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