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