:
1690                         ext4_msg(sb, KERN_ERR,
1691                                 "journaled quota options not supported");
1692                         break;
1693                 case Opt_noquota:
1694                         break;
1695 #endif
1696                 case Opt_abort:
1697                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1698                         break;
1699                 case Opt_nobarrier:
1700                         clear_opt(sb, BARRIER);
1701                         break;
1702                 case Opt_barrier:
1703                         if (args[0].from) {
1704                                 if (match_int(&args[0], &option))
1705                                         return 0;
1706                         } else
1707                                 option = 1;     /* No argument, default to 1 */
1708                         if (option)
1709                                 set_opt(sb, BARRIER);
1710                         else
1711                                 clear_opt(sb, BARRIER);
1712                         break;
1713                 case Opt_ignore:
1714                         break;
1715                 case Opt_resize:
1716                         if (!is_remount) {
1717                                 ext4_msg(sb, KERN_ERR,
1718                                         "resize option only available "
1719                                         "for remount");
1720                                 return 0;
1721                         }
1722                         if (match_int(&args[0], &option) != 0)
1723                                 return 0;
1724                         *n_blocks_count = option;
1725                         break;
1726                 case Opt_nobh:
1727                         ext4_msg(sb, KERN_WARNING,
1728                                  "Ignoring deprecated nobh option");
1729                         break;
1730                 case Opt_bh:
1731                         ext4_msg(sb, KERN_WARNING,
1732                                  "Ignoring deprecated bh option");
1733                         break;
1734                 case Opt_i_version:
1735                         set_opt(sb, I_VERSION);
1736                         sb->s_flags |= MS_I_VERSION;
1737                         break;
1738                 case Opt_nodelalloc:
1739                         clear_opt(sb, DELALLOC);
1740                         break;
1741                 case Opt_mblk_io_submit:
1742                         set_opt(sb, MBLK_IO_SUBMIT);
1743                         break;
1744                 case Opt_nomblk_io_submit:
1745                         clear_opt(sb, MBLK_IO_SUBMIT);
1746                         break;
1747                 case Opt_stripe:
1748                         if (match_int(&args[0], &option))
1749                                 return 0;
1750                         if (option < 0)
1751                                 return 0;
1752                         sbi->s_stripe = option;
1753                         break;
1754                 case Opt_delalloc:
1755                         set_opt(sb, DELALLOC);
1756                         break;
1757                 case Opt_block_validity:
1758                         set_opt(sb, BLOCK_VALIDITY);
1759                         break;
1760                 case Opt_noblock_validity:
1761                         clear_opt(sb, BLOCK_VALIDITY);
1762                         break;
1763                 case Opt_inode_readahead_blks:
1764                         if (match_int(&args[0], &option))
1765                                 return 0;
1766                         if (option < 0 || option > (1 << 30))
1767                                 return 0;
1768                         if (!is_power_of_2(option)) {
1769                                 ext4_msg(sb, KERN_ERR,
1770                                          "EXT4-fs: inode_readahead_blks"
1771                                          " must be a power of 2");
1772                                 return 0;
1773                         }
1774                         sbi->s_inode_readahead_blks = option;
1775                         break;
1776                 case Opt_journal_ioprio:
1777                         if (match_int(&args[0], &option))
1778                                 return 0;
1779                         if (option < 0 || option > 7)
1780                                 break;
1781                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1782                                                             option);
1783                         break;
1784                 case Opt_noauto_da_alloc:
1785                         set_opt(sb, NO_AUTO_DA_ALLOC);
1786                         break;
1787                 case Opt_auto_da_alloc:
1788                         if (args[0].from) {
1789                                 if (match_int(&args[0], &option))
1790                                         return 0;
1791                         } else
1792                                 option = 1;     /* No argument, default to 1 */
1793                         if (option)
1794                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1795                         else
1796                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1797                         break;
1798                 case Opt_discard:
1799                         set_opt(sb, DISCARD);
1800                         break;
1801                 case Opt_nodiscard:
1802                         clear_opt(sb, DISCARD);
1803                         break;
1804                 case Opt_dioread_nolock:
1805                         set_opt(sb, DIOREAD_NOLOCK);
1806                         break;
1807                 case Opt_dioread_lock:
1808                         clear_opt(sb, DIOREAD_NOLOCK);
1809                         break;
1810                 case Opt_init_inode_table:
1811                         set_opt(sb, INIT_INODE_TABLE);
1812                         if (args[0].from) {
1813                                 if (match_int(&args[0], &option))
1814                                         return 0;
1815                         } else
1816                                 option = EXT4_DEF_LI_WAIT_MULT;
1817                         if (option < 0)
1818                                 return 0;
1819                         sbi->s_li_wait_mult = option;
1820                         break;
1821                 case Opt_noinit_inode_table:
1822                         clear_opt(sb, INIT_INODE_TABLE);
1823                         break;
1824                 default:
1825                         ext4_msg(sb, KERN_ERR,
1826                                "Unrecognized mount option \"%s\" "
1827                                "or missing value", p);
1828                         return 0;
1829                 }
1830         }
1831 #ifdef CONFIG_QUOTA
1832         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1833                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1834                         clear_opt(sb, USRQUOTA);
1835
1836                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1837                         clear_opt(sb, GRPQUOTA);
1838
1839                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1840                         ext4_msg(sb, KERN_ERR, "old and new quota "
1841                                         "format mixing");
1842                         return 0;
1843                 }
1844
1845                 if (!sbi->s_jquota_fmt) {
1846                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1847                                         "not specified");
1848                         return 0;
1849                 }
1850         } else {
1851                 if (sbi->s_jquota_fmt) {
1852                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1853                                         "specified with no journaling "
1854                                         "enabled");
1855                         return 0;
1856                 }
1857         }
1858 #endif
1859         return 1;
1860 }
1861
1862 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1863                             int read_only)
1864 {
1865         struct ext4_sb_info *sbi = EXT4_SB(sb);
1866         int res = 0;
1867
1868         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1869                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1870                          "forcing read-only mode");
1871                 res = MS_RDONLY;
1872         }
1873         if (read_only)
1874                 return res;
1875         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1876                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1877                          "running e2fsck is recommended");
1878         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1879                 ext4_msg(sb, KERN_WARNING,
1880                          "warning: mounting fs with errors, "
1881                          "running e2fsck is recommended");
1882         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1883                  le16_to_cpu(es->s_mnt_count) >=
1884                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1885                 ext4_msg(sb, KERN_WARNING,
1886                          "warning: maximal mount count reached, "
1887                          "running e2fsck is recommended");
1888         else if (le32_to_cpu(es->s_checkinterval) &&
1889                 (le32_to_cpu(es->s_lastcheck) +
1890                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1891                 ext4_msg(sb, KERN_WARNING,
1892                          "warning: checktime reached, "
1893                          "running e2fsck is recommended");
1894         if (!sbi->s_journal)
1895                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1896         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1897                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1898         le16_add_cpu(&es->s_mnt_count, 1);
1899         es->s_mtime = cpu_to_le32(get_seconds());
1900         ext4_update_dynamic_rev(sb);
1901         if (sbi->s_journal)
1902                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1903
1904         ext4_commit_super(sb, 1);
1905         if (test_opt(sb, DEBUG))
1906                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1907                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1908                         sb->s_blocksize,
1909                         sbi->s_groups_count,
1910                         EXT4_BLOCKS_PER_GROUP(sb),
1911                         EXT4_INODES_PER_GROUP(sb),
1912                         sbi->s_mount_opt, sbi->s_mount_opt2);
1913
1914         return res;
1915 }
1916
1917 static int ext4_fill_flex_info(struct super_block *sb)
1918 {
1919         struct ext4_sb_info *sbi = EXT4_SB(sb);
1920         struct ext4_group_desc *gdp = NULL;
1921         ext4_group_t flex_group_count;
1922         ext4_group_t flex_group;
1923         int groups_per_flex = 0;
1924         size_t size;
1925         int i;
1926
1927         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1928         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1929
1930         if (groups_per_flex < 2) {
1931                 sbi->s_log_groups_per_flex = 0;
1932                 return 1;
1933         }
1934
1935         /* We allocate both existing and potentially added groups */
1936         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1937                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1938                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1939         size = flex_group_count * sizeof(struct flex_groups);
1940         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1941         if (sbi->s_flex_groups == NULL) {
1942                 sbi->s_flex_groups = vzalloc(size);
1943                 if (sbi->s_flex_groups == NULL) {
1944                         ext4_msg(sb, KERN_ERR,
1945                                  "not enough memory for %u flex groups",
1946                                  flex_group_count);
1947                         goto failed;
1948                 }
1949         }
1950
1951         for (i = 0; i < sbi->s_groups_count; i++) {
1952                 gdp = ext4_get_group_desc(sb, i, NULL);
1953
1954                 flex_group = ext4_flex_group(sbi, i);
1955                 atomic_add(ext4_free_inodes_count(sb, gdp),
1956                            &sbi->s_flex_groups[flex_group].free_inodes);
1957                 atomic_add(ext4_free_blks_count(sb, gdp),
1958                            &sbi->s_flex_groups[flex_group].free_blocks);
1959                 atomic_add(ext4_used_dirs_count(sb, gdp),
1960                            &sbi->s_flex_groups[flex_group].used_dirs);
1961         }
1962
1963         return 1;
1964 failed:
1965         return 0;
1966 }
1967
1968 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1969                             struct ext4_group_desc *gdp)
1970 {
1971         __u16 crc = 0;
1972
1973         if (sbi->s_es->s_feature_ro_compat &
1974             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1975                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1976                 __le32 le_group = cpu_to_le32(block_group);
1977
1978                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1979                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1980                 crc = crc16(crc, (__u8 *)gdp, offset);
1981                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1982                 /* for checksum of struct ext4_group_desc do the rest...*/
1983                 if ((sbi->s_es->s_feature_incompat &
1984                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1985                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1986                         crc = crc16(crc, (__u8 *)gdp + offset,
1987                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1988                                         offset);
1989         }
1990
1991         return cpu_to_le16(crc);
1992 }
1993
1994 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1995                                 struct ext4_group_desc *gdp)
1996 {
1997         if ((sbi->s_es->s_feature_ro_compat &
1998              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1999             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2000                 return 0;
2001
2002         return 1;
2003 }
2004
2005 /* Called at mount-time, super-block is locked */
2006 static int ext4_check_descriptors(struct super_block *sb,
2007                                   ext4_group_t *first_not_zeroed)
2008 {
2009         struct ext4_sb_info *sbi = EXT4_SB(sb);
2010         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2011         ext4_fsblk_t last_block;
2012         ext4_fsblk_t block_bitmap;
2013         ext4_fsblk_t inode_bitmap;
2014         ext4_fsblk_t inode_table;
2015         int flexbg_flag = 0;
2016         ext4_group_t i, grp = sbi->s_groups_count;
2017
2018         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2019                 flexbg_flag = 1;
2020
2021         ext4_debug("Checking group descriptors");
2022
2023         for (i = 0; i < sbi->s_groups_count; i++) {
2024                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2025
2026                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2027                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2028                 else
2029                         last_block = first_block +
2030                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2031
2032                 if ((grp == sbi->s_groups_count) &&
2033                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2034                         grp = i;
2035
2036                 block_bitmap = ext4_block_bitmap(sb, gdp);
2037                 if (block_bitmap < first_block || block_bitmap > last_block) {
2038                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2039                                "Block bitmap for group %u not in group "
2040                                "(block %llu)!", i, block_bitmap);
2041                         return 0;
2042                 }
2043                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2044                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2045                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2046                                "Inode bitmap for group %u not in group "
2047                                "(block %llu)!", i, inode_bitmap);
2048                         return 0;
2049                 }
2050                 inode_table = ext4_inode_table(sb, gdp);
2051                 if (inode_table < first_block ||
2052                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2053                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2054                                "Inode table for group %u not in group "
2055                                "(block %llu)!", i, inode_table);
2056                         return 0;
2057                 }
2058                 ext4_lock_group(sb, i);
2059                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2060                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2061                                  "Checksum for group %u failed (%u!=%u)",
2062                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2063                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2064                         if (!(sb->s_flags & MS_RDONLY)) {
2065                                 ext4_unlock_group(sb, i);
2066                                 return 0;
2067                         }
2068                 }
2069                 ext4_unlock_group(sb, i);
2070                 if (!flexbg_flag)
2071                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2072         }
2073         if (NULL != first_not_zeroed)
2074                 *first_not_zeroed = grp;
2075
2076         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
2077         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2078         return 1;
2079 }
2080
2081 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2082  * the superblock) which were deleted from all directories, but held open by
2083  * a process at the time of a crash.  We walk the list and try to delete these
2084  * inodes at recovery time (only with a read-write filesystem).
2085  *
2086  * In order to keep the orphan inode chain consistent during traversal (in
2087  * case of crash during recovery), we link each inode into the superblock
2088  * orphan list_head and handle it the same way as an inode deletion during
2089  * normal operation (which journals the operations for us).
2090  *
2091  * We only do an iget() and an iput() on each inode, which is very safe if we
2092  * accidentally point at an in-use or already deleted inode.  The worst that
2093  * can happen in this case is that we get a "bit already cleared" message from
2094  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2095  * e2fsck was run on this filesystem, and it must have already done the orphan
2096  * inode cleanup for us, so we can safely abort without any further action.
2097  */
2098 static void ext4_orphan_cleanup(struct super_block *sb,
2099                                 struct ext4_super_block *es)
2100 {
2101         unsigned int s_flags = sb->s_flags;
2102         int nr_orphans = 0, nr_truncates = 0;
2103 #ifdef CONFIG_QUOTA
2104         int i;
2105 #endif
2106         if (!es->s_last_orphan) {
2107                 jbd_debug(4, "no orphan inodes to clean up\n");
2108                 return;
2109         }
2110
2111         if (bdev_read_only(sb->s_bdev)) {
2112                 ext4_msg(sb, KERN_ERR, "write access "
2113                         "unavailable, skipping orphan cleanup");
2114                 return;
2115         }
2116
2117         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2118                 if (es->s_last_orphan)
2119                         jbd_debug(1, "Errors on filesystem, "
2120                                   "clearing orphan list.\n");
2121                 es->s_last_orphan = 0;
2122                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2123                 return;
2124         }
2125
2126         if (s_flags & MS_RDONLY) {
2127                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2128                 sb->s_flags &= ~MS_RDONLY;
2129         }
2130 #ifdef CONFIG_QUOTA
2131         /* Needed for iput() to work correctly and not trash data */
2132         sb->s_flags |= MS_ACTIVE;
2133         /* Turn on quotas so that they are updated correctly */
2134         for (i = 0; i < MAXQUOTAS; i++) {
2135                 if (EXT4_SB(sb)->s_qf_names[i]) {
2136                         int ret = ext4_quota_on_mount(sb, i);
2137                         if (ret < 0)
2138                                 ext4_msg(sb, KERN_ERR,
2139                                         "Cannot turn on journaled "
2140                                         "quota: error %d", ret);
2141                 }
2142         }
2143 #endif
2144
2145         while (es->s_last_orphan) {
2146                 struct inode *inode;
2147
2148                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2149                 if (IS_ERR(inode)) {
2150                         es->s_last_orphan = 0;
2151                         break;
2152                 }
2153
2154                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2155                 dquot_initialize(inode);
2156                 if (inode->i_nlink) {
2157                         ext4_msg(sb, KERN_DEBUG,
2158                                 "%s: truncating inode %lu to %lld bytes",
2159                                 __func__, inode->i_ino, inode->i_size);
2160                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2161                                   inode->i_ino, inode->i_size);
2162                         ext4_truncate(inode);
2163                         nr_truncates++;
2164                 } else {
2165                         ext4_msg(sb, KERN_DEBUG,
2166                                 "%s: deleting unreferenced inode %lu",
2167                                 __func__, inode->i_ino);
2168                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2169                                   inode->i_ino);
2170                         nr_orphans++;
2171                 }
2172                 iput(inode);  /* The delete magic happens here! */
2173         }
2174
2175 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2176
2177         if (nr_orphans)
2178                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2179                        PLURAL(nr_orphans));
2180         if (nr_truncates)
2181                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2182                        PLURAL(nr_truncates));
2183 #ifdef CONFIG_QUOTA
2184         /* Turn quotas off */
2185         for (i = 0; i < MAXQUOTAS; i++) {
2186                 if (sb_dqopt(sb)->files[i])
2187                         dquot_quota_off(sb, i);
2188         }
2189 #endif
2190         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2191 }
2192
2193 /*
2194  * Maximal extent format file size.
2195  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2196  * extent format containers, within a sector_t, and within i_blocks
2197  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2198  * so that won't be a limiting factor.
2199  *
2200  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2201  */
2202 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2203 {
2204         loff_t res;
2205         loff_t upper_limit = MAX_LFS_FILESIZE;
2206
2207         /* small i_blocks in vfs inode? */
2208         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2209                 /*
2210                  * CONFIG_LBDAF is not enabled implies the inode
2211                  * i_block represent total blocks in 512 bytes
2212                  * 32 == size of vfs inode i_blocks * 8
2213                  */
2214                 upper_limit = (1LL << 32) - 1;
2215
2216                 /* total blocks in file system block size */
2217                 upper_limit >>= (blkbits - 9);
2218                 upper_limit <<= blkbits;
2219         }
2220
2221         /* 32-bit extent-start container, ee_block */
2222         res = 1LL << 32;
2223         res <<= blkbits;
2224         res -= 1;
2225
2226         /* Sanity check against vm- & vfs- imposed limits */
2227         if (res > upper_limit)
2228                 res = upper_limit;
2229
2230         return res;
2231 }
2232
2233 /*
2234  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2235  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2236  * We need to be 1 filesystem block less than the 2^48 sector limit.
2237  */
2238 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2239 {
2240         loff_t res = EXT4_NDIR_BLOCKS;
2241         int meta_blocks;
2242         loff_t upper_limit;
2243         /* This is calculated to be the largest file size for a dense, block
2244          * mapped file such that the file's total number of 512-byte sectors,
2245          * including data and all indirect blocks, does not exceed (2^48 - 1).
2246          *
2247          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2248          * number of 512-byte sectors of the file.
2249          */
2250
2251         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2252                 /*
2253                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2254                  * the inode i_block field represents total file blocks in
2255                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2256                  */
2257                 upper_limit = (1LL << 32) - 1;
2258
2259                 /* total blocks in file system block size */
2260                 upper_limit >>= (bits - 9);
2261
2262         } else {
2263                 /*
2264                  * We use 48 bit ext4_inode i_blocks
2265                  * With EXT4_HUGE_FILE_FL set the i_blocks
2266                  * represent total number of blocks in
2267                  * file system block size
2268                  */
2269                 upper_limit = (1LL << 48) - 1;
2270
2271         }
2272
2273         /* indirect blocks */
2274         meta_blocks = 1;
2275         /* double indirect blocks */
2276         meta_blocks += 1 + (1LL << (bits-2));
2277         /* tripple indirect blocks */
2278         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2279
2280         upper_limit -= meta_blocks;
2281         upper_limit <<= bits;
2282
2283         res += 1LL << (bits-2);
2284         res += 1LL << (2*(bits-2));
2285         res += 1LL << (3*(bits-2));
2286         res <<= bits;
2287         if (res > upper_limit)
2288                 res = upper_limit;
2289
2290         if (res > MAX_LFS_FILESIZE)
2291                 res = MAX_LFS_FILESIZE;
2292
2293         return res;
2294 }
2295
2296 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2297                                    ext4_fsblk_t logical_sb_block, int nr)
2298 {
2299         struct ext4_sb_info *sbi = EXT4_SB(sb);
2300         ext4_group_t bg, first_meta_bg;
2301         int has_super = 0;
2302
2303         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2304
2305         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2306             nr < first_meta_bg)
2307                 return logical_sb_block + nr + 1;
2308         bg = sbi->s_desc_per_block * nr;
2309         if (ext4_bg_has_super(sb, bg))
2310                 has_super = 1;
2311
2312         return (has_super + ext4_group_first_block_no(sb, bg));
2313 }
2314
2315 /**
2316  * ext4_get_stripe_size: Get the stripe size.
2317  * @sbi: In memory super block info
2318  *
2319  * If we have specified it via mount option, then
2320  * use the mount option value. If the value specified at mount time is
2321  * greater than the blocks per group use the super block value.
2322  * If the super block value is greater than blocks per group return 0.
2323  * Allocator needs it be less than blocks per group.
2324  *
2325  */
2326 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2327 {
2328         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2329         unsigned long stripe_width =
2330                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2331
2332         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2333                 return sbi->s_stripe;
2334
2335         if (stripe_width <= sbi->s_blocks_per_group)
2336                 return stripe_width;
2337
2338         if (stride <= sbi->s_blocks_per_group)
2339                 return stride;
2340
2341         return 0;
2342 }
2343
2344 /* sysfs supprt */
2345
2346 struct ext4_attr {
2347         struct attribute attr;
2348         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2349         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2350                          const char *, size_t);
2351         int offset;
2352 };
2353
2354 static int parse_strtoul(const char *buf,
2355                 unsigned long max, unsigned long *value)
2356 {
2357         char *endp;
2358
2359         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2360         endp = skip_spaces(endp);
2361         if (*endp || *value > max)
2362                 return -EINVAL;
2363
2364         return 0;
2365 }
2366
2367 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2368                                               struct ext4_sb_info *sbi,
2369                                               char *buf)
2370 {
2371         return snprintf(buf, PAGE_SIZE, "%llu\n",
2372                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2373 }
2374
2375 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2376                                          struct ext4_sb_info *sbi, char *buf)
2377 {
2378         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2379
2380         if (!sb->s_bdev->bd_part)
2381                 return snprintf(buf, PAGE_SIZE, "0\n");
2382         return snprintf(buf, PAGE_SIZE, "%lu\n",
2383                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2384                          sbi->s_sectors_written_start) >> 1);
2385 }
2386
2387 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2388                                           struct ext4_sb_info *sbi, char *buf)
2389 {
2390         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2391
2392         if (!sb->s_bdev->bd_part)
2393                 return snprintf(buf, PAGE_SIZE, "0\n");
2394         return snprintf(buf, PAGE_SIZE, "%llu\n",
2395                         (unsigned long long)(sbi->s_kbytes_written +
2396                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2397                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2398 }
2399
2400 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2401                                           struct ext4_sb_info *sbi,
2402                                           const char *buf, size_t count)
2403 {
2404         unsigned long t;
2405
2406         if (parse_strtoul(buf, 0x40000000, &t))
2407                 return -EINVAL;
2408
2409         if (!is_power_of_2(t))
2410                 return -EINVAL;
2411
2412         sbi->s_inode_readahead_blks = t;
2413         return count;
2414 }
2415
2416 static ssize_t sbi_ui_show(struct ext4_attr *a,
2417                            struct ext4_sb_info *sbi, char *buf)
2418 {
2419         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2420
2421         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2422 }
2423
2424 static ssize_t sbi_ui_store(struct ext4_attr *a,
2425                             struct ext4_sb_info *sbi,
2426                             const char *buf, size_t count)
2427 {
2428         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2429         unsigned long t;
2430
2431         if (parse_strtoul(buf, 0xffffffff, &t))
2432                 return -EINVAL;
2433         *ui = t;
2434         return count;
2435 }
2436
2437 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2438 static struct ext4_attr ext4_attr_##_name = {                   \
2439         .attr = {.name = __stringify(_name), .mode = _mode },   \
2440         .show   = _show,                                        \
2441         .store  = _store,                                       \
2442         .offset = offsetof(struct ext4_sb_info, _elname),       \
2443 }
2444 #define EXT4_ATTR(name, mode, show, store) \
2445 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2446
2447 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2448 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2449 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2450 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2451         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2452 #define ATTR_LIST(name) &ext4_attr_##name.attr
2453
2454 EXT4_RO_ATTR(delayed_allocation_blocks);
2455 EXT4_RO_ATTR(session_write_kbytes);
2456 EXT4_RO_ATTR(lifetime_write_kbytes);
2457 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2458                  inode_readahead_blks_store, s_inode_readahead_blks);
2459 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2460 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2461 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2462 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2463 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2464 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2465 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2466 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2467
2468 static struct attribute *ext4_attrs[] = {
2469         ATTR_LIST(delayed_allocation_blocks),
2470         ATTR_LIST(session_write_kbytes),
2471         ATTR_LIST(lifetime_write_kbytes),
2472         ATTR_LIST(inode_readahead_blks),
2473         ATTR_LIST(inode_goal),
2474         ATTR_LIST(mb_stats),
2475         ATTR_LIST(mb_max_to_scan),
2476         ATTR_LIST(mb_min_to_scan),
2477         ATTR_LIST(mb_order2_req),
2478         ATTR_LIST(mb_stream_req),
2479         ATTR_LIST(mb_group_prealloc),
2480         ATTR_LIST(max_writeback_mb_bump),
2481         NULL,
2482 };
2483
2484 /* Features this copy of ext4 supports */
2485 EXT4_INFO_ATTR(lazy_itable_init);
2486 EXT4_INFO_ATTR(batched_discard);
2487
2488 static struct attribute *ext4_feat_attrs[] = {
2489         ATTR_LIST(lazy_itable_init),
2490         ATTR_LIST(batched_discard),
2491         NULL,
2492 };
2493
2494 static ssize_t ext4_attr_show(struct kobject *kobj,
2495                               struct attribute *attr, char *buf)
2496 {
2497         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2498                                                 s_kobj);
2499         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2500
2501         return a->show ? a->show(a, sbi, buf) : 0;
2502 }
2503
2504 static ssize_t ext4_attr_store(struct kobject *kobj,
2505                                struct attribute *attr,
2506                                const char *buf, size_t len)
2507 {
2508         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2509                                                 s_kobj);
2510         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2511
2512         return a->store ? a->store(a, sbi, buf, len) : 0;
2513 }
2514
2515 static void ext4_sb_release(struct kobject *kobj)
2516 {
2517         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2518                                                 s_kobj);
2519         complete(&sbi->s_kobj_unregister);
2520 }
2521
2522 static const struct sysfs_ops ext4_attr_ops = {
2523         .show   = ext4_attr_show,
2524         .store  = ext4_attr_store,
2525 };
2526
2527 static struct kobj_type ext4_ktype = {
2528         .default_attrs  = ext4_attrs,
2529         .sysfs_ops      = &ext4_attr_ops,
2530         .release        = ext4_sb_release,
2531 };
2532
2533 static void ext4_feat_release(struct kobject *kobj)
2534 {
2535         complete(&ext4_feat->f_kobj_unregister);
2536 }
2537
2538 static struct kobj_type ext4_feat_ktype = {
2539         .default_attrs  = ext4_feat_attrs,
2540         .sysfs_ops      = &ext4_attr_ops,
2541         .release        = ext4_feat_release,
2542 };
2543
2544 /*
2545  * Check whether this filesystem can be mounted based on
2546  * the features present and the RDONLY/RDWR mount requested.
2547  * Returns 1 if this filesystem can be mounted as requested,
2548  * 0 if it cannot be.
2549  */
2550 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2551 {
2552         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2553                 ext4_msg(sb, KERN_ERR,
2554                         "Couldn't mount because of "
2555                         "unsupported optional features (%x)",
2556                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2557                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2558                 return 0;
2559         }
2560
2561         if (readonly)
2562                 return 1;
2563
2564         /* Check that feature set is OK for a read-write mount */
2565         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2566                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2567                          "unsupported optional features (%x)",
2568                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2569                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2570                 return 0;
2571         }
2572         /*
2573          * Large file size enabled file system can only be mounted
2574          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2575          */
2576         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2577                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2578                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2579                                  "cannot be mounted RDWR without "
2580                                  "CONFIG_LBDAF");
2581                         return 0;
2582                 }
2583         }
2584         return 1;
2585 }
2586
2587 /*
2588  * This function is called once a day if we have errors logged
2589  * on the file system
2590  */
2591 static void print_daily_error_info(unsigned long arg)
2592 {
2593         struct super_block *sb = (struct super_block *) arg;
2594         struct ext4_sb_info *sbi;
2595         struct ext4_super_block *es;
2596
2597         sbi = EXT4_SB(sb);
2598         es = sbi->s_es;
2599
2600         if (es->s_error_count)
2601                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2602                          le32_to_cpu(es->s_error_count));
2603         if (es->s_first_error_time) {
2604                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2605                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2606                        (int) sizeof(es->s_first_error_func),
2607                        es->s_first_error_func,
2608                        le32_to_cpu(es->s_first_error_line));
2609                 if (es->s_first_error_ino)
2610                         printk(": inode %u",
2611                                le32_to_cpu(es->s_first_error_ino));
2612                 if (es->s_first_error_block)
2613                         printk(": block %llu", (unsigned long long)
2614                                le64_to_cpu(es->s_first_error_block));
2615                 printk("\n");
2616         }
2617         if (es->s_last_error_time) {
2618                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2619                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2620                        (int) sizeof(es->s_last_error_func),
2621                        es->s_last_error_func,
2622                        le32_to_cpu(es->s_last_error_line));
2623                 if (es->s_last_error_ino)
2624                         printk(": inode %u",
2625                                le32_to_cpu(es->s_last_error_ino));
2626                 if (es->s_last_error_block)
2627                         printk(": block %llu", (unsigned long long)
2628                                le64_to_cpu(es->s_last_error_block));
2629                 printk("\n");
2630         }
2631         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2632 }
2633
2634 static void ext4_lazyinode_timeout(unsigned long data)
2635 {
2636         struct task_struct *p = (struct task_struct *)data;
2637         wake_up_process(p);
2638 }
2639
2640 /* Find next suitable group and run ext4_init_inode_table */
2641 static int ext4_run_li_request(struct ext4_li_request *elr)
2642 {
2643         struct ext4_group_desc *gdp = NULL;
2644         ext4_group_t group, ngroups;
2645         struct super_block *sb;
2646         unsigned long timeout = 0;
2647         int ret = 0;
2648
2649         sb = elr->lr_super;
2650         ngroups = EXT4_SB(sb)->s_groups_count;
2651
2652         for (group = elr->lr_next_group; group < ngroups; group++) {
2653                 gdp = ext4_get_group_desc(sb, group, NULL);
2654                 if (!gdp) {
2655                         ret = 1;
2656                         break;
2657                 }
2658
2659                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2660                         break;
2661         }
2662
2663         if (group == ngroups)
2664                 ret = 1;
2665
2666         if (!ret) {
2667                 timeout = jiffies;
2668                 ret = ext4_init_inode_table(sb, group,
2669                                             elr->lr_timeout ? 0 : 1);
2670                 if (elr->lr_timeout == 0) {
2671                         timeout = jiffies - timeout;
2672                         if (elr->lr_sbi->s_li_wait_mult)
2673                                 timeout *= elr->lr_sbi->s_li_wait_mult;
2674                         else
2675                                 timeout *= 20;
2676                         elr->lr_timeout = timeout;
2677                 }
2678                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2679                 elr->lr_next_group = group + 1;
2680         }
2681
2682         return ret;
2683 }
2684
2685 /*
2686  * Remove lr_request from the list_request and free the
2687  * request tructure. Should be called with li_list_mtx held
2688  */
2689 static void ext4_remove_li_request(struct ext4_li_request *elr)
2690 {
2691         struct ext4_sb_info *sbi;
2692
2693         if (!elr)
2694                 return;
2695
2696         sbi = elr->lr_sbi;
2697
2698         list_del(&elr->lr_request);
2699         sbi->s_li_request = NULL;
2700         kfree(elr);
2701 }
2702
2703 static void ext4_unregister_li_request(struct super_block *sb)
2704 {
2705         struct ext4_li_request *elr = EXT4_SB(sb)->s_li_request;
2706
2707         if (!ext4_li_info)
2708                 return;
2709
2710         mutex_lock(&ext4_li_info->li_list_mtx);
2711         ext4_remove_li_request(elr);
2712         mutex_unlock(&ext4_li_info->li_list_mtx);
2713 }
2714
2715 /*
2716  * This is the function where ext4lazyinit thread lives. It walks
2717  * through the request list searching for next scheduled filesystem.
2718  * When such a fs is found, run the lazy initialization request
2719  * (ext4_rn_li_request) and keep track of the time spend in this
2720  * function. Based on that time we compute next schedule time of
2721  * the request. When walking through the list is complete, compute
2722  * next waking time and put itself into sleep.
2723  */
2724 static int ext4_lazyinit_thread(void *arg)
2725 {
2726         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2727         struct list_head *pos, *n;
2728         struct ext4_li_request *elr;
2729         unsigned long next_wakeup;
2730         DEFINE_WAIT(wait);
2731
2732         BUG_ON(NULL == eli);
2733
2734         eli->li_timer.data = (unsigned long)current;
2735         eli->li_timer.function = ext4_lazyinode_timeout;
2736
2737         eli->li_task = current;
2738         wake_up(&eli->li_wait_task);
2739
2740 cont_thread:
2741         while (true) {
2742                 next_wakeup = MAX_JIFFY_OFFSET;
2743
2744                 mutex_lock(&eli->li_list_mtx);
2745                 if (list_empty(&eli->li_request_list)) {
2746                         mutex_unlock(&eli->li_list_mtx);
2747                         goto exit_thread;
2748                 }
2749
2750                 list_for_each_safe(pos, n, &eli->li_request_list) {
2751                         elr = list_entry(pos, struct ext4_li_request,
2752                                          lr_request);
2753
2754                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2755                                 if (ext4_run_li_request(elr) != 0) {
2756                                         /* error, remove the lazy_init job */
2757                                         ext4_remove_li_request(elr);
2758                                         continue;
2759                                 }
2760                         }
2761
2762                         if (time_before(elr->lr_next_sched, next_wakeup))
2763                                 next_wakeup = elr->lr_next_sched;
2764                 }
2765                 mutex_unlock(&eli->li_list_mtx);
2766
2767                 if (freezing(current))
2768                         refrigerator();
2769
2770                 if ((time_after_eq(jiffies, next_wakeup)) ||
2771                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2772                         cond_resched();
2773                         continue;
2774                 }
2775
2776                 eli->li_timer.expires = next_wakeup;
2777                 add_timer(&eli->li_timer);
2778                 prepare_to_wait(&eli->li_wait_daemon, &wait,
2779                                 TASK_INTERRUPTIBLE);
2780                 if (time_before(jiffies, next_wakeup))
2781                         schedule();
2782                 finish_wait(&eli->li_wait_daemon, &wait);
2783         }
2784
2785 exit_thread:
2786         /*
2787          * It looks like the request list is empty, but we need
2788          * to check it under the li_list_mtx lock, to prevent any
2789          * additions into it, and of course we should lock ext4_li_mtx
2790          * to atomically free the list and ext4_li_info, because at
2791          * this point another ext4 filesystem could be registering
2792          * new one.
2793          */
2794         mutex_lock(&ext4_li_mtx);
2795         mutex_lock(&eli->li_list_mtx);
2796         if (!list_empty(&eli->li_request_list)) {
2797                 mutex_unlock(&eli->li_list_mtx);
2798                 mutex_unlock(&ext4_li_mtx);
2799                 goto cont_thread;
2800         }
2801         mutex_unlock(&eli->li_list_mtx);
2802         del_timer_sync(&ext4_li_info->li_timer);
2803         eli->li_task = NULL;
2804         wake_up(&eli->li_wait_task);
2805
2806         kfree(ext4_li_info);
2807         ext4_li_info = NULL;
2808         mutex_unlock(&ext4_li_mtx);
2809
2810         return 0;
2811 }
2812
2813 static void ext4_clear_request_list(void)
2814 {
2815         struct list_head *pos, *n;
2816         struct ext4_li_request *elr;
2817
2818         mutex_lock(&ext4_li_info->li_list_mtx);
2819         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2820                 elr = list_entry(pos, struct ext4_li_request,
2821                                  lr_request);
2822                 ext4_remove_li_request(elr);
2823         }
2824         mutex_unlock(&ext4_li_info->li_list_mtx);
2825 }
2826
2827 static int ext4_run_lazyinit_thread(void)
2828 {
2829         struct task_struct *t;
2830
2831         t = kthread_run(ext4_lazyinit_thread, ext4_li_info, "ext4lazyinit");
2832         if (IS_ERR(t)) {
2833                 int err = PTR_ERR(t);
2834                 ext4_clear_request_list();
2835                 del_timer_sync(&ext4_li_info->li_timer);
2836                 kfree(ext4_li_info);
2837                 ext4_li_info = NULL;
2838                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2839                                  "initialization thread\n",
2840                                  err);
2841                 return err;
2842         }
2843         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2844
2845         wait_event(ext4_li_info->li_wait_task, ext4_li_info->li_task != NULL);
2846         return 0;
2847 }
2848
2849 /*
2850  * Check whether it make sense to run itable init. thread or not.
2851  * If there is at least one uninitialized inode table, return
2852  * corresponding group number, else the loop goes through all
2853  * groups and return total number of groups.
2854  */
2855 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2856 {
2857         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2858         struct ext4_group_desc *gdp = NULL;
2859
2860         for (group = 0; group < ngroups; group++) {
2861                 gdp = ext4_get_group_desc(sb, group, NULL);
2862                 if (!gdp)
2863                         continue;
2864
2865                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2866                         break;
2867         }
2868
2869         return group;
2870 }
2871
2872 static int ext4_li_info_new(void)
2873 {
2874         struct ext4_lazy_init *eli = NULL;
2875
2876         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2877         if (!eli)
2878                 return -ENOMEM;
2879
2880         eli->li_task = NULL;
2881         INIT_LIST_HEAD(&eli->li_request_list);
2882         mutex_init(&eli->li_list_mtx);
2883
2884         init_waitqueue_head(&eli->li_wait_daemon);
2885         init_waitqueue_head(&eli->li_wait_task);
2886         init_timer(&eli->li_timer);
2887         eli->li_state |= EXT4_LAZYINIT_QUIT;
2888
2889         ext4_li_info = eli;
2890
2891         return 0;
2892 }
2893
2894 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2895                                             ext4_group_t start)
2896 {
2897         struct ext4_sb_info *sbi = EXT4_SB(sb);
2898         struct ext4_li_request *elr;
2899         unsigned long rnd;
2900
2901         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2902         if (!elr)
2903                 return NULL;
2904
2905         elr->lr_super = sb;
2906         elr->lr_sbi = sbi;
2907         elr->lr_next_group = start;
2908
2909         /*
2910          * Randomize first schedule time of the request to
2911          * spread the inode table initialization requests
2912          * better.
2913          */
2914         get_random_bytes(&rnd, sizeof(rnd));
2915         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2916                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2917
2918         return elr;
2919 }
2920
2921 static int ext4_register_li_request(struct super_block *sb,
2922                                     ext4_group_t first_not_zeroed)
2923 {
2924         struct ext4_sb_info *sbi = EXT4_SB(sb);
2925         struct ext4_li_request *elr;
2926         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2927         int ret;
2928
2929         if (sbi->s_li_request != NULL)
2930                 return 0;
2931
2932         if (first_not_zeroed == ngroups ||
2933             (sb->s_flags & MS_RDONLY) ||
2934             !test_opt(sb, INIT_INODE_TABLE)) {
2935                 sbi->s_li_request = NULL;
2936                 return 0;
2937         }
2938
2939         if (first_not_zeroed == ngroups) {
2940                 sbi->s_li_request = NULL;
2941                 return 0;
2942         }
2943
2944         elr = ext4_li_request_new(sb, first_not_zeroed);
2945         if (!elr)
2946                 return -ENOMEM;
2947
2948         mutex_lock(&ext4_li_mtx);
2949
2950         if (NULL == ext4_li_info) {
2951                 ret = ext4_li_info_new();
2952                 if (ret)
2953                         goto out;
2954         }
2955
2956         mutex_lock(&ext4_li_info->li_list_mtx);
2957         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2958         mutex_unlock(&ext4_li_info->li_list_mtx);
2959
2960         sbi->s_li_request = elr;
2961
2962         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2963                 ret = ext4_run_lazyinit_thread();
2964                 if (ret)
2965                         goto out;
2966         }
2967 out:
2968         mutex_unlock(&ext4_li_mtx);
2969         if (ret)
2970                 kfree(elr);
2971         return ret;
2972 }
2973
2974 /*
2975  * We do not need to lock anything since this is called on
2976  * module unload.
2977  */
2978 static void ext4_destroy_lazyinit_thread(void)
2979 {
2980         /*
2981          * If thread exited earlier
2982          * there's nothing to be done.
2983          */
2984         if (!ext4_li_info)
2985                 return;
2986
2987         ext4_clear_request_list();
2988
2989         while (ext4_li_info->li_task) {
2990                 wake_up(&ext4_li_info->li_wait_daemon);
2991                 wait_event(ext4_li_info->li_wait_task,
2992                            ext4_li_info->li_task == NULL);
2993         }
2994 }
2995
2996 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2997                                 __releases(kernel_lock)
2998                                 __acquires(kernel_lock)
2999 {
3000         char *orig_data = kstrdup(data, GFP_KERNEL);
3001         struct buffer_head *bh;
3002         struct ext4_super_block *es = NULL;
3003         struct ext4_sb_info *sbi;
3004         ext4_fsblk_t block;
3005         ext4_fsblk_t sb_block = get_sb_block(&data);
3006         ext4_fsblk_t logical_sb_block;
3007         unsigned long offset = 0;
3008         unsigned long journal_devnum = 0;
3009         unsigned long def_mount_opts;
3010         struct inode *root;
3011         char *cp;
3012         const char *descr;
3013         int ret = -ENOMEM;
3014         int blocksize;
3015         unsigned int db_count;
3016         unsigned int i;
3017         int needs_recovery, has_huge_files;
3018         __u64 blocks_count;
3019         int err;
3020         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3021         ext4_group_t first_not_zeroed;
3022
3023         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3024         if (!sbi)
3025                 goto out_free_orig;
3026
3027         sbi->s_blockgroup_lock =
3028                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3029         if (!sbi->s_blockgroup_lock) {
3030                 kfree(sbi);
3031                 goto out_free_orig;
3032         }
3033         sb->s_fs_info = sbi;
3034         sbi->s_mount_opt = 0;
3035         sbi->s_resuid = EXT4_DEF_RESUID;
3036         sbi->s_resgid = EXT4_DEF_RESGID;
3037         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3038         sbi->s_sb_block = sb_block;
3039         if (sb->s_bdev->bd_part)
3040                 sbi->s_sectors_written_start =
3041                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3042
3043         /* Cleanup superblock name */
3044         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3045                 *cp = '!';
3046
3047         ret = -EINVAL;
3048         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3049         if (!blocksize) {
3050                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3051                 goto out_fail;
3052         }
3053
3054         /*
3055          * The ext4 superblock will not be buffer aligned for other than 1kB
3056          * block sizes.  We need to calculate the offset from buffer start.
3057          */
3058         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3059                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3060                 offset = do_div(logical_sb_block, blocksize);
3061         } else {
3062                 logical_sb_block = sb_block;
3063         }
3064
3065         if (!(bh = sb_bread(sb, logical_sb_block))) {
3066                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3067                 goto out_fail;
3068         }
3069         /*
3070          * Note: s_es must be initialized as soon as possible because
3071          *       some ext4 macro-instructions depend on its value
3072          */
3073         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3074         sbi->s_es = es;
3075         sb->s_magic = le16_to_cpu(es->s_magic);
3076         if (sb->s_magic != EXT4_SUPER_MAGIC)
3077                 goto cantfind_ext4;
3078         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3079
3080         /* Set defaults before we parse the mount options */
3081         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3082         set_opt(sb, INIT_INODE_TABLE);
3083         if (def_mount_opts & EXT4_DEFM_DEBUG)
3084                 set_opt(sb, DEBUG);
3085         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3086                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3087                         "2.6.38");
3088                 set_opt(sb, GRPID);
3089         }
3090         if (def_mount_opts & EXT4_DEFM_UID16)
3091                 set_opt(sb, NO_UID32);
3092 #ifdef CONFIG_EXT4_FS_XATTR
3093         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
3094                 set_opt(sb, XATTR_USER);
3095 #endif
3096 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3097         if (def_mount_opts & EXT4_DEFM_ACL)
3098                 set_opt(sb, POSIX_ACL);
3099 #endif
3100         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3101                 set_opt(sb, JOURNAL_DATA);
3102         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3103                 set_opt(sb, ORDERED_DATA);
3104         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3105                 set_opt(sb, WRITEBACK_DATA);
3106
3107         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3108                 set_opt(sb, ERRORS_PANIC);
3109         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3110                 set_opt(sb, ERRORS_CONT);
3111         else
3112                 set_opt(sb, ERRORS_RO);
3113         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3114                 set_opt(sb, BLOCK_VALIDITY);
3115         if (def_mount_opts & EXT4_DEFM_DISCARD)
3116                 set_opt(sb, DISCARD);
3117
3118         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3119         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3120         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3121         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3122         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3123
3124         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3125                 set_opt(sb, BARRIER);
3126
3127         /*
3128          * enable delayed allocation by default
3129          * Use -o nodelalloc to turn it off
3130          */
3131         if (!IS_EXT3_SB(sb) &&
3132             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3133                 set_opt(sb, DELALLOC);
3134
3135         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3136                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3137                 ext4_msg(sb, KERN_WARNING,
3138                          "failed to parse options in superblock: %s",
3139                          sbi->s_es->s_mount_opts);
3140         }
3141         if (!parse_options((char *) data, sb, &journal_devnum,
3142                            &journal_ioprio, NULL, 0))
3143                 goto failed_mount;
3144
3145         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3146                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3147
3148         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3149             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3150              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3151              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3152                 ext4_msg(sb, KERN_WARNING,
3153                        "feature flags set on rev 0 fs, "
3154                        "running e2fsck is recommended");
3155
3156         /*
3157          * Check feature flags regardless of the revision level, since we
3158          * previously didn't change the revision level when setting the flags,
3159          * so there is a chance incompat flags are set on a rev 0 filesystem.
3160          */
3161         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3162                 goto failed_mount;
3163
3164         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3165
3166         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3167             blocksize > EXT4_MAX_BLOCK_SIZE) {
3168                 ext4_msg(sb, KERN_ERR,
3169                        "Unsupported filesystem blocksize %d", blocksize);
3170                 goto failed_mount;
3171         }
3172
3173         if (sb->s_blocksize != blocksize) {
3174                 /* Validate the filesystem blocksize */
3175                 if (!sb_set_blocksize(sb, blocksize)) {
3176                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3177                                         blocksize);
3178                         goto failed_mount;
3179                 }
3180
3181                 brelse(bh);
3182                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3183                 offset = do_div(logical_sb_block, blocksize);
3184                 bh = sb_bread(sb, logical_sb_block);
3185                 if (!bh) {
3186                         ext4_msg(sb, KERN_ERR,
3187                                "Can't read superblock on 2nd try");
3188                         goto failed_mount;
3189                 }
3190                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3191                 sbi->s_es = es;
3192                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3193                         ext4_msg(sb, KERN_ERR,
3194                                "Magic mismatch, very weird!");
3195                         goto failed_mount;
3196                 }
3197         }
3198
3199         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3200                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3201         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3202                                                       has_huge_files);
3203         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3204
3205         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3206                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3207                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3208         } else {
3209                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3210                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3211                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3212                     (!is_power_of_2(sbi->s_inode_size)) ||
3213                     (sbi->s_inode_size > blocksize)) {
3214                         ext4_msg(sb, KERN_ERR,
3215                                "unsupported inode size: %d",
3216                                sbi->s_inode_size);
3217                         goto failed_mount;
3218                 }
3219                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3220                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3221         }
3222
3223         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3224         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3225                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3226                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3227                     !is_power_of_2(sbi->s_desc_size)) {
3228                         ext4_msg(sb, KERN_ERR,
3229                                "unsupported descriptor size %lu",
3230                                sbi->s_desc_size);
3231                         goto failed_mount;
3232                 }
3233         } else
3234                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3235
3236         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3237         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3238         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3239                 goto cantfind_ext4;
3240
3241         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3242         if (sbi->s_inodes_per_block == 0)
3243                 goto cantfind_ext4;
3244         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3245                                         sbi->s_inodes_per_block;
3246         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3247         sbi->s_sbh = bh;
3248         sbi->s_mount_state = le16_to_cpu(es->s_state);
3249         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3250         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3251
3252         for (i = 0; i < 4; i++)
3253                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3254         sbi->s_def_hash_version = es->s_def_hash_version;
3255         i = le32_to_cpu(es->s_flags);
3256         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3257                 sbi->s_hash_unsigned = 3;
3258         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3259 #ifdef __CHAR_UNSIGNED__
3260                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3261                 sbi->s_hash_unsigned = 3;
3262 #else
3263                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3264 #endif
3265                 sb->s_dirt = 1;
3266         }
3267
3268         if (sbi->s_blocks_per_group > blocksize * 8) {
3269                 ext4_msg(sb, KERN_ERR,
3270                        "#blocks per group too big: %lu",
3271                        sbi->s_blocks_per_group);
3272                 goto failed_mount;
3273         }
3274         if (sbi->s_inodes_per_group > blocksize * 8) {
3275                 ext4_msg(sb, KERN_ERR,
3276                        "#inodes per group too big: %lu",
3277                        sbi->s_inodes_per_group);
3278                 goto failed_mount;
3279         }
3280
3281         /*
3282          * Test whether we have more sectors than will fit in sector_t,
3283          * and whether the max offset is addressable by the page cache.
3284          */
3285         err = generic_check_addressable(sb->s_blocksize_bits,
3286                                         ext4_blocks_count(es));
3287         if (err) {
3288                 ext4_msg(sb, KERN_ERR, "filesystem"
3289                          " too large to mount safely on this system");
3290                 if (sizeof(sector_t) < 8)
3291                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3292                 ret = err;
3293                 goto failed_mount;
3294         }
3295
3296         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3297                 goto cantfind_ext4;
3298
3299         /* check blocks count against device size */
3300         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3301         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3302                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3303                        "exceeds size of device (%llu blocks)",
3304                        ext4_blocks_count(es), blocks_count);
3305                 goto failed_mount;
3306         }
3307
3308         /*
3309          * It makes no sense for the first data block to be beyond the end
3310          * of the filesystem.
3311          */
3312         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3313                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3314                          "block %u is beyond end of filesystem (%llu)",
3315                          le32_to_cpu(es->s_first_data_block),
3316                          ext4_blocks_count(es));
3317                 goto failed_mount;
3318         }
3319         blocks_count = (ext4_blocks_count(es) -
3320                         le32_to_cpu(es->s_first_data_block) +
3321                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3322         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3323         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3324                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3325                        "(block count %llu, first data block %u, "
3326                        "blocks per group %lu)", sbi->s_groups_count,
3327                        ext4_blocks_count(es),
3328                        le32_to_cpu(es->s_first_data_block),
3329                        EXT4_BLOCKS_PER_GROUP(sb));
3330                 goto failed_mount;
3331         }
3332         sbi->s_groups_count = blocks_count;
3333         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3334                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3335         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3336                    EXT4_DESC_PER_BLOCK(sb);
3337         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
3338                                     GFP_KERNEL);
3339         if (sbi->s_group_desc == NULL) {
3340                 ext4_msg(sb, KERN_ERR, "not enough memory");
3341                 goto failed_mount;
3342         }
3343
3344 #ifdef CONFIG_PROC_FS
3345         if (ext4_proc_root)
3346                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3347 #endif
3348
3349         bgl_lock_init(sbi->s_blockgroup_lock);
3350
3351         for (i = 0; i < db_count; i++) {
3352                 block = descriptor_loc(sb, logical_sb_block, i);
3353                 sbi->s_group_desc[i] = sb_bread(sb, block);
3354                 if (!sbi->s_group_desc[i]) {
3355                         ext4_msg(sb, KERN_ERR,
3356                                "can't read group descriptor %d", i);
3357                         db_count = i;
3358                         goto failed_mount2;
3359                 }
3360         }
3361         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3362                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3363                 goto failed_mount2;
3364         }
3365         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3366                 if (!ext4_fill_flex_info(sb)) {
3367                         ext4_msg(sb, KERN_ERR,
3368                                "unable to initialize "
3369                                "flex_bg meta info!");
3370                         goto failed_mount2;
3371                 }
3372
3373         sbi->s_gdb_count = db_count;
3374         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3375         spin_lock_init(&sbi->s_next_gen_lock);
3376
3377         err = percpu_counter_init(&sbi->s_freeblocks_counter,
3378                         ext4_count_free_blocks(sb));
3379         if (!err) {
3380                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3381                                 ext4_count_free_inodes(sb));
3382         }
3383         if (!err) {
3384                 err = percpu_counter_init(&sbi->s_dirs_counter,
3385                                 ext4_count_dirs(sb));
3386         }
3387         if (!err) {
3388                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
3389         }
3390         if (err) {
3391                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3392                 goto failed_mount3;
3393         }
3394
3395         sbi->s_stripe = ext4_get_stripe_size(sbi);
3396         sbi->s_max_writeback_mb_bump = 128;
3397
3398         /*
3399          * set up enough so that it can read an inode
3400          */
3401         if (!test_opt(sb, NOLOAD) &&
3402             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3403                 sb->s_op = &ext4_sops;
3404         else
3405                 sb->s_op = &ext4_nojournal_sops;
3406         sb->s_export_op = &ext4_export_ops;
3407         sb->s_xattr = ext4_xattr_handlers;
3408 #ifdef CONFIG_QUOTA
3409         sb->s_qcop = &ext4_qctl_operations;
3410         sb->dq_op = &ext4_quota_operations;
3411 #endif
3412         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3413         mutex_init(&sbi->s_orphan_lock);
3414         mutex_init(&sbi->s_resize_lock);
3415
3416         sb->s_root = NULL;
3417
3418         needs_recovery = (es->s_last_orphan != 0 ||
3419                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3420                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3421
3422         /*
3423          * The first inode we look at is the journal inode.  Don't try
3424          * root first: it may be modified in the journal!
3425          */
3426         if (!test_opt(sb, NOLOAD) &&
3427             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3428                 if (ext4_load_journal(sb, es, journal_devnum))
3429                         goto failed_mount3;
3430         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3431               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3432                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3433                        "suppressed and not mounted read-only");
3434                 goto failed_mount_wq;
3435         } else {
3436                 clear_opt(sb, DATA_FLAGS);
3437                 set_opt(sb, WRITEBACK_DATA);
3438                 sbi->s_journal = NULL;
3439                 needs_recovery = 0;
3440                 goto no_journal;
3441         }
3442
3443         if (ext4_blocks_count(es) > 0xffffffffULL &&
3444             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3445                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3446                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3447                 goto failed_mount_wq;
3448         }
3449
3450         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3451                 jbd2_journal_set_features(sbi->s_journal,
3452                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3453                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3454         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3455                 jbd2_journal_set_features(sbi->s_journal,
3456                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3457                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3458                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3459         } else {
3460                 jbd2_journal_clear_features(sbi->s_journal,
3461                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3462                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3463         }
3464
3465         /* We have now updated the journal if required, so we can
3466          * validate the data journaling mode. */
3467         switch (test_opt(sb, DATA_FLAGS)) {
3468         case 0:
3469                 /* No mode set, assume a default based on the journal
3470                  * capabilities: ORDERED_DATA if the journal can
3471                  * cope, else JOURNAL_DATA
3472                  */
3473                 if (jbd2_journal_check_available_features
3474                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3475                         set_opt(sb, ORDERED_DATA);
3476                 else
3477                         set_opt(sb, JOURNAL_DATA);
3478                 break;
3479
3480         case EXT4_MOUNT_ORDERED_DATA:
3481         case EXT4_MOUNT_WRITEBACK_DATA:
3482                 if (!jbd2_journal_check_available_features
3483                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3484                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3485                                "requested data journaling mode");
3486                         goto failed_mount_wq;
3487                 }
3488         default:
3489                 break;
3490         }
3491         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3492
3493         /*
3494          * The journal may have updated the bg summary counts, so we
3495          * need to update the global counters.
3496          */
3497         percpu_counter_set(&sbi->s_freeblocks_counter,
3498                            ext4_count_free_blocks(sb));
3499         percpu_counter_set(&sbi->s_freeinodes_counter,
3500                            ext4_count_free_inodes(sb));
3501         percpu_counter_set(&sbi->s_dirs_counter,
3502                            ext4_count_dirs(sb));
3503         percpu_counter_set(&sbi->s_dirtyblocks_counter, 0);
3504
3505 no_journal:
3506         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
3507         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3508                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3509                 goto failed_mount_wq;
3510         }
3511
3512         /*
3513          * The jbd2_journal_load will have done any necessary log recovery,
3514          * so we can safely mount the rest of the filesystem now.
3515          */
3516
3517         root = ext4_iget(sb, EXT4_ROOT_INO);
3518         if (IS_ERR(root)) {
3519                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3520                 ret = PTR_ERR(root);
3521                 goto failed_mount4;
3522         }
3523         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3524                 iput(root);
3525                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3526                 goto failed_mount4;
3527         }
3528         sb->s_root = d_alloc_root(root);
3529         if (!sb->s_root) {
3530                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3531                 iput(root);
3532                 ret = -ENOMEM;
3533                 goto failed_mount4;
3534         }
3535
3536         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3537
3538         /* determine the minimum size of new large inodes, if present */
3539         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3540                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3541                                                      EXT4_GOOD_OLD_INODE_SIZE;
3542                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3543                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3544                         if (sbi->s_want_extra_isize <
3545                             le16_to_cpu(es->s_want_extra_isize))
3546                                 sbi->s_want_extra_isize =
3547                                         le16_to_cpu(es->s_want_extra_isize);
3548                         if (sbi->s_want_extra_isize <
3549                             le16_to_cpu(es->s_min_extra_isize))
3550                                 sbi->s_want_extra_isize =
3551                                         le16_to_cpu(es->s_min_extra_isize);
3552                 }
3553         }
3554         /* Check if enough inode space is available */
3555         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3556                                                         sbi->s_inode_size) {
3557                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3558                                                        EXT4_GOOD_OLD_INODE_SIZE;
3559                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3560                          "available");
3561         }
3562
3563         if (test_opt(sb, DELALLOC) &&
3564             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
3565                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
3566                          "requested data journaling mode");
3567                 clear_opt(sb, DELALLOC);
3568         }
3569         if (test_opt(sb, DIOREAD_NOLOCK)) {
3570                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3571                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3572                                 "option - requested data journaling mode");
3573                         clear_opt(sb, DIOREAD_NOLOCK);
3574                 }
3575                 if (sb->s_blocksize < PAGE_SIZE) {
3576                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
3577                                 "option - block size is too small");
3578                         clear_opt(sb, DIOREAD_NOLOCK);
3579                 }
3580         }
3581
3582         err = ext4_setup_system_zone(sb);
3583         if (err) {
3584                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3585                          "zone (%d)", err);
3586                 goto failed_mount4;
3587         }
3588
3589         ext4_ext_init(sb);
3590         err = ext4_mb_init(sb, needs_recovery);
3591         if (err) {
3592                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3593                          err);
3594                 goto failed_mount4;
3595         }
3596
3597         err = ext4_register_li_request(sb, first_not_zeroed);
3598         if (err)
3599                 goto failed_mount4;
3600
3601         sbi->s_kobj.kset = ext4_kset;
3602         init_completion(&sbi->s_kobj_unregister);
3603         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3604                                    "%s", sb->s_id);
3605         if (err) {
3606                 ext4_mb_release(sb);
3607                 ext4_ext_release(sb);
3608                 goto failed_mount4;
3609         };
3610
3611         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3612         ext4_orphan_cleanup(sb, es);
3613         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3614         if (needs_recovery) {
3615                 ext4_msg(sb, KERN_INFO, "recovery complete");
3616                 ext4_mark_recovery_complete(sb, es);
3617         }
3618         if (EXT4_SB(sb)->s_journal) {
3619                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3620                         descr = " journalled data mode";
3621                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3622                         descr = " ordered data mode";
3623                 else
3624                         descr = " writeback data mode";
3625         } else
3626                 descr = "out journal";
3627
3628         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3629                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3630                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3631
3632         init_timer(&sbi->s_err_report);
3633         sbi->s_err_report.function = print_daily_error_info;
3634         sbi->s_err_report.data = (unsigned long) sb;
3635         if (es->s_error_count)
3636                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3637
3638         kfree(orig_data);
3639         return 0;
3640
3641 cantfind_ext4:
3642         if (!silent)
3643                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3644         goto failed_mount;
3645
3646 failed_mount4:
3647         ext4_msg(sb, KERN_ERR, "mount failed");
3648         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3649 failed_mount_wq:
3650         ext4_release_system_zone(sb);
3651         if (sbi->s_journal) {
3652                 jbd2_journal_destroy(sbi->s_journal);
3653                 sbi->s_journal = NULL;
3654         }
3655 failed_mount3:
3656         if (sbi->s_flex_groups) {
3657                 if (is_vmalloc_addr(sbi->s_flex_groups))
3658                         vfree(sbi->s_flex_groups);
3659                 else
3660                         kfree(sbi->s_flex_groups);
3661         }
3662         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3663         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3664         percpu_counter_destroy(&sbi->s_dirs_counter);
3665         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3666 failed_mount2:
3667         for (i = 0; i < db_count; i++)
3668                 brelse(sbi->s_group_desc[i]);
3669         kfree(sbi->s_group_desc);
3670 failed_mount:
3671         if (sbi->s_proc) {
3672                 remove_proc_entry(sb->s_id, ext4_proc_root);
3673         }
3674 #ifdef CONFIG_QUOTA
3675         for (i = 0; i < MAXQUOTAS; i++)
3676                 kfree(sbi->s_qf_names[i]);
3677 #endif
3678         ext4_blkdev_remove(sbi);
3679         brelse(bh);
3680 out_fail:
3681         sb->s_fs_info = NULL;
3682         kfree(sbi->s_blockgroup_lock);
3683         kfree(sbi);
3684 out_free_orig:
3685         kfree(orig_data);
3686         return ret;
3687 }
3688
3689 /*
3690  * Setup any per-fs journal parameters now.  We'll do this both on
3691  * initial mount, once the journal has been initialised but before we've
3692  * done any recovery; and again on any subsequent remount.
3693  */
3694 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3695 {
3696         struct ext4_sb_info *sbi = EXT4_SB(sb);
3697
3698         journal->j_commit_interval = sbi->s_commit_interval;
3699         journal->j_min_batch_time = sbi->s_min_batch_time;
3700         journal->j_max_batch_time = sbi->s_max_batch_time;
3701
3702         write_lock(&journal->j_state_lock);
3703         if (test_opt(sb, BARRIER))
3704                 journal->j_flags |= JBD2_BARRIER;
3705         else
3706                 journal->j_flags &= ~JBD2_BARRIER;
3707         if (test_opt(sb, DATA_ERR_ABORT))
3708                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3709         else
3710                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3711         write_unlock(&journal->j_state_lock);
3712 }
3713
3714 static journal_t *ext4_get_journal(struct super_block *sb,
3715                                    unsigned int journal_inum)
3716 {
3717         struct inode *journal_inode;
3718         journal_t *journal;
3719
3720         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3721
3722         /* First, test for the existence of a valid inode on disk.  Bad
3723          * things happen if we iget() an unused inode, as the subsequent
3724          * iput() will try to delete it. */
3725
3726         journal_inode = ext4_iget(sb, journal_inum);
3727         if (IS_ERR(journal_inode)) {
3728                 ext4_msg(sb, KERN_ERR, "no journal found");
3729                 return NULL;
3730         }
3731         if (!journal_inode->i_nlink) {
3732                 make_bad_inode(journal_inode);
3733                 iput(journal_inode);
3734                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3735                 return NULL;
3736         }
3737
3738         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3739                   journal_inode, journal_inode->i_size);
3740         if (!S_ISREG(journal_inode->i_mode)) {
3741                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3742                 iput(journal_inode);
3743                 return NULL;
3744         }
3745
3746         journal = jbd2_journal_init_inode(journal_inode);
3747         if (!journal) {
3748                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3749                 iput(journal_inode);
3750                 return NULL;
3751         }
3752         journal->j_private = sb;
3753         ext4_init_journal_params(sb, journal);
3754         return journal;
3755 }
3756
3757 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3758                                        dev_t j_dev)
3759 {
3760         struct buffer_head *bh;
3761         journal_t *journal;
3762         ext4_fsblk_t start;
3763         ext4_fsblk_t len;
3764         int hblock, blocksize;
3765         ext4_fsblk_t sb_block;
3766         unsigned long offset;
3767         struct ext4_super_block *es;
3768         struct block_device *bdev;
3769
3770         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3771
3772         bdev = ext4_blkdev_get(j_dev, sb);
3773         if (bdev == NULL)
3774                 return NULL;
3775
3776         if (bd_claim(bdev, sb)) {
3777                 ext4_msg(sb, KERN_ERR,
3778                         "failed to claim external journal device");
3779                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3780                 return NULL;
3781         }
3782
3783         blocksize = sb->s_blocksize;
3784         hblock = bdev_logical_block_size(bdev);
3785         if (blocksize < hblock) {
3786                 ext4_msg(sb, KERN_ERR,
3787                         "blocksize too small for journal device");
3788                 goto out_bdev;
3789         }
3790
3791         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3792         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3793         set_blocksize(bdev, blocksize);
3794         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3795                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3796                        "external journal");
3797                 goto out_bdev;
3798         }
3799
3800         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3801         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3802             !(le32_to_cpu(es->s_feature_incompat) &
3803               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3804                 ext4_msg(sb, KERN_ERR, "external journal has "
3805                                         "bad superblock");
3806                 brelse(bh);
3807                 goto out_bdev;
3808         }
3809
3810         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3811                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3812                 brelse(bh);
3813                 goto out_bdev;
3814         }
3815
3816         len = ext4_blocks_count(es);
3817         start = sb_block + 1;
3818         brelse(bh);     /* we're done with the superblock */
3819
3820         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3821                                         start, len, blocksize);
3822         if (!journal) {
3823                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3824                 goto out_bdev;
3825         }
3826         journal->j_private = sb;
3827         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3828         wait_on_buffer(journal->j_sb_buffer);
3829         if (!buffer_uptodate(journal->j_sb_buffer)) {
3830                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3831                 goto out_journal;
3832         }
3833         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3834                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3835                                         "user (unsupported) - %d",
3836                         be32_to_cpu(journal->j_superblock->s_nr_users));
3837                 goto out_journal;
3838         }
3839         EXT4_SB(sb)->journal_bdev = bdev;
3840         ext4_init_journal_params(sb, journal);
3841         return journal;
3842
3843 out_journal:
3844         jbd2_journal_destroy(journal);
3845 out_bdev:
3846         ext4_blkdev_put(bdev);
3847         return NULL;
3848 }
3849
3850 static int ext4_load_journal(struct super_block *sb,
3851                              struct ext4_super_block *es,
3852                              unsigned long journal_devnum)
3853 {
3854         journal_t *journal;
3855         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3856         dev_t journal_dev;
3857         int err = 0;
3858         int really_read_only;
3859
3860         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3861
3862         if (journal_devnum &&
3863             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3864                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3865                         "numbers have changed");
3866                 journal_dev = new_decode_dev(journal_devnum);
3867         } else
3868                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3869
3870         really_read_only = bdev_read_only(sb->s_bdev);
3871
3872         /*
3873          * Are we loading a blank journal or performing recovery after a
3874          * crash?  For recovery, we need to check in advance whether we
3875          * can get read-write access to the device.
3876          */
3877         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3878                 if (sb->s_flags & MS_RDONLY) {
3879                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3880                                         "required on readonly filesystem");
3881                         if (really_read_only) {
3882                                 ext4_msg(sb, KERN_ERR, "write access "
3883                                         "unavailable, cannot proceed");
3884                                 return -EROFS;
3885                         }
3886                         ext4_msg(sb, KERN_INFO, "write access will "
3887                                "be enabled during recovery");
3888                 }
3889         }
3890
3891         if (journal_inum && journal_dev) {
3892                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3893                        "and inode journals!");
3894                 return -EINVAL;
3895         }
3896
3897         if (journal_inum) {
3898                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3899                         return -EINVAL;
3900         } else {
3901                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3902                         return -EINVAL;
3903         }
3904
3905         if (!(journal->j_flags & JBD2_BARRIER))
3906                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3907
3908         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3909                 err = jbd2_journal_update_format(journal);
3910                 if (err)  {
3911                         ext4_msg(sb, KERN_ERR, "error updating journal");
3912                         jbd2_journal_destroy(journal);
3913                         return err;
3914                 }
3915         }
3916
3917         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3918                 err = jbd2_journal_wipe(journal, !really_read_only);
3919         if (!err) {
3920                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3921                 if (save)
3922                         memcpy(save, ((char *) es) +
3923                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3924                 err = jbd2_journal_load(journal);
3925                 if (save)
3926                         memcpy(((char *) es) + EXT4_S_ERR_START,
3927                                save, EXT4_S_ERR_LEN);
3928                 kfree(save);
3929         }
3930
3931         if (err) {
3932                 ext4_msg(sb, KERN_ERR, "error loading journal");
3933                 jbd2_journal_destroy(journal);
3934                 return err;
3935         }
3936
3937         EXT4_SB(sb)->s_journal = journal;
3938         ext4_clear_journal_err(sb, es);
3939
3940         if (!really_read_only && journal_devnum &&
3941             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3942                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3943
3944                 /* Make sure we flush the recovery flag to disk. */
3945                 ext4_commit_super(sb, 1);
3946         }
3947
3948         return 0;
3949 }
3950
3951 static int ext4_commit_super(struct super_block *sb, int sync)
3952 {
3953         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3954         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3955         int error = 0;
3956
3957         if (!sbh)
3958                 return error;
3959         if (buffer_write_io_error(sbh)) {
3960                 /*
3961                  * Oh, dear.  A previous attempt to write the
3962                  * superblock failed.  This could happen because the
3963                  * USB device was yanked out.  Or it could happen to
3964                  * be a transient write error and maybe the block will
3965                  * be remapped.  Nothing we can do but to retry the
3966                  * write and hope for the best.
3967                  */
3968                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3969                        "superblock detected");
3970                 clear_buffer_write_io_error(sbh);
3971                 set_buffer_uptodate(sbh);
3972         }
3973         /*
3974          * If the file system is mounted read-only, don't update the
3975          * superblock write time.  This avoids updating the superblock
3976          * write time when we are mounting the root file system
3977          * read/only but we need to replay the journal; at that point,
3978          * for people who are east of GMT and who make their clock
3979          * tick in localtime for Windows bug-for-bug compatibility,
3980          * the clock is set in the future, and this will cause e2fsck
3981          * to complain and force a full file system check.
3982          */
3983         if (!(sb->s_flags & MS_RDONLY))
3984                 es->s_wtime = cpu_to_le32(get_seconds());
3985         if (sb->s_bdev->bd_part)
3986                 es->s_kbytes_written =
3987                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3988                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3989                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3990         else
3991                 es->s_kbytes_written =
3992                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
3993         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3994                                            &EXT4_SB(sb)->s_freeblocks_counter));
3995         es->s_free_inodes_count =
3996                 cpu_to_le32(percpu_counter_sum_positive(
3997                                 &EXT4_SB(sb)->s_freeinodes_counter));
3998         sb->s_dirt = 0;
3999         BUFFER_TRACE(sbh, "marking dirty");
4000         mark_buffer_dirty(sbh);
4001         if (sync) {
4002                 error = sync_dirty_buffer(sbh);
4003                 if (error)
4004                         return error;
4005
4006                 error = buffer_write_io_error(sbh);
4007                 if (error) {
4008                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4009                                "superblock");
4010                         clear_buffer_write_io_error(sbh);
4011                         set_buffer_uptodate(sbh);
4012                 }
4013         }
4014         return error;
4015 }
4016
4017 /*
4018  * Have we just finished recovery?  If so, and if we are mounting (or
4019  * remounting) the filesystem readonly, then we will end up with a
4020  * consistent fs on disk.  Record that fact.
4021  */
4022 static void ext4_mark_recovery_complete(struct super_block *sb,
4023                                         struct ext4_super_block *es)
4024 {
4025         journal_t *journal = EXT4_SB(sb)->s_journal;
4026
4027         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4028                 BUG_ON(journal != NULL);
4029                 return;
4030         }
4031         jbd2_journal_lock_updates(journal);
4032         if (jbd2_journal_flush(journal) < 0)
4033                 goto out;
4034
4035         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4036             sb->s_flags & MS_RDONLY) {
4037                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4038                 ext4_commit_super(sb, 1);
4039         }
4040
4041 out:
4042         jbd2_journal_unlock_updates(journal);
4043 }
4044
4045 /*
4046  * If we are mounting (or read-write remounting) a filesystem whose journal
4047  * has recorded an error from a previous lifetime, move that error to the
4048  * main filesystem now.
4049  */
4050 static void ext4_clear_journal_err(struct super_block *sb,
4051                                    struct ext4_super_block *es)
4052 {
4053         journal_t *journal;
4054         int j_errno;
4055         const char *errstr;
4056
4057         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4058
4059         journal = EXT4_SB(sb)->s_journal;
4060
4061         /*
4062          * Now check for any error status which may have been recorded in the
4063          * journal by a prior ext4_error() or ext4_abort()
4064          */
4065
4066         j_errno = jbd2_journal_errno(journal);
4067         if (j_errno) {
4068                 char nbuf[16];
4069
4070                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4071                 ext4_warning(sb, "Filesystem error recorded "
4072                              "from previous mount: %s", errstr);
4073                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4074
4075                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4076                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4077                 ext4_commit_super(sb, 1);
4078
4079                 jbd2_journal_clear_err(journal);
4080         }
4081 }
4082
4083 /*
4084  * Force the running and committing transactions to commit,
4085  * and wait on the commit.
4086  */
4087 int ext4_force_commit(struct super_block *sb)
4088 {
4089         journal_t *journal;
4090         int ret = 0;
4091
4092         if (sb->s_flags & MS_RDONLY)
4093                 return 0;
4094
4095         journal = EXT4_SB(sb)->s_journal;
4096         if (journal) {
4097                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4098                 ret = ext4_journal_force_commit(journal);
4099         }
4100
4101         return ret;
4102 }
4103
4104 static void ext4_write_super(struct super_block *sb)
4105 {
4106         lock_super(sb);
4107         ext4_commit_super(sb, 1);
4108         unlock_super(sb);
4109 }
4110
4111 static int ext4_sync_fs(struct super_block *sb, int wait)
4112 {
4113         int ret = 0;
4114         tid_t target;
4115         struct ext4_sb_info *sbi = EXT4_SB(sb);
4116
4117         trace_ext4_sync_fs(sb, wait);
4118         flush_workqueue(sbi->dio_unwritten_wq);
4119         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4120                 if (wait)
4121                         jbd2_log_wait_commit(sbi->s_journal, target);
4122         }
4123         return ret;
4124 }
4125
4126 /*
4127  * LVM calls this function before a (read-only) snapshot is created.  This
4128  * gives us a chance to flush the journal completely and mark the fs clean.
4129  */
4130 static int ext4_freeze(struct super_block *sb)
4131 {
4132         int error = 0;
4133         journal_t *journal;
4134
4135         if (sb->s_flags & MS_RDONLY)
4136                 return 0;
4137
4138         journal = EXT4_SB(sb)->s_journal;
4139
4140         /* Now we set up the journal barrier. */
4141         jbd2_journal_lock_updates(journal);
4142
4143         /*
4144          * Don't clear the needs_recovery flag if we failed to flush
4145          * the journal.
4146          */
4147         error = jbd2_journal_flush(journal);
4148         if (error < 0)
4149                 goto out;
4150
4151         /* Journal blocked and flushed, clear needs_recovery flag. */
4152         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4153         error = ext4_commit_super(sb, 1);
4154 out:
4155         /* we rely on s_frozen to stop further updates */
4156         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4157         return error;
4158 }
4159
4160 /*
4161  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4162  * flag here, even though the filesystem is not technically dirty yet.
4163  */
4164 static int ext4_unfreeze(struct super_block *sb)
4165 {
4166         if (sb->s_flags & MS_RDONLY)
4167                 return 0;
4168
4169         lock_super(sb);
4170         /* Reset the needs_recovery flag before the fs is unlocked. */
4171         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4172         ext4_commit_super(sb, 1);
4173         unlock_super(sb);
4174         return 0;
4175 }
4176
4177 /*
4178  * Structure to save mount options for ext4_remount's benefit
4179  */
4180 struct ext4_mount_options {
4181         unsigned long s_mount_opt;
4182         unsigned long s_mount_opt2;
4183         uid_t s_resuid;
4184         gid_t s_resgid;
4185         unsigned long s_commit_interval;
4186         u32 s_min_batch_time, s_max_batch_time;
4187 #ifdef CONFIG_QUOTA
4188         int s_jquota_fmt;
4189         char *s_qf_names[MAXQUOTAS];
4190 #endif
4191 };
4192
4193 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4194 {
4195         struct ext4_super_block *es;
4196         struct ext4_sb_info *sbi = EXT4_SB(sb);
4197         ext4_fsblk_t n_blocks_count = 0;
4198         unsigned long old_sb_flags;
4199         struct ext4_mount_options old_opts;
4200         int enable_quota = 0;
4201         ext4_group_t g;
4202         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4203         int err;
4204 #ifdef CONFIG_QUOTA
4205         int i;
4206 #endif
4207         char *orig_data = kstrdup(data, GFP_KERNEL);
4208
4209         /* Store the original options */
4210         lock_super(sb);
4211         old_sb_flags = sb->s_flags;
4212         old_opts.s_mount_opt = sbi->s_mount_opt;
4213         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4214         old_opts.s_resuid = sbi->s_resuid;
4215         old_opts.s_resgid = sbi->s_resgid;
4216         old_opts.s_commit_interval = sbi->s_commit_interval;
4217         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4218         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4219 #ifdef CONFIG_QUOTA
4220         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4221         for (i = 0; i < MAXQUOTAS; i++)
4222                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4223 #endif
4224         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4225                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4226
4227         /*
4228          * Allow the "check" option to be passed as a remount option.
4229          */
4230         if (!parse_options(data, sb, NULL, &journal_ioprio,
4231                            &n_blocks_count, 1)) {
4232                 err = -EINVAL;
4233                 goto restore_opts;
4234         }
4235
4236         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4237                 ext4_abort(sb, "Abort forced by user");
4238
4239         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4240                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4241
4242         es = sbi->s_es;
4243
4244         if (sbi->s_journal) {
4245                 ext4_init_journal_params(sb, sbi->s_journal);
4246                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4247         }
4248
4249         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4250                 n_blocks_count > ext4_blocks_count(es)) {
4251                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4252                         err = -EROFS;
4253                         goto restore_opts;
4254                 }
4255
4256                 if (*flags & MS_RDONLY) {
4257                         err = dquot_suspend(sb, -1);
4258                         if (err < 0)
4259                                 goto restore_opts;
4260
4261                         /*
4262                          * First of all, the unconditional stuff we have to do
4263                          * to disable replay of the journal when we next remount
4264                          */
4265                         sb->s_flags |= MS_RDONLY;
4266
4267                         /*
4268                          * OK, test if we are remounting a valid rw partition
4269                          * readonly, and if so set the rdonly flag and then
4270                          * mark the partition as valid again.
4271                          */
4272                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4273                             (sbi->s_mount_state & EXT4_VALID_FS))
4274                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4275
4276                         if (sbi->s_journal)
4277                                 ext4_mark_recovery_complete(sb, es);
4278                 } else {
4279                         /* Make sure we can mount this feature set readwrite */
4280                         if (!ext4_feature_set_ok(sb, 0)) {
4281                                 err = -EROFS;
4282                                 goto restore_opts;
4283                         }
4284                         /*
4285                          * Make sure the group descriptor checksums
4286                          * are sane.  If they aren't, refuse to remount r/w.
4287                          */
4288                         for (g = 0; g < sbi->s_groups_count; g++) {
4289                                 struct ext4_group_desc *gdp =
4290                                         ext4_get_group_desc(sb, g, NULL);
4291
4292                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4293                                         ext4_msg(sb, KERN_ERR,
4294                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4295                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4296                                                le16_to_cpu(gdp->bg_checksum));
4297                                         err = -EINVAL;
4298                                         goto restore_opts;
4299                                 }
4300                         }
4301
4302                         /*
4303                          * If we have an unprocessed orphan list hanging
4304                          * around from a previously readonly bdev mount,
4305                          * require a full umount/remount for now.
4306                          */
4307                         if (es->s_last_orphan) {
4308                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4309                                        "remount RDWR because of unprocessed "
4310                                        "orphan inode list.  Please "
4311                                        "umount/remount instead");
4312                                 err = -EINVAL;
4313                                 goto restore_opts;
4314                         }
4315
4316                         /*
4317                          * Mounting a RDONLY partition read-write, so reread
4318                          * and store the current valid flag.  (It may have
4319                          * been changed by e2fsck since we originally mounted
4320                          * the partition.)
4321                          */
4322                         if (sbi->s_journal)
4323                                 ext4_clear_journal_err(sb, es);
4324                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4325                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4326                                 goto restore_opts;
4327                         if (!ext4_setup_super(sb, es, 0))
4328                                 sb->s_flags &= ~MS_RDONLY;
4329                         enable_quota = 1;
4330                 }
4331         }
4332
4333         /*
4334          * Reinitialize lazy itable initialization thread based on
4335          * current settings
4336          */
4337         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4338                 ext4_unregister_li_request(sb);
4339         else {
4340                 ext4_group_t first_not_zeroed;
4341                 first_not_zeroed = ext4_has_uninit_itable(sb);
4342                 ext4_register_li_request(sb, first_not_zeroed);
4343         }
4344
4345         ext4_setup_system_zone(sb);
4346         if (sbi->s_journal == NULL)
4347                 ext4_commit_super(sb, 1);
4348
4349 #ifdef CONFIG_QUOTA
4350         /* Release old quota file names */
4351         for (i = 0; i < MAXQUOTAS; i++)
4352                 if (old_opts.s_qf_names[i] &&
4353                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4354                         kfree(old_opts.s_qf_names[i]);
4355 #endif
4356         unlock_super(sb);
4357         if (enable_quota)
4358                 dquot_resume(sb, -1);
4359
4360         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4361         kfree(orig_data);
4362         return 0;
4363
4364 restore_opts:
4365         sb->s_flags = old_sb_flags;
4366         sbi->s_mount_opt = old_opts.s_mount_opt;
4367         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4368         sbi->s_resuid = old_opts.s_resuid;
4369         sbi->s_resgid = old_opts.s_resgid;
4370         sbi->s_commit_interval = old_opts.s_commit_interval;
4371         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4372         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4373 #ifdef CONFIG_QUOTA
4374         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4375         for (i = 0; i < MAXQUOTAS; i++) {
4376                 if (sbi->s_qf_names[i] &&
4377                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4378                         kfree(sbi->s_qf_names[i]);
4379                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4380         }
4381 #endif
4382         unlock_super(sb);
4383         kfree(orig_data);
4384         return err;
4385 }
4386
4387 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4388 {
4389         struct super_block *sb = dentry->d_sb;
4390         struct ext4_sb_info *sbi = EXT4_SB(sb);
4391         struct ext4_super_block *es = sbi->s_es;
4392         u64 fsid;
4393
4394         if (test_opt(sb, MINIX_DF)) {
4395                 sbi->s_overhead_last = 0;
4396         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4397                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4398                 ext4_fsblk_t overhead = 0;
4399
4400                 /*
4401                  * Compute the overhead (FS structures).  This is constant
4402                  * for a given filesystem unless the number of block groups
4403                  * changes so we cache the previous value until it does.
4404                  */
4405
4406                 /*
4407                  * All of the blocks before first_data_block are
4408                  * overhead
4409                  */
4410                 overhead = le32_to_cpu(es->s_first_data_block);
4411
4412                 /*
4413                  * Add the overhead attributed to the superblock and
4414                  * block group descriptors.  If the sparse superblocks
4415                  * feature is turned on, then not all groups have this.
4416                  */
4417                 for (i = 0; i < ngroups; i++) {
4418                         overhead += ext4_bg_has_super(sb, i) +
4419                                 ext4_bg_num_gdb(sb, i);
4420                         cond_resched();
4421                 }
4422
4423                 /*
4424                  * Every block group has an inode bitmap, a block
4425                  * bitmap, and an inode table.
4426                  */
4427                 overhead += ngroups * (2 + sbi->s_itb_per_group);
4428                 sbi->s_overhead_last = overhead;
4429                 smp_wmb();
4430                 sbi->s_blocks_last = ext4_blocks_count(es);
4431         }
4432
4433         buf->f_type = EXT4_SUPER_MAGIC;
4434         buf->f_bsize = sb->s_blocksize;
4435         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
4436         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
4437                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
4438         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4439         if (buf->f_bfree < ext4_r_blocks_count(es))
4440                 buf->f_bavail = 0;
4441         buf->f_files = le32_to_cpu(es->s_inodes_count);
4442         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4443         buf->f_namelen = EXT4_NAME_LEN;
4444         fsid = le64_to_cpup((void *)es->s_uuid) ^
4445                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4446         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4447         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4448
4449         return 0;
4450 }
4451
4452 /* Helper function for writing quotas on sync - we need to start transaction
4453  * before quota file is locked for write. Otherwise the are possible deadlocks:
4454  * Process 1                         Process 2
4455  * ext4_create()                     quota_sync()
4456  *   jbd2_journal_start()                  write_dquot()
4457  *   dquot_initialize()                         down(dqio_mutex)
4458  *     down(dqio_mutex)                    jbd2_journal_start()
4459  *
4460  */
4461
4462 #ifdef CONFIG_QUOTA
4463
4464 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4465 {
4466         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4467 }
4468
4469 static int ext4_write_dquot(struct dquot *dquot)
4470 {
4471         int ret, err;
4472         handle_t *handle;
4473         struct inode *inode;
4474
4475         inode = dquot_to_inode(dquot);
4476         handle = ext4_journal_start(inode,
4477                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4478         if (IS_ERR(handle))
4479                 return PTR_ERR(handle);
4480         ret = dquot_commit(dquot);
4481         err = ext4_journal_stop(handle);
4482         if (!ret)
4483                 ret = err;
4484         return ret;
4485 }
4486
4487 static int ext4_acquire_dquot(struct dquot *dquot)
4488 {
4489         int ret, err;
4490         handle_t *handle;
4491
4492         handle = ext4_journal_start(dquot_to_inode(dquot),
4493                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4494         if (IS_ERR(handle))
4495                 return PTR_ERR(handle);
4496         ret = dquot_acquire(dquot);
4497         err = ext4_journal_stop(handle);
4498         if (!ret)
4499                 ret = err;
4500         return ret;
4501 }
4502
4503 static int ext4_release_dquot(struct dquot *dquot)
4504 {
4505         int ret, err;
4506         handle_t *handle;
4507
4508         handle = ext4_journal_start(dquot_to_inode(dquot),
4509                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4510         if (IS_ERR(handle)) {
4511                 /* Release dquot anyway to avoid endless cycle in dqput() */
4512                 dquot_release(dquot);
4513                 return PTR_ERR(handle);
4514         }
4515         ret = dquot_release(dquot);
4516         err = ext4_journal_stop(handle);
4517         if (!ret)
4518                 ret = err;
4519         return ret;
4520 }
4521
4522 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4523 {
4524         /* Are we journaling quotas? */
4525         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4526             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4527                 dquot_mark_dquot_dirty(dquot);
4528                 return ext4_write_dquot(dquot);
4529         } else {
4530                 return dquot_mark_dquot_dirty(dquot);
4531         }
4532 }
4533
4534 static int ext4_write_info(struct super_block *sb, int type)
4535 {
4536         int ret, err;
4537         handle_t *handle;
4538
4539         /* Data block + inode block */
4540         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4541         if (IS_ERR(handle))
4542                 return PTR_ERR(handle);
4543         ret = dquot_commit_info(sb, type);
4544         err = ext4_journal_stop(handle);
4545         if (!ret)
4546                 ret = err;
4547         return ret;
4548 }
4549
4550 /*
4551  * Turn on quotas during mount time - we need to find
4552  * the quota file and such...
4553  */
4554 static int ext4_quota_on_mount(struct super_block *sb, int type)
4555 {
4556         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4557                                         EXT4_SB(sb)->s_jquota_fmt, type);
4558 }
4559
4560 /*
4561  * Standard function to be called on quota_on
4562  */
4563 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4564                          char *name)
4565 {
4566         int err;
4567         struct path path;
4568
4569         if (!test_opt(sb, QUOTA))
4570                 return -EINVAL;
4571
4572         err = kern_path(name, LOOKUP_FOLLOW, &path);
4573         if (err)
4574                 return err;
4575
4576         /* Quotafile not on the same filesystem? */
4577         if (path.mnt->mnt_sb != sb) {
4578                 path_put(&path);
4579                 return -EXDEV;
4580         }
4581         /* Journaling quota? */
4582         if (EXT4_SB(sb)->s_qf_names[type]) {
4583                 /* Quotafile not in fs root? */
4584                 if (path.dentry->d_parent != sb->s_root)
4585                         ext4_msg(sb, KERN_WARNING,
4586                                 "Quota file not on filesystem root. "
4587                                 "Journaled quota will not work");
4588         }
4589
4590         /*
4591          * When we journal data on quota file, we have to flush journal to see
4592          * all updates to the file when we bypass pagecache...
4593          */
4594         if (EXT4_SB(sb)->s_journal &&
4595             ext4_should_journal_data(path.dentry->d_inode)) {
4596                 /*
4597                  * We don't need to lock updates but journal_flush() could
4598                  * otherwise be livelocked...
4599                  */
4600                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4601                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4602                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4603                 if (err) {
4604                         path_put(&path);
4605                         return err;
4606                 }
4607         }
4608
4609         err = dquot_quota_on_path(sb, type, format_id, &path);
4610         path_put(&path);
4611         return err;
4612 }
4613
4614 static int ext4_quota_off(struct super_block *sb, int type)
4615 {
4616         /* Force all delayed allocation blocks to be allocated.
4617          * Caller already holds s_umount sem */
4618         if (test_opt(sb, DELALLOC))
4619                 sync_filesystem(sb);
4620
4621         return dquot_quota_off(sb, type);
4622 }
4623
4624 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4625  * acquiring the locks... As quota files are never truncated and quota code
4626  * itself serializes the operations (and noone else should touch the files)
4627  * we don't have to be afraid of races */
4628 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4629                                size_t len, loff_t off)
4630 {
4631         struct inode *inode = sb_dqopt(sb)->files[type];
4632         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4633         int err = 0;
4634         int offset = off & (sb->s_blocksize - 1);
4635         int tocopy;
4636         size_t toread;
4637         struct buffer_head *bh;
4638         loff_t i_size = i_size_read(inode);
4639
4640         if (off > i_size)
4641                 return 0;
4642         if (off+len > i_size)
4643                 len = i_size-off;
4644         toread = len;
4645         while (toread > 0) {
4646                 tocopy = sb->s_blocksize - offset < toread ?
4647                                 sb->s_blocksize - offset : toread;
4648                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4649                 if (err)
4650                         return err;
4651                 if (!bh)        /* A hole? */
4652                         memset(data, 0, tocopy);
4653                 else
4654                         memcpy(data, bh->b_data+offset, tocopy);
4655                 brelse(bh);
4656                 offset = 0;
4657                 toread -= tocopy;
4658                 data += tocopy;
4659                 blk++;
4660         }
4661         return len;
4662 }
4663
4664 /* Write to quotafile (we know the transaction is already started and has
4665  * enough credits) */
4666 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4667                                 const char *data, size_t len, loff_t off)
4668 {
4669         struct inode *inode = sb_dqopt(sb)->files[type];
4670         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4671         int err = 0;
4672         int offset = off & (sb->s_blocksize - 1);
4673         struct buffer_head *bh;
4674         handle_t *handle = journal_current_handle();
4675
4676         if (EXT4_SB(sb)->s_journal && !handle) {
4677                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4678                         " cancelled because transaction is not started",
4679                         (unsigned long long)off, (unsigned long long)len);
4680                 return -EIO;
4681         }
4682         /*
4683          * Since we account only one data block in transaction credits,
4684          * then it is impossible to cross a block boundary.
4685          */
4686         if (sb->s_blocksize - offset < len) {
4687                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4688                         " cancelled because not block aligned",
4689                         (unsigned long long)off, (unsigned long long)len);
4690                 return -EIO;
4691         }
4692
4693         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4694         bh = ext4_bread(handle, inode, blk, 1, &err);
4695         if (!bh)
4696                 goto out;
4697         err = ext4_journal_get_write_access(handle, bh);
4698         if (err) {
4699                 brelse(bh);
4700                 goto out;
4701         }
4702         lock_buffer(bh);
4703         memcpy(bh->b_data+offset, data, len);
4704         flush_dcache_page(bh->b_page);
4705         unlock_buffer(bh);
4706         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4707         brelse(bh);
4708 out:
4709         if (err) {
4710                 mutex_unlock(&inode->i_mutex);
4711                 return err;
4712         }
4713         if (inode->i_size < off + len) {
4714                 i_size_write(inode, off + len);
4715                 EXT4_I(inode)->i_disksize = inode->i_size;
4716         }
4717         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4718         ext4_mark_inode_dirty(handle, inode);
4719         mutex_unlock(&inode->i_mutex);
4720         return len;
4721 }
4722
4723 #endif
4724
4725 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4726                        const char *dev_name, void *data)
4727 {
4728         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4729 }
4730
4731 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4732 static struct file_system_type ext2_fs_type = {
4733         .owner          = THIS_MODULE,
4734         .name           = "ext2",
4735         .mount          = ext4_mount,
4736         .kill_sb        = kill_block_super,
4737         .fs_flags       = FS_REQUIRES_DEV,
4738 };
4739
4740 static inline void register_as_ext2(void)
4741 {
4742         int err = register_filesystem(&ext2_fs_type);
4743         if (err)
4744                 printk(KERN_WARNING
4745                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4746 }
4747
4748 static inline void unregister_as_ext2(void)
4749 {
4750         unregister_filesystem(&ext2_fs_type);
4751 }
4752 MODULE_ALIAS("ext2");
4753 #else
4754 static inline void register_as_ext2(void) { }
4755 static inline void unregister_as_ext2(void) { }
4756 #endif
4757
4758 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4759 static inline void register_as_ext3(void)
4760 {
4761         int err = register_filesystem(&ext3_fs_type);
4762         if (err)
4763                 printk(KERN_WARNING
4764                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4765 }
4766
4767 static inline void unregister_as_ext3(void)
4768 {
4769         unregister_filesystem(&ext3_fs_type);
4770 }
4771 MODULE_ALIAS("ext3");
4772 #else
4773 static inline void register_as_ext3(void) { }
4774 static inline void unregister_as_ext3(void) { }
4775 #endif
4776
4777 static struct file_system_type ext4_fs_type = {
4778         .owner          = THIS_MODULE,
4779         .name           = "ext4",
4780         .mount          = ext4_mount,
4781         .kill_sb        = kill_block_super,
4782         .fs_flags       = FS_REQUIRES_DEV,
4783 };
4784
4785 int __init ext4_init_feat_adverts(void)
4786 {
4787         struct ext4_features *ef;
4788         int ret = -ENOMEM;
4789
4790         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4791         if (!ef)
4792                 goto out;
4793
4794         ef->f_kobj.kset = ext4_kset;
4795         init_completion(&ef->f_kobj_unregister);
4796         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4797                                    "features");
4798         if (ret) {
4799                 kfree(ef);
4800                 goto out;
4801         }
4802
4803         ext4_feat = ef;
4804         ret = 0;
4805 out:
4806         return ret;
4807 }
4808
4809 static int __init ext4_init_fs(void)
4810 {
4811         int err;
4812
4813         ext4_check_flag_values();
4814         err = ext4_init_pageio();
4815         if (err)
4816                 return err;
4817         err = ext4_init_system_zone();
4818         if (err)
4819                 goto out5;
4820         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4821         if (!ext4_kset)
4822                 goto out4;
4823         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4824
4825         err = ext4_init_feat_adverts();
4826
4827         err = ext4_init_mballoc();
4828         if (err)
4829                 goto out3;
4830
4831         err = ext4_init_xattr();
4832         if (err)
4833                 goto out2;
4834         err = init_inodecache();
4835         if (err)
4836                 goto out1;
4837         register_as_ext2();
4838         register_as_ext3();
4839         err = register_filesystem(&ext4_fs_type);
4840         if (err)
4841                 goto out;
4842
4843         ext4_li_info = NULL;
4844         mutex_init(&ext4_li_mtx);
4845         return 0;
4846 out:
4847         unregister_as_ext2();
4848         unregister_as_ext3();
4849         destroy_inodecache();
4850 out1:
4851         ext4_exit_xattr();
4852 out2:
4853         ext4_exit_mballoc();
4854 out3:
4855         kfree(ext4_feat);
4856         remove_proc_entry("fs/ext4", NULL);
4857         kset_unregister(ext4_kset);
4858 out4:
4859         ext4_exit_system_zone();
4860 out5:
4861         ext4_exit_pageio();
4862         return err;
4863 }
4864
4865 static void __exit ext4_exit_fs(void)
4866 {
4867         ext4_destroy_lazyinit_thread();
4868         unregister_as_ext2();
4869         unregister_as_ext3();
4870         unregister_filesystem(&ext4_fs_type);
4871         destroy_inodecache();
4872         ext4_exit_xattr();
4873         ext4_exit_mballoc();
4874         remove_proc_entry("fs/ext4", NULL);
4875         kset_unregister(ext4_kset);
4876         ext4_exit_system_zone();
4877         ext4_exit_pageio();
4878 }
4879
4880 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4881 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4882 MODULE_LICENSE("GPL");
4883 module_init(ext4_init_fs)
4884 module_exit(ext4_exit_fs)