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/smp_lock.h>
  30 #include <linux/buffer_head.h>
  31 #include <linux/exportfs.h>
  32 #include <linux/vfs.h>
  33 #include <linux/random.h>
  34 #include <linux/mount.h>
  35 #include <linux/namei.h>
  36 #include <linux/quotaops.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/proc_fs.h>
  39 #include <linux/ctype.h>
  40 #include <linux/log2.h>
  41 #include <linux/crc16.h>
  42 #include <asm/uaccess.h>
  43
  44 #include "ext4.h"
  45 #include "ext4_jbd2.h"
  46 #include "xattr.h"
  47 #include "acl.h"
  48 #include "mballoc.h"
  49
  50 #define CREATE_TRACE_POINTS
  51 #include <trace/events/ext4.h>
  52
  53 struct proc_dir_entry *ext4_proc_root;
  54 static struct kset *ext4_kset;
  55
  56 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  57                              unsigned long journal_devnum);
  58 static int ext4_commit_super(struct super_block *sb, int sync);
  59 static void ext4_mark_recovery_complete(struct super_block *sb,
  60                                         struct ext4_super_block *es);
  61 static void ext4_clear_journal_err(struct super_block *sb,
  62                                    struct ext4_super_block *es);
  63 static int ext4_sync_fs(struct super_block *sb, int wait);
  64 static const char *ext4_decode_error(struct super_block *sb, int errno,
  65                                      char nbuf[16]);
  66 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  67 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  68 static int ext4_unfreeze(struct super_block *sb);
  69 static void ext4_write_super(struct super_block *sb);
  70 static int ext4_freeze(struct super_block *sb);
  71 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
  72                        const char *dev_name, void *data, struct vfsmount *mnt);
  73
  74 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  75 static struct file_system_type ext3_fs_type = {
  76         .owner          = THIS_MODULE,
  77         .name           = "ext3",
  78         .get_sb         = ext4_get_sb,
  79         .kill_sb        = kill_block_super,
  80         .fs_flags       = FS_REQUIRES_DEV,
  81 };
  82 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
  83 #else
  84 #define IS_EXT3_SB(sb) (0)
  85 #endif
  86
  87 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
  88                                struct ext4_group_desc *bg)
  89 {
  90         return le32_to_cpu(bg->bg_block_bitmap_lo) |
  91                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
  92                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
  93 }
  94
  95 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
  96                                struct ext4_group_desc *bg)
  97 {
  98         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
  99                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 100                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 101 }
 102
 103 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 104                               struct ext4_group_desc *bg)
 105 {
 106         return le32_to_cpu(bg->bg_inode_table_lo) |
 107                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 108                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 109 }
 110
 111 __u32 ext4_free_blks_count(struct super_block *sb,
 112                               struct ext4_group_desc *bg)
 113 {
 114         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 115                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 116                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 117 }
 118
 119 __u32 ext4_free_inodes_count(struct super_block *sb,
 120                               struct ext4_group_desc *bg)
 121 {
 122         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 123                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 124                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 125 }
 126
 127 __u32 ext4_used_dirs_count(struct super_block *sb,
 128                               struct ext4_group_desc *bg)
 129 {
 130         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 131                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 132                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 133 }
 134
 135 __u32 ext4_itable_unused_count(struct super_block *sb,
 136                               struct ext4_group_desc *bg)
 137 {
 138         return le16_to_cpu(bg->bg_itable_unused_lo) |
 139                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 140                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 141 }
 142
 143 void ext4_block_bitmap_set(struct super_block *sb,
 144                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 145 {
 146         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 147         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 148                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 149 }
 150
 151 void ext4_inode_bitmap_set(struct super_block *sb,
 152                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 153 {
 154         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 155         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 156                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 157 }
 158
 159 void ext4_inode_table_set(struct super_block *sb,
 160                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 161 {
 162         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 163         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 164                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 165 }
 166
 167 void ext4_free_blks_set(struct super_block *sb,
 168                           struct ext4_group_desc *bg, __u32 count)
 169 {
 170         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 171         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 172                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 173 }
 174
 175 void ext4_free_inodes_set(struct super_block *sb,
 176                           struct ext4_group_desc *bg, __u32 count)
 177 {
 178         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 179         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 180                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 181 }
 182
 183 void ext4_used_dirs_set(struct super_block *sb,
 184                           struct ext4_group_desc *bg, __u32 count)
 185 {
 186         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 187         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 188                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 189 }
 190
 191 void ext4_itable_unused_set(struct super_block *sb,
 192                           struct ext4_group_desc *bg, __u32 count)
 193 {
 194         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 195         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 196                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 197 }
 198
 199
 200 /* Just increment the non-pointer handle value */
 201 static handle_t *ext4_get_nojournal(void)
 202 {
 203         handle_t *handle = current->journal_info;
 204         unsigned long ref_cnt = (unsigned long)handle;
 205
 206         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 207
 208         ref_cnt++;
 209         handle = (handle_t *)ref_cnt;
 210
 211         current->journal_info = handle;
 212         return handle;
 213 }
 214
 215
 216 /* Decrement the non-pointer handle value */
 217 static void ext4_put_nojournal(handle_t *handle)
 218 {
 219         unsigned long ref_cnt = (unsigned long)handle;
 220
 221         BUG_ON(ref_cnt == 0);
 222
 223         ref_cnt--;
 224         handle = (handle_t *)ref_cnt;
 225
 226         current->journal_info = handle;
 227 }
 228
 229 /*
 230  * Wrappers for jbd2_journal_start/end.
 231  *
 232  * The only special thing we need to do here is to make sure that all
 233  * journal_end calls result in the superblock being marked dirty, so
 234  * that sync() will call the filesystem's write_super callback if
 235  * appropriate.
 236  */
 237 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 238 {
 239         journal_t *journal;
 240
 241         if (sb->s_flags & MS_RDONLY)
 242                 return ERR_PTR(-EROFS);
 243
 244         vfs_check_frozen(sb, SB_FREEZE_WRITE);
 245         /* Special case here: if the journal has aborted behind our
 246          * backs (eg. EIO in the commit thread), then we still need to
 247          * take the FS itself readonly cleanly. */
 248         journal = EXT4_SB(sb)->s_journal;
 249         if (journal) {
 250                 if (is_journal_aborted(journal)) {
 251                         ext4_abort(sb, __func__, "Detected aborted journal");
 252                         return ERR_PTR(-EROFS);
 253                 }
 254                 return jbd2_journal_start(journal, nblocks);
 255         }
 256         return ext4_get_nojournal();
 257 }
 258
 259 /*
 260  * The only special thing we need to do here is to make sure that all
 261  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 262  * that sync() will call the filesystem's write_super callback if
 263  * appropriate.
 264  */
 265 int __ext4_journal_stop(const char *where, handle_t *handle)
 266 {
 267         struct super_block *sb;
 268         int err;
 269         int rc;
 270
 271         if (!ext4_handle_valid(handle)) {
 272                 ext4_put_nojournal(handle);
 273                 return 0;
 274         }
 275         sb = handle->h_transaction->t_journal->j_private;
 276         err = handle->h_err;
 277         rc = jbd2_journal_stop(handle);
 278
 279         if (!err)
 280                 err = rc;
 281         if (err)
 282                 __ext4_std_error(sb, where, err);
 283         return err;
 284 }
 285
 286 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
 287                 struct buffer_head *bh, handle_t *handle, int err)
 288 {
 289         char nbuf[16];
 290         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 291
 292         BUG_ON(!ext4_handle_valid(handle));
 293
 294         if (bh)
 295                 BUFFER_TRACE(bh, "abort");
 296
 297         if (!handle->h_err)
 298                 handle->h_err = err;
 299
 300         if (is_handle_aborted(handle))
 301                 return;
 302
 303         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 304                caller, errstr, err_fn);
 305
 306         jbd2_journal_abort_handle(handle);
 307 }
 308
 309 /* Deal with the reporting of failure conditions on a filesystem such as
 310  * inconsistencies detected or read IO failures.
 311  *
 312  * On ext2, we can store the error state of the filesystem in the
 313  * superblock.  That is not possible on ext4, because we may have other
 314  * write ordering constraints on the superblock which prevent us from
 315  * writing it out straight away; and given that the journal is about to
 316  * be aborted, we can't rely on the current, or future, transactions to
 317  * write out the superblock safely.
 318  *
 319  * We'll just use the jbd2_journal_abort() error code to record an error in
 320  * the journal instead.  On recovery, the journal will complain about
 321  * that error until we've noted it down and cleared it.
 322  */
 323
 324 static void ext4_handle_error(struct super_block *sb)
 325 {
 326         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 327
 328         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 329         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 330
 331         if (sb->s_flags & MS_RDONLY)
 332                 return;
 333
 334         if (!test_opt(sb, ERRORS_CONT)) {
 335                 journal_t *journal = EXT4_SB(sb)->s_journal;
 336
 337                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 338                 if (journal)
 339                         jbd2_journal_abort(journal, -EIO);
 340         }
 341         if (test_opt(sb, ERRORS_RO)) {
 342                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 343                 sb->s_flags |= MS_RDONLY;
 344         }
 345         ext4_commit_super(sb, 1);
 346         if (test_opt(sb, ERRORS_PANIC))
 347                 panic("EXT4-fs (device %s): panic forced after error\n",
 348                         sb->s_id);
 349 }
 350
 351 void __ext4_error(struct super_block *sb, const char *function,
 352                 const char *fmt, ...)
 353 {
 354         va_list args;
 355
 356         va_start(args, fmt);
 357         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 358         vprintk(fmt, args);
 359         printk("\n");
 360         va_end(args);
 361
 362         ext4_handle_error(sb);
 363 }
 364
 365 void ext4_error_inode(const char *function, struct inode *inode,
 366                       const char *fmt, ...)
 367 {
 368         va_list args;
 369
 370         va_start(args, fmt);
 371         printk(KERN_CRIT "EXT4-fs error (device %s): %s: inode #%lu: (comm %s) ",
 372                inode->i_sb->s_id, function, inode->i_ino, current->comm);
 373         vprintk(fmt, args);
 374         printk("\n");
 375         va_end(args);
 376
 377         ext4_handle_error(inode->i_sb);
 378 }
 379
 380 void ext4_error_file(const char *function, struct file *file,
 381                      const char *fmt, ...)
 382 {
 383         va_list args;
 384         struct inode *inode = file->f_dentry->d_inode;
 385         char pathname[80], *path;
 386
 387         va_start(args, fmt);
 388         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 389         if (!path)
 390                 path = "(unknown)";
 391         printk(KERN_CRIT
 392                "EXT4-fs error (device %s): %s: inode #%lu (comm %s path %s): ",
 393                inode->i_sb->s_id, function, inode->i_ino, current->comm, path);
 394         vprintk(fmt, args);
 395         printk("\n");
 396         va_end(args);
 397
 398         ext4_handle_error(inode->i_sb);
 399 }
 400
 401 static const char *ext4_decode_error(struct super_block *sb, int errno,
 402                                      char nbuf[16])
 403 {
 404         char *errstr = NULL;
 405
 406         switch (errno) {
 407         case -EIO:
 408                 errstr = "IO failure";
 409                 break;
 410         case -ENOMEM:
 411                 errstr = "Out of memory";
 412                 break;
 413         case -EROFS:
 414                 if (!sb || (EXT4_SB(sb)->s_journal &&
 415                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 416                         errstr = "Journal has aborted";
 417                 else
 418                         errstr = "Readonly filesystem";
 419                 break;
 420         default:
 421                 /* If the caller passed in an extra buffer for unknown
 422                  * errors, textualise them now.  Else we just return
 423                  * NULL. */
 424                 if (nbuf) {
 425                         /* Check for truncated error codes... */
 426                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 427                                 errstr = nbuf;
 428                 }
 429                 break;
 430         }
 431
 432         return errstr;
 433 }
 434
 435 /* __ext4_std_error decodes expected errors from journaling functions
 436  * automatically and invokes the appropriate error response.  */
 437
 438 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
 439 {
 440         char nbuf[16];
 441         const char *errstr;
 442
 443         /* Special case: if the error is EROFS, and we're not already
 444          * inside a transaction, then there's really no point in logging
 445          * an error. */
 446         if (errno == -EROFS && journal_current_handle() == NULL &&
 447             (sb->s_flags & MS_RDONLY))
 448                 return;
 449
 450         errstr = ext4_decode_error(sb, errno, nbuf);
 451         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
 452                sb->s_id, function, errstr);
 453
 454         ext4_handle_error(sb);
 455 }
 456
 457 /*
 458  * ext4_abort is a much stronger failure handler than ext4_error.  The
 459  * abort function may be used to deal with unrecoverable failures such
 460  * as journal IO errors or ENOMEM at a critical moment in log management.
 461  *
 462  * We unconditionally force the filesystem into an ABORT|READONLY state,
 463  * unless the error response on the fs has been set to panic in which
 464  * case we take the easy way out and panic immediately.
 465  */
 466
 467 void ext4_abort(struct super_block *sb, const char *function,
 468                 const char *fmt, ...)
 469 {
 470         va_list args;
 471
 472         va_start(args, fmt);
 473         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 474         vprintk(fmt, args);
 475         printk("\n");
 476         va_end(args);
 477
 478         if (test_opt(sb, ERRORS_PANIC))
 479                 panic("EXT4-fs panic from previous error\n");
 480
 481         if (sb->s_flags & MS_RDONLY)
 482                 return;
 483
 484         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 485         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 486         sb->s_flags |= MS_RDONLY;
 487         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 488         if (EXT4_SB(sb)->s_journal)
 489                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 490 }
 491
 492 void ext4_msg (struct super_block * sb, const char *prefix,
 493                    const char *fmt, ...)
 494 {
 495         va_list args;
 496
 497         va_start(args, fmt);
 498         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
 499         vprintk(fmt, args);
 500         printk("\n");
 501         va_end(args);
 502 }
 503
 504 void __ext4_warning(struct super_block *sb, const char *function,
 505                   const char *fmt, ...)
 506 {
 507         va_list args;
 508
 509         va_start(args, fmt);
 510         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
 511                sb->s_id, function);
 512         vprintk(fmt, args);
 513         printk("\n");
 514         va_end(args);
 515 }
 516
 517 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
 518                            const char *function, const char *fmt, ...)
 519 __releases(bitlock)
 520 __acquires(bitlock)
 521 {
 522         va_list args;
 523         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 524
 525         va_start(args, fmt);
 526         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
 527         vprintk(fmt, args);
 528         printk("\n");
 529         va_end(args);
 530
 531         if (test_opt(sb, ERRORS_CONT)) {
 532                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 533                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 534                 ext4_commit_super(sb, 0);
 535                 return;
 536         }
 537         ext4_unlock_group(sb, grp);
 538         ext4_handle_error(sb);
 539         /*
 540          * We only get here in the ERRORS_RO case; relocking the group
 541          * may be dangerous, but nothing bad will happen since the
 542          * filesystem will have already been marked read/only and the
 543          * journal has been aborted.  We return 1 as a hint to callers
 544          * who might what to use the return value from
 545          * ext4_grp_locked_error() to distinguish beween the
 546          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 547          * aggressively from the ext4 function in question, with a
 548          * more appropriate error code.
 549          */
 550         ext4_lock_group(sb, grp);
 551         return;
 552 }
 553
 554 void ext4_update_dynamic_rev(struct super_block *sb)
 555 {
 556         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 557
 558         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 559                 return;
 560
 561         ext4_warning(sb,
 562                      "updating to rev %d because of new feature flag, "
 563                      "running e2fsck is recommended",
 564                      EXT4_DYNAMIC_REV);
 565
 566         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 567         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 568         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 569         /* leave es->s_feature_*compat flags alone */
 570         /* es->s_uuid will be set by e2fsck if empty */
 571
 572         /*
 573          * The rest of the superblock fields should be zero, and if not it
 574          * means they are likely already in use, so leave them alone.  We
 575          * can leave it up to e2fsck to clean up any inconsistencies there.
 576          */
 577 }
 578
 579 /*
 580  * Open the external journal device
 581  */
 582 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 583 {
 584         struct block_device *bdev;
 585         char b[BDEVNAME_SIZE];
 586
 587         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 588         if (IS_ERR(bdev))
 589                 goto fail;
 590         return bdev;
 591
 592 fail:
 593         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 594                         __bdevname(dev, b), PTR_ERR(bdev));
 595         return NULL;
 596 }
 597
 598 /*
 599  * Release the journal device
 600  */
 601 static int ext4_blkdev_put(struct block_device *bdev)
 602 {
 603         bd_release(bdev);
 604         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
 605 }
 606
 607 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 608 {
 609         struct block_device *bdev;
 610         int ret = -ENODEV;
 611
 612         bdev = sbi->journal_bdev;
 613         if (bdev) {
 614                 ret = ext4_blkdev_put(bdev);
 615                 sbi->journal_bdev = NULL;
 616         }
 617         return ret;
 618 }
 619
 620 static inline struct inode *orphan_list_entry(struct list_head *l)
 621 {
 622         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 623 }
 624
 625 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 626 {
 627         struct list_head *l;
 628
 629         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 630                  le32_to_cpu(sbi->s_es->s_last_orphan));
 631
 632         printk(KERN_ERR "sb_info orphan list:\n");
 633         list_for_each(l, &sbi->s_orphan) {
 634                 struct inode *inode = orphan_list_entry(l);
 635                 printk(KERN_ERR "  "
 636                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 637                        inode->i_sb->s_id, inode->i_ino, inode,
 638                        inode->i_mode, inode->i_nlink,
 639                        NEXT_ORPHAN(inode));
 640         }
 641 }
 642
 643 static void ext4_put_super(struct super_block *sb)
 644 {
 645         struct ext4_sb_info *sbi = EXT4_SB(sb);
 646         struct ext4_super_block *es = sbi->s_es;
 647         int i, err;
 648
 649         flush_workqueue(sbi->dio_unwritten_wq);
 650         destroy_workqueue(sbi->dio_unwritten_wq);
 651
 652         lock_super(sb);
 653         lock_kernel();
 654         if (sb->s_dirt)
 655                 ext4_commit_super(sb, 1);
 656
 657         if (sbi->s_journal) {
 658                 err = jbd2_journal_destroy(sbi->s_journal);
 659                 sbi->s_journal = NULL;
 660                 if (err < 0)
 661                         ext4_abort(sb, __func__,
 662                                    "Couldn't clean up the journal");
 663         }
 664
 665         ext4_release_system_zone(sb);
 666         ext4_mb_release(sb);
 667         ext4_ext_release(sb);
 668         ext4_xattr_put_super(sb);
 669
 670         if (!(sb->s_flags & MS_RDONLY)) {
 671                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 672                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 673                 ext4_commit_super(sb, 1);
 674         }
 675         if (sbi->s_proc) {
 676                 remove_proc_entry(sb->s_id, ext4_proc_root);
 677         }
 678         kobject_del(&sbi->s_kobj);
 679
 680         for (i = 0; i < sbi->s_gdb_count; i++)
 681                 brelse(sbi->s_group_desc[i]);
 682         kfree(sbi->s_group_desc);
 683         if (is_vmalloc_addr(sbi->s_flex_groups))
 684                 vfree(sbi->s_flex_groups);
 685         else
 686                 kfree(sbi->s_flex_groups);
 687         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 688         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 689         percpu_counter_destroy(&sbi->s_dirs_counter);
 690         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 691         brelse(sbi->s_sbh);
 692 #ifdef CONFIG_QUOTA
 693         for (i = 0; i < MAXQUOTAS; i++)
 694                 kfree(sbi->s_qf_names[i]);
 695 #endif
 696
 697         /* Debugging code just in case the in-memory inode orphan list
 698          * isn't empty.  The on-disk one can be non-empty if we've
 699          * detected an error and taken the fs readonly, but the
 700          * in-memory list had better be clean by this point. */
 701         if (!list_empty(&sbi->s_orphan))
 702                 dump_orphan_list(sb, sbi);
 703         J_ASSERT(list_empty(&sbi->s_orphan));
 704
 705         invalidate_bdev(sb->s_bdev);
 706         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 707                 /*
 708                  * Invalidate the journal device's buffers.  We don't want them
 709                  * floating about in memory - the physical journal device may
 710                  * hotswapped, and it breaks the `ro-after' testing code.
 711                  */
 712                 sync_blockdev(sbi->journal_bdev);
 713                 invalidate_bdev(sbi->journal_bdev);
 714                 ext4_blkdev_remove(sbi);
 715         }
 716         sb->s_fs_info = NULL;
 717         /*
 718          * Now that we are completely done shutting down the
 719          * superblock, we need to actually destroy the kobject.
 720          */
 721         unlock_kernel();
 722         unlock_super(sb);
 723         kobject_put(&sbi->s_kobj);
 724         wait_for_completion(&sbi->s_kobj_unregister);
 725         kfree(sbi->s_blockgroup_lock);
 726         kfree(sbi);
 727 }
 728
 729 static struct kmem_cache *ext4_inode_cachep;
 730
 731 /*
 732  * Called inside transaction, so use GFP_NOFS
 733  */
 734 static struct inode *ext4_alloc_inode(struct super_block *sb)
 735 {
 736         struct ext4_inode_info *ei;
 737
 738         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 739         if (!ei)
 740                 return NULL;
 741
 742         ei->vfs_inode.i_version = 1;
 743         ei->vfs_inode.i_data.writeback_index = 0;
 744         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 745         INIT_LIST_HEAD(&ei->i_prealloc_list);
 746         spin_lock_init(&ei->i_prealloc_lock);
 747         /*
 748          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 749          * therefore it can be null here.  Don't check it, just initialize
 750          * jinode.
 751          */
 752         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 753         ei->i_reserved_data_blocks = 0;
 754         ei->i_reserved_meta_blocks = 0;
 755         ei->i_allocated_meta_blocks = 0;
 756         ei->i_da_metadata_calc_len = 0;
 757         ei->i_delalloc_reserved_flag = 0;
 758         spin_lock_init(&(ei->i_block_reservation_lock));
 759 #ifdef CONFIG_QUOTA
 760         ei->i_reserved_quota = 0;
 761 #endif
 762         INIT_LIST_HEAD(&ei->i_completed_io_list);
 763         spin_lock_init(&ei->i_completed_io_lock);
 764         ei->cur_aio_dio = NULL;
 765         ei->i_sync_tid = 0;
 766         ei->i_datasync_tid = 0;
 767
 768         return &ei->vfs_inode;
 769 }
 770
 771 static void ext4_destroy_inode(struct inode *inode)
 772 {
 773         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 774                 ext4_msg(inode->i_sb, KERN_ERR,
 775                          "Inode %lu (%p): orphan list check failed!",
 776                          inode->i_ino, EXT4_I(inode));
 777                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 778                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 779                                 true);
 780                 dump_stack();
 781         }
 782         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 783 }
 784
 785 static void init_once(void *foo)
 786 {
 787         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 788
 789         INIT_LIST_HEAD(&ei->i_orphan);
 790 #ifdef CONFIG_EXT4_FS_XATTR
 791         init_rwsem(&ei->xattr_sem);
 792 #endif
 793         init_rwsem(&ei->i_data_sem);
 794         inode_init_once(&ei->vfs_inode);
 795 }
 796
 797 static int init_inodecache(void)
 798 {
 799         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 800                                              sizeof(struct ext4_inode_info),
 801                                              0, (SLAB_RECLAIM_ACCOUNT|
 802                                                 SLAB_MEM_SPREAD),
 803                                              init_once);
 804         if (ext4_inode_cachep == NULL)
 805                 return -ENOMEM;
 806         return 0;
 807 }
 808
 809 static void destroy_inodecache(void)
 810 {
 811         kmem_cache_destroy(ext4_inode_cachep);
 812 }
 813
 814 static void ext4_clear_inode(struct inode *inode)
 815 {
 816         dquot_drop(inode);
 817         ext4_discard_preallocations(inode);
 818         if (EXT4_JOURNAL(inode))
 819                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 820                                        &EXT4_I(inode)->jinode);
 821 }
 822
 823 static inline void ext4_show_quota_options(struct seq_file *seq,
 824                                            struct super_block *sb)
 825 {
 826 #if defined(CONFIG_QUOTA)
 827         struct ext4_sb_info *sbi = EXT4_SB(sb);
 828
 829         if (sbi->s_jquota_fmt) {
 830                 char *fmtname = "";
 831
 832                 switch (sbi->s_jquota_fmt) {
 833                 case QFMT_VFS_OLD:
 834                         fmtname = "vfsold";
 835                         break;
 836                 case QFMT_VFS_V0:
 837                         fmtname = "vfsv0";
 838                         break;
 839                 case QFMT_VFS_V1:
 840                         fmtname = "vfsv1";
 841                         break;
 842                 }
 843                 seq_printf(seq, ",jqfmt=%s", fmtname);
 844         }
 845
 846         if (sbi->s_qf_names[USRQUOTA])
 847                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 848
 849         if (sbi->s_qf_names[GRPQUOTA])
 850                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 851
 852         if (test_opt(sb, USRQUOTA))
 853                 seq_puts(seq, ",usrquota");
 854
 855         if (test_opt(sb, GRPQUOTA))
 856                 seq_puts(seq, ",grpquota");
 857 #endif
 858 }
 859
 860 /*
 861  * Show an option if
 862  *  - it's set to a non-default value OR
 863  *  - if the per-sb default is different from the global default
 864  */
 865 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 866 {
 867         int def_errors;
 868         unsigned long def_mount_opts;
 869         struct super_block *sb = vfs->mnt_sb;
 870         struct ext4_sb_info *sbi = EXT4_SB(sb);
 871         struct ext4_super_block *es = sbi->s_es;
 872
 873         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 874         def_errors     = le16_to_cpu(es->s_errors);
 875
 876         if (sbi->s_sb_block != 1)
 877                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 878         if (test_opt(sb, MINIX_DF))
 879                 seq_puts(seq, ",minixdf");
 880         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 881                 seq_puts(seq, ",grpid");
 882         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 883                 seq_puts(seq, ",nogrpid");
 884         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 885             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 886                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 887         }
 888         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 889             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 890                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 891         }
 892         if (test_opt(sb, ERRORS_RO)) {
 893                 if (def_errors == EXT4_ERRORS_PANIC ||
 894                     def_errors == EXT4_ERRORS_CONTINUE) {
 895                         seq_puts(seq, ",errors=remount-ro");
 896                 }
 897         }
 898         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 899                 seq_puts(seq, ",errors=continue");
 900         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 901                 seq_puts(seq, ",errors=panic");
 902         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 903                 seq_puts(seq, ",nouid32");
 904         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 905                 seq_puts(seq, ",debug");
 906         if (test_opt(sb, OLDALLOC))
 907                 seq_puts(seq, ",oldalloc");
 908 #ifdef CONFIG_EXT4_FS_XATTR
 909         if (test_opt(sb, XATTR_USER) &&
 910                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 911                 seq_puts(seq, ",user_xattr");
 912         if (!test_opt(sb, XATTR_USER) &&
 913             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 914                 seq_puts(seq, ",nouser_xattr");
 915         }
 916 #endif
 917 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 918         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 919                 seq_puts(seq, ",acl");
 920         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 921                 seq_puts(seq, ",noacl");
 922 #endif
 923         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 924                 seq_printf(seq, ",commit=%u",
 925                            (unsigned) (sbi->s_commit_interval / HZ));
 926         }
 927         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 928                 seq_printf(seq, ",min_batch_time=%u",
 929                            (unsigned) sbi->s_min_batch_time);
 930         }
 931         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 932                 seq_printf(seq, ",max_batch_time=%u",
 933                            (unsigned) sbi->s_min_batch_time);
 934         }
 935
 936         /*
 937          * We're changing the default of barrier mount option, so
 938          * let's always display its mount state so it's clear what its
 939          * status is.
 940          */
 941         seq_puts(seq, ",barrier=");
 942         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 943         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 944                 seq_puts(seq, ",journal_async_commit");
 945         else if (test_opt(sb, JOURNAL_CHECKSUM))
 946                 seq_puts(seq, ",journal_checksum");
 947         if (test_opt(sb, NOBH))
 948                 seq_puts(seq, ",nobh");
 949         if (test_opt(sb, I_VERSION))
 950                 seq_puts(seq, ",i_version");
 951         if (!test_opt(sb, DELALLOC))
 952                 seq_puts(seq, ",nodelalloc");
 953
 954
 955         if (sbi->s_stripe)
 956                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 957         /*
 958          * journal mode get enabled in different ways
 959          * So just print the value even if we didn't specify it
 960          */
 961         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 962                 seq_puts(seq, ",data=journal");
 963         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 964                 seq_puts(seq, ",data=ordered");
 965         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 966                 seq_puts(seq, ",data=writeback");
 967
 968         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 969                 seq_printf(seq, ",inode_readahead_blks=%u",
 970                            sbi->s_inode_readahead_blks);
 971
 972         if (test_opt(sb, DATA_ERR_ABORT))
 973                 seq_puts(seq, ",data_err=abort");
 974
 975         if (test_opt(sb, NO_AUTO_DA_ALLOC))
 976                 seq_puts(seq, ",noauto_da_alloc");
 977
 978         if (test_opt(sb, DISCARD))
 979                 seq_puts(seq, ",discard");
 980
 981         if (test_opt(sb, NOLOAD))
 982                 seq_puts(seq, ",norecovery");
 983
 984         if (test_opt(sb, DIOREAD_NOLOCK))
 985                 seq_puts(seq, ",dioread_nolock");
 986
 987         ext4_show_quota_options(seq, sb);
 988
 989         return 0;
 990 }
 991
 992 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 993                                         u64 ino, u32 generation)
 994 {
 995         struct inode *inode;
 996
 997         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 998                 return ERR_PTR(-ESTALE);
 999         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1000                 return ERR_PTR(-ESTALE);
1001
1002         /* iget isn't really right if the inode is currently unallocated!!
1003          *
1004          * ext4_read_inode will return a bad_inode if the inode had been
1005          * deleted, so we should be safe.
1006          *
1007          * Currently we don't know the generation for parent directory, so
1008          * a generation of 0 means "accept any"
1009          */
1010         inode = ext4_iget(sb, ino);
1011         if (IS_ERR(inode))
1012                 return ERR_CAST(inode);
1013         if (generation && inode->i_generation != generation) {
1014                 iput(inode);
1015                 return ERR_PTR(-ESTALE);
1016         }
1017
1018         return inode;
1019 }
1020
1021 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1022                                         int fh_len, int fh_type)
1023 {
1024         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1025                                     ext4_nfs_get_inode);
1026 }
1027
1028 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1029                                         int fh_len, int fh_type)
1030 {
1031         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1032                                     ext4_nfs_get_inode);
1033 }
1034
1035 /*
1036  * Try to release metadata pages (indirect blocks, directories) which are
1037  * mapped via the block device.  Since these pages could have journal heads
1038  * which would prevent try_to_free_buffers() from freeing them, we must use
1039  * jbd2 layer's try_to_free_buffers() function to release them.
1040  */
1041 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1042                                  gfp_t wait)
1043 {
1044         journal_t *journal = EXT4_SB(sb)->s_journal;
1045
1046         WARN_ON(PageChecked(page));
1047         if (!page_has_buffers(page))
1048                 return 0;
1049         if (journal)
1050                 return jbd2_journal_try_to_free_buffers(journal, page,
1051                                                         wait & ~__GFP_WAIT);
1052         return try_to_free_buffers(page);
1053 }
1054
1055 #ifdef CONFIG_QUOTA
1056 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1057 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1058
1059 static int ext4_write_dquot(struct dquot *dquot);
1060 static int ext4_acquire_dquot(struct dquot *dquot);
1061 static int ext4_release_dquot(struct dquot *dquot);
1062 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1063 static int ext4_write_info(struct super_block *sb, int type);
1064 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1065                                 char *path, int remount);
1066 static int ext4_quota_on_mount(struct super_block *sb, int type);
1067 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1068                                size_t len, loff_t off);
1069 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1070                                 const char *data, size_t len, loff_t off);
1071
1072 static const struct dquot_operations ext4_quota_operations = {
1073 #ifdef CONFIG_QUOTA
1074         .get_reserved_space = ext4_get_reserved_space,
1075 #endif
1076         .write_dquot    = ext4_write_dquot,
1077         .acquire_dquot  = ext4_acquire_dquot,
1078         .release_dquot  = ext4_release_dquot,
1079         .mark_dirty     = ext4_mark_dquot_dirty,
1080         .write_info     = ext4_write_info,
1081         .alloc_dquot    = dquot_alloc,
1082         .destroy_dquot  = dquot_destroy,
1083 };
1084
1085 static const struct quotactl_ops ext4_qctl_operations = {
1086         .quota_on       = ext4_quota_on,
1087         .quota_off      = vfs_quota_off,
1088         .quota_sync     = vfs_quota_sync,
1089         .get_info       = vfs_get_dqinfo,
1090         .set_info       = vfs_set_dqinfo,
1091         .get_dqblk      = vfs_get_dqblk,
1092         .set_dqblk      = vfs_set_dqblk
1093 };
1094 #endif
1095
1096 static const struct super_operations ext4_sops = {
1097         .alloc_inode    = ext4_alloc_inode,
1098         .destroy_inode  = ext4_destroy_inode,
1099         .write_inode    = ext4_write_inode,
1100         .dirty_inode    = ext4_dirty_inode,
1101         .delete_inode   = ext4_delete_inode,
1102         .put_super      = ext4_put_super,
1103         .sync_fs        = ext4_sync_fs,
1104         .freeze_fs      = ext4_freeze,
1105         .unfreeze_fs    = ext4_unfreeze,
1106         .statfs         = ext4_statfs,
1107         .remount_fs     = ext4_remount,
1108         .clear_inode    = ext4_clear_inode,
1109         .show_options   = ext4_show_options,
1110 #ifdef CONFIG_QUOTA
1111         .quota_read     = ext4_quota_read,
1112         .quota_write    = ext4_quota_write,
1113 #endif
1114         .bdev_try_to_free_page = bdev_try_to_free_page,
1115 };
1116
1117 static const struct super_operations ext4_nojournal_sops = {
1118         .alloc_inode    = ext4_alloc_inode,
1119         .destroy_inode  = ext4_destroy_inode,
1120         .write_inode    = ext4_write_inode,
1121         .dirty_inode    = ext4_dirty_inode,
1122         .delete_inode   = ext4_delete_inode,
1123         .write_super    = ext4_write_super,
1124         .put_super      = ext4_put_super,
1125         .statfs         = ext4_statfs,
1126         .remount_fs     = ext4_remount,
1127         .clear_inode    = ext4_clear_inode,
1128         .show_options   = ext4_show_options,
1129 #ifdef CONFIG_QUOTA
1130         .quota_read     = ext4_quota_read,
1131         .quota_write    = ext4_quota_write,
1132 #endif
1133         .bdev_try_to_free_page = bdev_try_to_free_page,
1134 };
1135
1136 static const struct export_operations ext4_export_ops = {
1137         .fh_to_dentry = ext4_fh_to_dentry,
1138         .fh_to_parent = ext4_fh_to_parent,
1139         .get_parent = ext4_get_parent,
1140 };
1141
1142 enum {
1143         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1144         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1145         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1146         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1147         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1148         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1149         Opt_journal_update, Opt_journal_dev,
1150         Opt_journal_checksum, Opt_journal_async_commit,
1151         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1152         Opt_data_err_abort, Opt_data_err_ignore,
1153         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1154         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1155         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1156         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1157         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1158         Opt_block_validity, Opt_noblock_validity,
1159         Opt_inode_readahead_blks, Opt_journal_ioprio,
1160         Opt_dioread_nolock, Opt_dioread_lock,
1161         Opt_discard, Opt_nodiscard,
1162 };
1163
1164 static const match_table_t tokens = {
1165         {Opt_bsd_df, "bsddf"},
1166         {Opt_minix_df, "minixdf"},
1167         {Opt_grpid, "grpid"},
1168         {Opt_grpid, "bsdgroups"},
1169         {Opt_nogrpid, "nogrpid"},
1170         {Opt_nogrpid, "sysvgroups"},
1171         {Opt_resgid, "resgid=%u"},
1172         {Opt_resuid, "resuid=%u"},
1173         {Opt_sb, "sb=%u"},
1174         {Opt_err_cont, "errors=continue"},
1175         {Opt_err_panic, "errors=panic"},
1176         {Opt_err_ro, "errors=remount-ro"},
1177         {Opt_nouid32, "nouid32"},
1178         {Opt_debug, "debug"},
1179         {Opt_oldalloc, "oldalloc"},
1180         {Opt_orlov, "orlov"},
1181         {Opt_user_xattr, "user_xattr"},
1182         {Opt_nouser_xattr, "nouser_xattr"},
1183         {Opt_acl, "acl"},
1184         {Opt_noacl, "noacl"},
1185         {Opt_noload, "noload"},
1186         {Opt_noload, "norecovery"},
1187         {Opt_nobh, "nobh"},
1188         {Opt_bh, "bh"},
1189         {Opt_commit, "commit=%u"},
1190         {Opt_min_batch_time, "min_batch_time=%u"},
1191         {Opt_max_batch_time, "max_batch_time=%u"},
1192         {Opt_journal_update, "journal=update"},
1193         {Opt_journal_dev, "journal_dev=%u"},
1194         {Opt_journal_checksum, "journal_checksum"},
1195         {Opt_journal_async_commit, "journal_async_commit"},
1196         {Opt_abort, "abort"},
1197         {Opt_data_journal, "data=journal"},
1198         {Opt_data_ordered, "data=ordered"},
1199         {Opt_data_writeback, "data=writeback"},
1200         {Opt_data_err_abort, "data_err=abort"},
1201         {Opt_data_err_ignore, "data_err=ignore"},
1202         {Opt_offusrjquota, "usrjquota="},
1203         {Opt_usrjquota, "usrjquota=%s"},
1204         {Opt_offgrpjquota, "grpjquota="},
1205         {Opt_grpjquota, "grpjquota=%s"},
1206         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1207         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1208         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1209         {Opt_grpquota, "grpquota"},
1210         {Opt_noquota, "noquota"},
1211         {Opt_quota, "quota"},
1212         {Opt_usrquota, "usrquota"},
1213         {Opt_barrier, "barrier=%u"},
1214         {Opt_barrier, "barrier"},
1215         {Opt_nobarrier, "nobarrier"},
1216         {Opt_i_version, "i_version"},
1217         {Opt_stripe, "stripe=%u"},
1218         {Opt_resize, "resize"},
1219         {Opt_delalloc, "delalloc"},
1220         {Opt_nodelalloc, "nodelalloc"},
1221         {Opt_block_validity, "block_validity"},
1222         {Opt_noblock_validity, "noblock_validity"},
1223         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1224         {Opt_journal_ioprio, "journal_ioprio=%u"},
1225         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1226         {Opt_auto_da_alloc, "auto_da_alloc"},
1227         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1228         {Opt_dioread_nolock, "dioread_nolock"},
1229         {Opt_dioread_lock, "dioread_lock"},
1230         {Opt_discard, "discard"},
1231         {Opt_nodiscard, "nodiscard"},
1232         {Opt_err, NULL},
1233 };
1234
1235 static ext4_fsblk_t get_sb_block(void **data)
1236 {
1237         ext4_fsblk_t    sb_block;
1238         char            *options = (char *) *data;
1239
1240         if (!options || strncmp(options, "sb=", 3) != 0)
1241                 return 1;       /* Default location */
1242
1243         options += 3;
1244         /* TODO: use simple_strtoll with >32bit ext4 */
1245         sb_block = simple_strtoul(options, &options, 0);
1246         if (*options && *options != ',') {
1247                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1248                        (char *) *data);
1249                 return 1;
1250         }
1251         if (*options == ',')
1252                 options++;
1253         *data = (void *) options;
1254
1255         return sb_block;
1256 }
1257
1258 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1259 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1260         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1261
1262 #ifdef CONFIG_QUOTA
1263 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1264 {
1265         struct ext4_sb_info *sbi = EXT4_SB(sb);
1266         char *qname;
1267
1268         if (sb_any_quota_loaded(sb) &&
1269                 !sbi->s_qf_names[qtype]) {
1270                 ext4_msg(sb, KERN_ERR,
1271                         "Cannot change journaled "
1272                         "quota options when quota turned on");
1273                 return 0;
1274         }
1275         qname = match_strdup(args);
1276         if (!qname) {
1277                 ext4_msg(sb, KERN_ERR,
1278                         "Not enough memory for storing quotafile name");
1279                 return 0;
1280         }
1281         if (sbi->s_qf_names[qtype] &&
1282                 strcmp(sbi->s_qf_names[qtype], qname)) {
1283                 ext4_msg(sb, KERN_ERR,
1284                         "%s quota file already specified", QTYPE2NAME(qtype));
1285                 kfree(qname);
1286                 return 0;
1287         }
1288         sbi->s_qf_names[qtype] = qname;
1289         if (strchr(sbi->s_qf_names[qtype], '/')) {
1290                 ext4_msg(sb, KERN_ERR,
1291                         "quotafile must be on filesystem root");
1292                 kfree(sbi->s_qf_names[qtype]);
1293                 sbi->s_qf_names[qtype] = NULL;
1294                 return 0;
1295         }
1296         set_opt(sbi->s_mount_opt, QUOTA);
1297         return 1;
1298 }
1299
1300 static int clear_qf_name(struct super_block *sb, int qtype)
1301 {
1302
1303         struct ext4_sb_info *sbi = EXT4_SB(sb);
1304
1305         if (sb_any_quota_loaded(sb) &&
1306                 sbi->s_qf_names[qtype]) {
1307                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1308                         " when quota turned on");
1309                 return 0;
1310         }
1311         /*
1312          * The space will be released later when all options are confirmed
1313          * to be correct
1314          */
1315         sbi->s_qf_names[qtype] = NULL;
1316         return 1;
1317 }
1318 #endif
1319
1320 static int parse_options(char *options, struct super_block *sb,
1321                          unsigned long *journal_devnum,
1322                          unsigned int *journal_ioprio,
1323                          ext4_fsblk_t *n_blocks_count, int is_remount)
1324 {
1325         struct ext4_sb_info *sbi = EXT4_SB(sb);
1326         char *p;
1327         substring_t args[MAX_OPT_ARGS];
1328         int data_opt = 0;
1329         int option;
1330 #ifdef CONFIG_QUOTA
1331         int qfmt;
1332 #endif
1333
1334         if (!options)
1335                 return 1;
1336
1337         while ((p = strsep(&options, ",")) != NULL) {
1338                 int token;
1339                 if (!*p)
1340                         continue;
1341
1342                 /*
1343                  * Initialize args struct so we know whether arg was
1344                  * found; some options take optional arguments.
1345                  */
1346                 args[0].to = args[0].from = 0;
1347                 token = match_token(p, tokens, args);
1348                 switch (token) {
1349                 case Opt_bsd_df:
1350                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1351                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1352                         break;
1353                 case Opt_minix_df:
1354                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1355                         set_opt(sbi->s_mount_opt, MINIX_DF);
1356
1357                         break;
1358                 case Opt_grpid:
1359                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1360                         set_opt(sbi->s_mount_opt, GRPID);
1361
1362                         break;
1363                 case Opt_nogrpid:
1364                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1365                         clear_opt(sbi->s_mount_opt, GRPID);
1366
1367                         break;
1368                 case Opt_resuid:
1369                         if (match_int(&args[0], &option))
1370                                 return 0;
1371                         sbi->s_resuid = option;
1372                         break;
1373                 case Opt_resgid:
1374                         if (match_int(&args[0], &option))
1375                                 return 0;
1376                         sbi->s_resgid = option;
1377                         break;
1378                 case Opt_sb:
1379                         /* handled by get_sb_block() instead of here */
1380                         /* *sb_block = match_int(&args[0]); */
1381                         break;
1382                 case Opt_err_panic:
1383                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1384                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1385                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1386                         break;
1387                 case Opt_err_ro:
1388                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1389                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1390                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1391                         break;
1392                 case Opt_err_cont:
1393                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1394                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1395                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1396                         break;
1397                 case Opt_nouid32:
1398                         set_opt(sbi->s_mount_opt, NO_UID32);
1399                         break;
1400                 case Opt_debug:
1401                         set_opt(sbi->s_mount_opt, DEBUG);
1402                         break;
1403                 case Opt_oldalloc:
1404                         set_opt(sbi->s_mount_opt, OLDALLOC);
1405                         break;
1406                 case Opt_orlov:
1407                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1408                         break;
1409 #ifdef CONFIG_EXT4_FS_XATTR
1410                 case Opt_user_xattr:
1411                         set_opt(sbi->s_mount_opt, XATTR_USER);
1412                         break;
1413                 case Opt_nouser_xattr:
1414                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1415                         break;
1416 #else
1417                 case Opt_user_xattr:
1418                 case Opt_nouser_xattr:
1419                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1420                         break;
1421 #endif
1422 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1423                 case Opt_acl:
1424                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1425                         break;
1426                 case Opt_noacl:
1427                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1428                         break;
1429 #else
1430                 case Opt_acl:
1431                 case Opt_noacl:
1432                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1433                         break;
1434 #endif
1435                 case Opt_journal_update:
1436                         /* @@@ FIXME */
1437                         /* Eventually we will want to be able to create
1438                            a journal file here.  For now, only allow the
1439                            user to specify an existing inode to be the
1440                            journal file. */
1441                         if (is_remount) {
1442                                 ext4_msg(sb, KERN_ERR,
1443                                          "Cannot specify journal on remount");
1444                                 return 0;
1445                         }
1446                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1447                         break;
1448                 case Opt_journal_dev:
1449                         if (is_remount) {
1450                                 ext4_msg(sb, KERN_ERR,
1451                                         "Cannot specify journal on remount");
1452                                 return 0;
1453                         }
1454                         if (match_int(&args[0], &option))
1455                                 return 0;
1456                         *journal_devnum = option;
1457                         break;
1458                 case Opt_journal_checksum:
1459                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1460                         break;
1461                 case Opt_journal_async_commit:
1462                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1463                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1464                         break;
1465                 case Opt_noload:
1466                         set_opt(sbi->s_mount_opt, NOLOAD);
1467                         break;
1468                 case Opt_commit:
1469                         if (match_int(&args[0], &option))
1470                                 return 0;
1471                         if (option < 0)
1472                                 return 0;
1473                         if (option == 0)
1474                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1475                         sbi->s_commit_interval = HZ * option;
1476                         break;
1477                 case Opt_max_batch_time:
1478                         if (match_int(&args[0], &option))
1479                                 return 0;
1480                         if (option < 0)
1481                                 return 0;
1482                         if (option == 0)
1483                                 option = EXT4_DEF_MAX_BATCH_TIME;
1484                         sbi->s_max_batch_time = option;
1485                         break;
1486                 case Opt_min_batch_time:
1487                         if (match_int(&args[0], &option))
1488                                 return 0;
1489                         if (option < 0)
1490                                 return 0;
1491                         sbi->s_min_batch_time = option;
1492                         break;
1493                 case Opt_data_journal:
1494                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1495                         goto datacheck;
1496                 case Opt_data_ordered:
1497                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1498                         goto datacheck;
1499                 case Opt_data_writeback:
1500                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1501                 datacheck:
1502                         if (is_remount) {
1503                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1504                                         ext4_msg(sb, KERN_ERR,
1505                                                 "Cannot change data mode on remount");
1506                                         return 0;
1507                                 }
1508                         } else {
1509                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1510                                 sbi->s_mount_opt |= data_opt;
1511                         }
1512                         break;
1513                 case Opt_data_err_abort:
1514                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1515                         break;
1516                 case Opt_data_err_ignore:
1517                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1518                         break;
1519 #ifdef CONFIG_QUOTA
1520                 case Opt_usrjquota:
1521                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1522                                 return 0;
1523                         break;
1524                 case Opt_grpjquota:
1525                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1526                                 return 0;
1527                         break;
1528                 case Opt_offusrjquota:
1529                         if (!clear_qf_name(sb, USRQUOTA))
1530                                 return 0;
1531                         break;
1532                 case Opt_offgrpjquota:
1533                         if (!clear_qf_name(sb, GRPQUOTA))
1534                                 return 0;
1535                         break;
1536
1537                 case Opt_jqfmt_vfsold:
1538                         qfmt = QFMT_VFS_OLD;
1539                         goto set_qf_format;
1540                 case Opt_jqfmt_vfsv0:
1541                         qfmt = QFMT_VFS_V0;
1542                         goto set_qf_format;
1543                 case Opt_jqfmt_vfsv1:
1544                         qfmt = QFMT_VFS_V1;
1545 set_qf_format:
1546                         if (sb_any_quota_loaded(sb) &&
1547                             sbi->s_jquota_fmt != qfmt) {
1548                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1549                                         "journaled quota options when "
1550                                         "quota turned on");
1551                                 return 0;
1552                         }
1553                         sbi->s_jquota_fmt = qfmt;
1554                         break;
1555                 case Opt_quota:
1556                 case Opt_usrquota:
1557                         set_opt(sbi->s_mount_opt, QUOTA);
1558                         set_opt(sbi->s_mount_opt, USRQUOTA);
1559                         break;
1560                 case Opt_grpquota:
1561                         set_opt(sbi->s_mount_opt, QUOTA);
1562                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1563                         break;
1564                 case Opt_noquota:
1565                         if (sb_any_quota_loaded(sb)) {
1566                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1567                                         "options when quota turned on");
1568                                 return 0;
1569                         }
1570                         clear_opt(sbi->s_mount_opt, QUOTA);
1571                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1572                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1573                         break;
1574 #else
1575                 case Opt_quota:
1576                 case Opt_usrquota:
1577                 case Opt_grpquota:
1578                         ext4_msg(sb, KERN_ERR,
1579                                 "quota options not supported");
1580                         break;
1581                 case Opt_usrjquota:
1582                 case Opt_grpjquota:
1583                 case Opt_offusrjquota:
1584                 case Opt_offgrpjquota:
1585                 case Opt_jqfmt_vfsold:
1586                 case Opt_jqfmt_vfsv0:
1587                 case Opt_jqfmt_vfsv1:
1588                         ext4_msg(sb, KERN_ERR,
1589                                 "journaled quota options not supported");
1590                         break;
1591                 case Opt_noquota:
1592                         break;
1593 #endif
1594                 case Opt_abort:
1595                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1596                         break;
1597                 case Opt_nobarrier:
1598                         clear_opt(sbi->s_mount_opt, BARRIER);
1599                         break;
1600                 case Opt_barrier:
1601                         if (args[0].from) {
1602                                 if (match_int(&args[0], &option))
1603                                         return 0;
1604                         } else
1605                                 option = 1;     /* No argument, default to 1 */
1606                         if (option)
1607                                 set_opt(sbi->s_mount_opt, BARRIER);
1608                         else
1609                                 clear_opt(sbi->s_mount_opt, BARRIER);
1610                         break;
1611                 case Opt_ignore:
1612                         break;
1613                 case Opt_resize:
1614                         if (!is_remount) {
1615                                 ext4_msg(sb, KERN_ERR,
1616                                         "resize option only available "
1617                                         "for remount");
1618                                 return 0;
1619                         }
1620                         if (match_int(&args[0], &option) != 0)
1621                                 return 0;
1622                         *n_blocks_count = option;
1623                         break;
1624                 case Opt_nobh:
1625                         set_opt(sbi->s_mount_opt, NOBH);
1626                         break;
1627                 case Opt_bh:
1628                         clear_opt(sbi->s_mount_opt, NOBH);
1629                         break;
1630                 case Opt_i_version:
1631                         set_opt(sbi->s_mount_opt, I_VERSION);
1632                         sb->s_flags |= MS_I_VERSION;
1633                         break;
1634                 case Opt_nodelalloc:
1635                         clear_opt(sbi->s_mount_opt, DELALLOC);
1636                         break;
1637                 case Opt_stripe:
1638                         if (match_int(&args[0], &option))
1639                                 return 0;
1640                         if (option < 0)
1641                                 return 0;
1642                         sbi->s_stripe = option;
1643                         break;
1644                 case Opt_delalloc:
1645                         set_opt(sbi->s_mount_opt, DELALLOC);
1646                         break;
1647                 case Opt_block_validity:
1648                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1649                         break;
1650                 case Opt_noblock_validity:
1651                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1652                         break;
1653                 case Opt_inode_readahead_blks:
1654                         if (match_int(&args[0], &option))
1655                                 return 0;
1656                         if (option < 0 || option > (1 << 30))
1657                                 return 0;
1658                         if (!is_power_of_2(option)) {
1659                                 ext4_msg(sb, KERN_ERR,
1660                                          "EXT4-fs: inode_readahead_blks"
1661                                          " must be a power of 2");
1662                                 return 0;
1663                         }
1664                         sbi->s_inode_readahead_blks = option;
1665                         break;
1666                 case Opt_journal_ioprio:
1667                         if (match_int(&args[0], &option))
1668                                 return 0;
1669                         if (option < 0 || option > 7)
1670                                 break;
1671                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1672                                                             option);
1673                         break;
1674                 case Opt_noauto_da_alloc:
1675                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1676                         break;
1677                 case Opt_auto_da_alloc:
1678                         if (args[0].from) {
1679                                 if (match_int(&args[0], &option))
1680                                         return 0;
1681                         } else
1682                                 option = 1;     /* No argument, default to 1 */
1683                         if (option)
1684                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1685                         else
1686                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1687                         break;
1688                 case Opt_discard:
1689                         set_opt(sbi->s_mount_opt, DISCARD);
1690                         break;
1691                 case Opt_nodiscard:
1692                         clear_opt(sbi->s_mount_opt, DISCARD);
1693                         break;
1694                 case Opt_dioread_nolock:
1695                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1696                         break;
1697                 case Opt_dioread_lock:
1698                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1699                         break;
1700                 default:
1701                         ext4_msg(sb, KERN_ERR,
1702                                "Unrecognized mount option \"%s\" "
1703                                "or missing value", p);
1704                         return 0;
1705                 }
1706         }
1707 #ifdef CONFIG_QUOTA
1708         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1709                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1710                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1711
1712                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1713                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1714
1715                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1716                         ext4_msg(sb, KERN_ERR, "old and new quota "
1717                                         "format mixing");
1718                         return 0;
1719                 }
1720
1721                 if (!sbi->s_jquota_fmt) {
1722                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1723                                         "not specified");
1724                         return 0;
1725                 }
1726         } else {
1727                 if (sbi->s_jquota_fmt) {
1728                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1729                                         "specified with no journaling "
1730                                         "enabled");
1731                         return 0;
1732                 }
1733         }
1734 #endif
1735         return 1;
1736 }
1737
1738 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1739                             int read_only)
1740 {
1741         struct ext4_sb_info *sbi = EXT4_SB(sb);
1742         int res = 0;
1743
1744         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1745                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1746                          "forcing read-only mode");
1747                 res = MS_RDONLY;
1748         }
1749         if (read_only)
1750                 return res;
1751         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1752                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1753                          "running e2fsck is recommended");
1754         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1755                 ext4_msg(sb, KERN_WARNING,
1756                          "warning: mounting fs with errors, "
1757                          "running e2fsck is recommended");
1758         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1759                  le16_to_cpu(es->s_mnt_count) >=
1760                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1761                 ext4_msg(sb, KERN_WARNING,
1762                          "warning: maximal mount count reached, "
1763                          "running e2fsck is recommended");
1764         else if (le32_to_cpu(es->s_checkinterval) &&
1765                 (le32_to_cpu(es->s_lastcheck) +
1766                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1767                 ext4_msg(sb, KERN_WARNING,
1768                          "warning: checktime reached, "
1769                          "running e2fsck is recommended");
1770         if (!sbi->s_journal)
1771                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1772         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1773                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1774         le16_add_cpu(&es->s_mnt_count, 1);
1775         es->s_mtime = cpu_to_le32(get_seconds());
1776         ext4_update_dynamic_rev(sb);
1777         if (sbi->s_journal)
1778                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1779
1780         ext4_commit_super(sb, 1);
1781         if (test_opt(sb, DEBUG))
1782                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1783                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1784                         sb->s_blocksize,
1785                         sbi->s_groups_count,
1786                         EXT4_BLOCKS_PER_GROUP(sb),
1787                         EXT4_INODES_PER_GROUP(sb),
1788                         sbi->s_mount_opt);
1789
1790         return res;
1791 }
1792
1793 static int ext4_fill_flex_info(struct super_block *sb)
1794 {
1795         struct ext4_sb_info *sbi = EXT4_SB(sb);
1796         struct ext4_group_desc *gdp = NULL;
1797         ext4_group_t flex_group_count;
1798         ext4_group_t flex_group;
1799         int groups_per_flex = 0;
1800         size_t size;
1801         int i;
1802
1803         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1804         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1805
1806         if (groups_per_flex < 2) {
1807                 sbi->s_log_groups_per_flex = 0;
1808                 return 1;
1809         }
1810
1811         /* We allocate both existing and potentially added groups */
1812         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1813                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1814                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1815         size = flex_group_count * sizeof(struct flex_groups);
1816         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1817         if (sbi->s_flex_groups == NULL) {
1818                 sbi->s_flex_groups = vmalloc(size);
1819                 if (sbi->s_flex_groups)
1820                         memset(sbi->s_flex_groups, 0, size);
1821         }
1822         if (sbi->s_flex_groups == NULL) {
1823                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1824                                 "%u flex groups", flex_group_count);
1825                 goto failed;
1826         }
1827
1828         for (i = 0; i < sbi->s_groups_count; i++) {
1829                 gdp = ext4_get_group_desc(sb, i, NULL);
1830
1831                 flex_group = ext4_flex_group(sbi, i);
1832                 atomic_add(ext4_free_inodes_count(sb, gdp),
1833                            &sbi->s_flex_groups[flex_group].free_inodes);
1834                 atomic_add(ext4_free_blks_count(sb, gdp),
1835                            &sbi->s_flex_groups[flex_group].free_blocks);
1836                 atomic_add(ext4_used_dirs_count(sb, gdp),
1837                            &sbi->s_flex_groups[flex_group].used_dirs);
1838         }
1839
1840         return 1;
1841 failed:
1842         return 0;
1843 }
1844
1845 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1846                             struct ext4_group_desc *gdp)
1847 {
1848         __u16 crc = 0;
1849
1850         if (sbi->s_es->s_feature_ro_compat &
1851             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1852                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1853                 __le32 le_group = cpu_to_le32(block_group);
1854
1855                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1856                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1857                 crc = crc16(crc, (__u8 *)gdp, offset);
1858                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1859                 /* for checksum of struct ext4_group_desc do the rest...*/
1860                 if ((sbi->s_es->s_feature_incompat &
1861                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1862                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1863                         crc = crc16(crc, (__u8 *)gdp + offset,
1864                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1865                                         offset);
1866         }
1867
1868         return cpu_to_le16(crc);
1869 }
1870
1871 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1872                                 struct ext4_group_desc *gdp)
1873 {
1874         if ((sbi->s_es->s_feature_ro_compat &
1875              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1876             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1877                 return 0;
1878
1879         return 1;
1880 }
1881
1882 /* Called at mount-time, super-block is locked */
1883 static int ext4_check_descriptors(struct super_block *sb)
1884 {
1885         struct ext4_sb_info *sbi = EXT4_SB(sb);
1886         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1887         ext4_fsblk_t last_block;
1888         ext4_fsblk_t block_bitmap;
1889         ext4_fsblk_t inode_bitmap;
1890         ext4_fsblk_t inode_table;
1891         int flexbg_flag = 0;
1892         ext4_group_t i;
1893
1894         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1895                 flexbg_flag = 1;
1896
1897         ext4_debug("Checking group descriptors");
1898
1899         for (i = 0; i < sbi->s_groups_count; i++) {
1900                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1901
1902                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1903                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1904                 else
1905                         last_block = first_block +
1906                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1907
1908                 block_bitmap = ext4_block_bitmap(sb, gdp);
1909                 if (block_bitmap < first_block || block_bitmap > last_block) {
1910                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1911                                "Block bitmap for group %u not in group "
1912                                "(block %llu)!", i, block_bitmap);
1913                         return 0;
1914                 }
1915                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1916                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1917                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1918                                "Inode bitmap for group %u not in group "
1919                                "(block %llu)!", i, inode_bitmap);
1920                         return 0;
1921                 }
1922                 inode_table = ext4_inode_table(sb, gdp);
1923                 if (inode_table < first_block ||
1924                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1925                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1926                                "Inode table for group %u not in group "
1927                                "(block %llu)!", i, inode_table);
1928                         return 0;
1929                 }
1930                 ext4_lock_group(sb, i);
1931                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1932                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1933                                  "Checksum for group %u failed (%u!=%u)",
1934                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1935                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1936                         if (!(sb->s_flags & MS_RDONLY)) {
1937                                 ext4_unlock_group(sb, i);
1938                                 return 0;
1939                         }
1940                 }
1941                 ext4_unlock_group(sb, i);
1942                 if (!flexbg_flag)
1943                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1944         }
1945
1946         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1947         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1948         return 1;
1949 }
1950
1951 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1952  * the superblock) which were deleted from all directories, but held open by
1953  * a process at the time of a crash.  We walk the list and try to delete these
1954  * inodes at recovery time (only with a read-write filesystem).
1955  *
1956  * In order to keep the orphan inode chain consistent during traversal (in
1957  * case of crash during recovery), we link each inode into the superblock
1958  * orphan list_head and handle it the same way as an inode deletion during
1959  * normal operation (which journals the operations for us).
1960  *
1961  * We only do an iget() and an iput() on each inode, which is very safe if we
1962  * accidentally point at an in-use or already deleted inode.  The worst that
1963  * can happen in this case is that we get a "bit already cleared" message from
1964  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1965  * e2fsck was run on this filesystem, and it must have already done the orphan
1966  * inode cleanup for us, so we can safely abort without any further action.
1967  */
1968 static void ext4_orphan_cleanup(struct super_block *sb,
1969                                 struct ext4_super_block *es)
1970 {
1971         unsigned int s_flags = sb->s_flags;
1972         int nr_orphans = 0, nr_truncates = 0;
1973 #ifdef CONFIG_QUOTA
1974         int i;
1975 #endif
1976         if (!es->s_last_orphan) {
1977                 jbd_debug(4, "no orphan inodes to clean up\n");
1978                 return;
1979         }
1980
1981         if (bdev_read_only(sb->s_bdev)) {
1982                 ext4_msg(sb, KERN_ERR, "write access "
1983                         "unavailable, skipping orphan cleanup");
1984                 return;
1985         }
1986
1987         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1988                 if (es->s_last_orphan)
1989                         jbd_debug(1, "Errors on filesystem, "
1990                                   "clearing orphan list.\n");
1991                 es->s_last_orphan = 0;
1992                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1993                 return;
1994         }
1995
1996         if (s_flags & MS_RDONLY) {
1997                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1998                 sb->s_flags &= ~MS_RDONLY;
1999         }
2000 #ifdef CONFIG_QUOTA
2001         /* Needed for iput() to work correctly and not trash data */
2002         sb->s_flags |= MS_ACTIVE;
2003         /* Turn on quotas so that they are updated correctly */
2004         for (i = 0; i < MAXQUOTAS; i++) {
2005                 if (EXT4_SB(sb)->s_qf_names[i]) {
2006                         int ret = ext4_quota_on_mount(sb, i);
2007                         if (ret < 0)
2008                                 ext4_msg(sb, KERN_ERR,
2009                                         "Cannot turn on journaled "
2010                                         "quota: error %d", ret);
2011                 }
2012         }
2013 #endif
2014
2015         while (es->s_last_orphan) {
2016                 struct inode *inode;
2017
2018                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2019                 if (IS_ERR(inode)) {
2020                         es->s_last_orphan = 0;
2021                         break;
2022                 }
2023
2024                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2025                 dquot_initialize(inode);
2026                 if (inode->i_nlink) {
2027                         ext4_msg(sb, KERN_DEBUG,
2028                                 "%s: truncating inode %lu to %lld bytes",
2029                                 __func__, inode->i_ino, inode->i_size);
2030                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2031                                   inode->i_ino, inode->i_size);
2032                         ext4_truncate(inode);
2033                         nr_truncates++;
2034                 } else {
2035                         ext4_msg(sb, KERN_DEBUG,
2036                                 "%s: deleting unreferenced inode %lu",
2037                                 __func__, inode->i_ino);
2038                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2039                                   inode->i_ino);
2040                         nr_orphans++;
2041                 }
2042                 iput(inode);  /* The delete magic happens here! */
2043         }
2044
2045 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2046
2047         if (nr_orphans)
2048                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2049                        PLURAL(nr_orphans));
2050         if (nr_truncates)
2051                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2052                        PLURAL(nr_truncates));
2053 #ifdef CONFIG_QUOTA
2054         /* Turn quotas off */
2055         for (i = 0; i < MAXQUOTAS; i++) {
2056                 if (sb_dqopt(sb)->files[i])
2057                         vfs_quota_off(sb, i, 0);
2058         }
2059 #endif
2060         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2061 }
2062
2063 /*
2064  * Maximal extent format file size.
2065  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2066  * extent format containers, within a sector_t, and within i_blocks
2067  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2068  * so that won't be a limiting factor.
2069  *
2070  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2071  */
2072 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2073 {
2074         loff_t res;
2075         loff_t upper_limit = MAX_LFS_FILESIZE;
2076
2077         /* small i_blocks in vfs inode? */
2078         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2079                 /*
2080                  * CONFIG_LBDAF is not enabled implies the inode
2081                  * i_block represent total blocks in 512 bytes
2082                  * 32 == size of vfs inode i_blocks * 8
2083                  */
2084                 upper_limit = (1LL << 32) - 1;
2085
2086                 /* total blocks in file system block size */
2087                 upper_limit >>= (blkbits - 9);
2088                 upper_limit <<= blkbits;
2089         }
2090
2091         /* 32-bit extent-start container, ee_block */
2092         res = 1LL << 32;
2093         res <<= blkbits;
2094         res -= 1;
2095
2096         /* Sanity check against vm- & vfs- imposed limits */
2097         if (res > upper_limit)
2098                 res = upper_limit;
2099
2100         return res;
2101 }
2102
2103 /*
2104  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2105  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2106  * We need to be 1 filesystem block less than the 2^48 sector limit.
2107  */
2108 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2109 {
2110         loff_t res = EXT4_NDIR_BLOCKS;
2111         int meta_blocks;
2112         loff_t upper_limit;
2113         /* This is calculated to be the largest file size for a dense, block
2114          * mapped file such that the file's total number of 512-byte sectors,
2115          * including data and all indirect blocks, does not exceed (2^48 - 1).
2116          *
2117          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2118          * number of 512-byte sectors of the file.
2119          */
2120
2121         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2122                 /*
2123                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2124                  * the inode i_block field represents total file blocks in
2125                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2126                  */
2127                 upper_limit = (1LL << 32) - 1;
2128
2129                 /* total blocks in file system block size */
2130                 upper_limit >>= (bits - 9);
2131
2132         } else {
2133                 /*
2134                  * We use 48 bit ext4_inode i_blocks
2135                  * With EXT4_HUGE_FILE_FL set the i_blocks
2136                  * represent total number of blocks in
2137                  * file system block size
2138                  */
2139                 upper_limit = (1LL << 48) - 1;
2140
2141         }
2142
2143         /* indirect blocks */
2144         meta_blocks = 1;
2145         /* double indirect blocks */
2146         meta_blocks += 1 + (1LL << (bits-2));
2147         /* tripple indirect blocks */
2148         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2149
2150         upper_limit -= meta_blocks;
2151         upper_limit <<= bits;
2152
2153         res += 1LL << (bits-2);
2154         res += 1LL << (2*(bits-2));
2155         res += 1LL << (3*(bits-2));
2156         res <<= bits;
2157         if (res > upper_limit)
2158                 res = upper_limit;
2159
2160         if (res > MAX_LFS_FILESIZE)
2161                 res = MAX_LFS_FILESIZE;
2162
2163         return res;
2164 }
2165
2166 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2167                                    ext4_fsblk_t logical_sb_block, int nr)
2168 {
2169         struct ext4_sb_info *sbi = EXT4_SB(sb);
2170         ext4_group_t bg, first_meta_bg;
2171         int has_super = 0;
2172
2173         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2174
2175         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2176             nr < first_meta_bg)
2177                 return logical_sb_block + nr + 1;
2178         bg = sbi->s_desc_per_block * nr;
2179         if (ext4_bg_has_super(sb, bg))
2180                 has_super = 1;
2181
2182         return (has_super + ext4_group_first_block_no(sb, bg));
2183 }
2184
2185 /**
2186  * ext4_get_stripe_size: Get the stripe size.
2187  * @sbi: In memory super block info
2188  *
2189  * If we have specified it via mount option, then
2190  * use the mount option value. If the value specified at mount time is
2191  * greater than the blocks per group use the super block value.
2192  * If the super block value is greater than blocks per group return 0.
2193  * Allocator needs it be less than blocks per group.
2194  *
2195  */
2196 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2197 {
2198         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2199         unsigned long stripe_width =
2200                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2201
2202         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2203                 return sbi->s_stripe;
2204
2205         if (stripe_width <= sbi->s_blocks_per_group)
2206                 return stripe_width;
2207
2208         if (stride <= sbi->s_blocks_per_group)
2209                 return stride;
2210
2211         return 0;
2212 }
2213
2214 /* sysfs supprt */
2215
2216 struct ext4_attr {
2217         struct attribute attr;
2218         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2219         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2220                          const char *, size_t);
2221         int offset;
2222 };
2223
2224 static int parse_strtoul(const char *buf,
2225                 unsigned long max, unsigned long *value)
2226 {
2227         char *endp;
2228
2229         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2230         endp = skip_spaces(endp);
2231         if (*endp || *value > max)
2232                 return -EINVAL;
2233
2234         return 0;
2235 }
2236
2237 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2238                                               struct ext4_sb_info *sbi,
2239                                               char *buf)
2240 {
2241         return snprintf(buf, PAGE_SIZE, "%llu\n",
2242                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2243 }
2244
2245 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2246                                          struct ext4_sb_info *sbi, char *buf)
2247 {
2248         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2249
2250         return snprintf(buf, PAGE_SIZE, "%lu\n",
2251                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2252                          sbi->s_sectors_written_start) >> 1);
2253 }
2254
2255 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2256                                           struct ext4_sb_info *sbi, char *buf)
2257 {
2258         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2259
2260         return snprintf(buf, PAGE_SIZE, "%llu\n",
2261                         (unsigned long long)(sbi->s_kbytes_written +
2262                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2263                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2264 }
2265
2266 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2267                                           struct ext4_sb_info *sbi,
2268                                           const char *buf, size_t count)
2269 {
2270         unsigned long t;
2271
2272         if (parse_strtoul(buf, 0x40000000, &t))
2273                 return -EINVAL;
2274
2275         if (!is_power_of_2(t))
2276                 return -EINVAL;
2277
2278         sbi->s_inode_readahead_blks = t;
2279         return count;
2280 }
2281
2282 static ssize_t sbi_ui_show(struct ext4_attr *a,
2283                            struct ext4_sb_info *sbi, char *buf)
2284 {
2285         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2286
2287         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2288 }
2289
2290 static ssize_t sbi_ui_store(struct ext4_attr *a,
2291                             struct ext4_sb_info *sbi,
2292                             const char *buf, size_t count)
2293 {
2294         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2295         unsigned long t;
2296
2297         if (parse_strtoul(buf, 0xffffffff, &t))
2298                 return -EINVAL;
2299         *ui = t;
2300         return count;
2301 }
2302
2303 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2304 static struct ext4_attr ext4_attr_##_name = {                   \
2305         .attr = {.name = __stringify(_name), .mode = _mode },   \
2306         .show   = _show,                                        \
2307         .store  = _store,                                       \
2308         .offset = offsetof(struct ext4_sb_info, _elname),       \
2309 }
2310 #define EXT4_ATTR(name, mode, show, store) \
2311 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2312
2313 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2314 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2315 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2316         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2317 #define ATTR_LIST(name) &ext4_attr_##name.attr
2318
2319 EXT4_RO_ATTR(delayed_allocation_blocks);
2320 EXT4_RO_ATTR(session_write_kbytes);
2321 EXT4_RO_ATTR(lifetime_write_kbytes);
2322 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2323                  inode_readahead_blks_store, s_inode_readahead_blks);
2324 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2325 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2326 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2327 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2328 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2329 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2330 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2331 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2332
2333 static struct attribute *ext4_attrs[] = {
2334         ATTR_LIST(delayed_allocation_blocks),
2335         ATTR_LIST(session_write_kbytes),
2336         ATTR_LIST(lifetime_write_kbytes),
2337         ATTR_LIST(inode_readahead_blks),
2338         ATTR_LIST(inode_goal),
2339         ATTR_LIST(mb_stats),
2340         ATTR_LIST(mb_max_to_scan),
2341         ATTR_LIST(mb_min_to_scan),
2342         ATTR_LIST(mb_order2_req),
2343         ATTR_LIST(mb_stream_req),
2344         ATTR_LIST(mb_group_prealloc),
2345         ATTR_LIST(max_writeback_mb_bump),
2346         NULL,
2347 };
2348
2349 static ssize_t ext4_attr_show(struct kobject *kobj,
2350                               struct attribute *attr, char *buf)
2351 {
2352         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2353                                                 s_kobj);
2354         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2355
2356         return a->show ? a->show(a, sbi, buf) : 0;
2357 }
2358
2359 static ssize_t ext4_attr_store(struct kobject *kobj,
2360                                struct attribute *attr,
2361                                const char *buf, size_t len)
2362 {
2363         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2364                                                 s_kobj);
2365         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2366
2367         return a->store ? a->store(a, sbi, buf, len) : 0;
2368 }
2369
2370 static void ext4_sb_release(struct kobject *kobj)
2371 {
2372         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2373                                                 s_kobj);
2374         complete(&sbi->s_kobj_unregister);
2375 }
2376
2377
2378 static const struct sysfs_ops ext4_attr_ops = {
2379         .show   = ext4_attr_show,
2380         .store  = ext4_attr_store,
2381 };
2382
2383 static struct kobj_type ext4_ktype = {
2384         .default_attrs  = ext4_attrs,
2385         .sysfs_ops      = &ext4_attr_ops,
2386         .release        = ext4_sb_release,
2387 };
2388
2389 /*
2390  * Check whether this filesystem can be mounted based on
2391  * the features present and the RDONLY/RDWR mount requested.
2392  * Returns 1 if this filesystem can be mounted as requested,
2393  * 0 if it cannot be.
2394  */
2395 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2396 {
2397         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2398                 ext4_msg(sb, KERN_ERR,
2399                         "Couldn't mount because of "
2400                         "unsupported optional features (%x)",
2401                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2402                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2403                 return 0;
2404         }
2405
2406         if (readonly)
2407                 return 1;
2408
2409         /* Check that feature set is OK for a read-write mount */
2410         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2411                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2412                          "unsupported optional features (%x)",
2413                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2414                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2415                 return 0;
2416         }
2417         /*
2418          * Large file size enabled file system can only be mounted
2419          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2420          */
2421         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2422                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2423                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2424                                  "cannot be mounted RDWR without "
2425                                  "CONFIG_LBDAF");
2426                         return 0;
2427                 }
2428         }
2429         return 1;
2430 }
2431
2432 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2433                                 __releases(kernel_lock)
2434                                 __acquires(kernel_lock)
2435 {
2436         char *orig_data = kstrdup(data, GFP_KERNEL);
2437         struct buffer_head *bh;
2438         struct ext4_super_block *es = NULL;
2439         struct ext4_sb_info *sbi;
2440         ext4_fsblk_t block;
2441         ext4_fsblk_t sb_block = get_sb_block(&data);
2442         ext4_fsblk_t logical_sb_block;
2443         unsigned long offset = 0;
2444         unsigned long journal_devnum = 0;
2445         unsigned long def_mount_opts;
2446         struct inode *root;
2447         char *cp;
2448         const char *descr;
2449         int ret = -EINVAL;
2450         int blocksize;
2451         unsigned int db_count;
2452         unsigned int i;
2453         int needs_recovery, has_huge_files;
2454         __u64 blocks_count;
2455         int err;
2456         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2457
2458         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2459         if (!sbi)
2460                 return -ENOMEM;
2461
2462         sbi->s_blockgroup_lock =
2463                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2464         if (!sbi->s_blockgroup_lock) {
2465                 kfree(sbi);
2466                 return -ENOMEM;
2467         }
2468         sb->s_fs_info = sbi;
2469         sbi->s_mount_opt = 0;
2470         sbi->s_resuid = EXT4_DEF_RESUID;
2471         sbi->s_resgid = EXT4_DEF_RESGID;
2472         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2473         sbi->s_sb_block = sb_block;
2474         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2475                                                       sectors[1]);
2476
2477         unlock_kernel();
2478
2479         /* Cleanup superblock name */
2480         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2481                 *cp = '!';
2482
2483         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2484         if (!blocksize) {
2485                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2486                 goto out_fail;
2487         }
2488
2489         /*
2490          * The ext4 superblock will not be buffer aligned for other than 1kB
2491          * block sizes.  We need to calculate the offset from buffer start.
2492          */
2493         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2494                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2495                 offset = do_div(logical_sb_block, blocksize);
2496         } else {
2497                 logical_sb_block = sb_block;
2498         }
2499
2500         if (!(bh = sb_bread(sb, logical_sb_block))) {
2501                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2502                 goto out_fail;
2503         }
2504         /*
2505          * Note: s_es must be initialized as soon as possible because
2506          *       some ext4 macro-instructions depend on its value
2507          */
2508         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2509         sbi->s_es = es;
2510         sb->s_magic = le16_to_cpu(es->s_magic);
2511         if (sb->s_magic != EXT4_SUPER_MAGIC)
2512                 goto cantfind_ext4;
2513         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2514
2515         /* Set defaults before we parse the mount options */
2516         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2517         if (def_mount_opts & EXT4_DEFM_DEBUG)
2518                 set_opt(sbi->s_mount_opt, DEBUG);
2519         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
2520                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
2521                         "2.6.38");
2522                 set_opt(sbi->s_mount_opt, GRPID);
2523         }
2524         if (def_mount_opts & EXT4_DEFM_UID16)
2525                 set_opt(sbi->s_mount_opt, NO_UID32);
2526 #ifdef CONFIG_EXT4_FS_XATTR
2527         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2528                 set_opt(sbi->s_mount_opt, XATTR_USER);
2529 #endif
2530 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2531         if (def_mount_opts & EXT4_DEFM_ACL)
2532                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2533 #endif
2534         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2535                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2536         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2537                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
2538         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2539                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2540
2541         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2542                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2543         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2544                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2545         else
2546                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2547
2548         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2549         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2550         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2551         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2552         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2553
2554         set_opt(sbi->s_mount_opt, BARRIER);
2555
2556         /*
2557          * enable delayed allocation by default
2558          * Use -o nodelalloc to turn it off
2559          */
2560         if (!IS_EXT3_SB(sb))
2561                 set_opt(sbi->s_mount_opt, DELALLOC);
2562
2563         if (!parse_options((char *) data, sb, &journal_devnum,
2564                            &journal_ioprio, NULL, 0))
2565                 goto failed_mount;
2566
2567         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2568                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2569
2570         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2571             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2572              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2573              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2574                 ext4_msg(sb, KERN_WARNING,
2575                        "feature flags set on rev 0 fs, "
2576                        "running e2fsck is recommended");
2577
2578         /*
2579          * Check feature flags regardless of the revision level, since we
2580          * previously didn't change the revision level when setting the flags,
2581          * so there is a chance incompat flags are set on a rev 0 filesystem.
2582          */
2583         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2584                 goto failed_mount;
2585
2586         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2587
2588         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2589             blocksize > EXT4_MAX_BLOCK_SIZE) {
2590                 ext4_msg(sb, KERN_ERR,
2591                        "Unsupported filesystem blocksize %d", blocksize);
2592                 goto failed_mount;
2593         }
2594
2595         if (sb->s_blocksize != blocksize) {
2596                 /* Validate the filesystem blocksize */
2597                 if (!sb_set_blocksize(sb, blocksize)) {
2598                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2599                                         blocksize);
2600                         goto failed_mount;
2601                 }
2602
2603                 brelse(bh);
2604                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2605                 offset = do_div(logical_sb_block, blocksize);
2606                 bh = sb_bread(sb, logical_sb_block);
2607                 if (!bh) {
2608                         ext4_msg(sb, KERN_ERR,
2609                                "Can't read superblock on 2nd try");
2610                         goto failed_mount;
2611                 }
2612                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2613                 sbi->s_es = es;
2614                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2615                         ext4_msg(sb, KERN_ERR,
2616                                "Magic mismatch, very weird!");
2617                         goto failed_mount;
2618                 }
2619         }
2620
2621         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2622                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2623         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2624                                                       has_huge_files);
2625         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2626
2627         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2628                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2629                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2630         } else {
2631                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2632                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2633                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2634                     (!is_power_of_2(sbi->s_inode_size)) ||
2635                     (sbi->s_inode_size > blocksize)) {
2636                         ext4_msg(sb, KERN_ERR,
2637                                "unsupported inode size: %d",
2638                                sbi->s_inode_size);
2639                         goto failed_mount;
2640                 }
2641                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2642                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2643         }
2644
2645         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2646         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2647                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2648                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2649                     !is_power_of_2(sbi->s_desc_size)) {
2650                         ext4_msg(sb, KERN_ERR,
2651                                "unsupported descriptor size %lu",
2652                                sbi->s_desc_size);
2653                         goto failed_mount;
2654                 }
2655         } else
2656                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2657
2658         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2659         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2660         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2661                 goto cantfind_ext4;
2662
2663         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2664         if (sbi->s_inodes_per_block == 0)
2665                 goto cantfind_ext4;
2666         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2667                                         sbi->s_inodes_per_block;
2668         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2669         sbi->s_sbh = bh;
2670         sbi->s_mount_state = le16_to_cpu(es->s_state);
2671         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2672         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2673
2674         for (i = 0; i < 4; i++)
2675                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2676         sbi->s_def_hash_version = es->s_def_hash_version;
2677         i = le32_to_cpu(es->s_flags);
2678         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2679                 sbi->s_hash_unsigned = 3;
2680         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2681 #ifdef __CHAR_UNSIGNED__
2682                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2683                 sbi->s_hash_unsigned = 3;
2684 #else
2685                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2686 #endif
2687                 sb->s_dirt = 1;
2688         }
2689
2690         if (sbi->s_blocks_per_group > blocksize * 8) {
2691                 ext4_msg(sb, KERN_ERR,
2692                        "#blocks per group too big: %lu",
2693                        sbi->s_blocks_per_group);
2694                 goto failed_mount;
2695         }
2696         if (sbi->s_inodes_per_group > blocksize * 8) {
2697                 ext4_msg(sb, KERN_ERR,
2698                        "#inodes per group too big: %lu",
2699                        sbi->s_inodes_per_group);
2700                 goto failed_mount;
2701         }
2702
2703         /*
2704          * Test whether we have more sectors than will fit in sector_t,
2705          * and whether the max offset is addressable by the page cache.
2706          */
2707         if ((ext4_blocks_count(es) >
2708              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2709             (ext4_blocks_count(es) >
2710              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2711                 ext4_msg(sb, KERN_ERR, "filesystem"
2712                          " too large to mount safely on this system");
2713                 if (sizeof(sector_t) < 8)
2714                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2715                 ret = -EFBIG;
2716                 goto failed_mount;
2717         }
2718
2719         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2720                 goto cantfind_ext4;
2721
2722         /* check blocks count against device size */
2723         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2724         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2725                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2726                        "exceeds size of device (%llu blocks)",
2727                        ext4_blocks_count(es), blocks_count);
2728                 goto failed_mount;
2729         }
2730
2731         /*
2732          * It makes no sense for the first data block to be beyond the end
2733          * of the filesystem.
2734          */
2735         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2736                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2737                          "block %u is beyond end of filesystem (%llu)",
2738                          le32_to_cpu(es->s_first_data_block),
2739                          ext4_blocks_count(es));
2740                 goto failed_mount;
2741         }
2742         blocks_count = (ext4_blocks_count(es) -
2743                         le32_to_cpu(es->s_first_data_block) +
2744                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2745         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2746         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2747                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2748                        "(block count %llu, first data block %u, "
2749                        "blocks per group %lu)", sbi->s_groups_count,
2750                        ext4_blocks_count(es),
2751                        le32_to_cpu(es->s_first_data_block),
2752                        EXT4_BLOCKS_PER_GROUP(sb));
2753                 goto failed_mount;
2754         }
2755         sbi->s_groups_count = blocks_count;
2756         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2757                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2758         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2759                    EXT4_DESC_PER_BLOCK(sb);
2760         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2761                                     GFP_KERNEL);
2762         if (sbi->s_group_desc == NULL) {
2763                 ext4_msg(sb, KERN_ERR, "not enough memory");
2764                 goto failed_mount;
2765         }
2766
2767 #ifdef CONFIG_PROC_FS
2768         if (ext4_proc_root)
2769                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2770 #endif
2771
2772         bgl_lock_init(sbi->s_blockgroup_lock);
2773
2774         for (i = 0; i < db_count; i++) {
2775                 block = descriptor_loc(sb, logical_sb_block, i);
2776                 sbi->s_group_desc[i] = sb_bread(sb, block);
2777                 if (!sbi->s_group_desc[i]) {
2778                         ext4_msg(sb, KERN_ERR,
2779                                "can't read group descriptor %d", i);
2780                         db_count = i;
2781                         goto failed_mount2;
2782                 }
2783         }
2784         if (!ext4_check_descriptors(sb)) {
2785                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2786                 goto failed_mount2;
2787         }
2788         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2789                 if (!ext4_fill_flex_info(sb)) {
2790                         ext4_msg(sb, KERN_ERR,
2791                                "unable to initialize "
2792                                "flex_bg meta info!");
2793                         goto failed_mount2;
2794                 }
2795
2796         sbi->s_gdb_count = db_count;
2797         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2798         spin_lock_init(&sbi->s_next_gen_lock);
2799
2800         sbi->s_stripe = ext4_get_stripe_size(sbi);
2801         sbi->s_max_writeback_mb_bump = 128;
2802
2803         /*
2804          * set up enough so that it can read an inode
2805          */
2806         if (!test_opt(sb, NOLOAD) &&
2807             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2808                 sb->s_op = &ext4_sops;
2809         else
2810                 sb->s_op = &ext4_nojournal_sops;
2811         sb->s_export_op = &ext4_export_ops;
2812         sb->s_xattr = ext4_xattr_handlers;
2813 #ifdef CONFIG_QUOTA
2814         sb->s_qcop = &ext4_qctl_operations;
2815         sb->dq_op = &ext4_quota_operations;
2816 #endif
2817         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2818         mutex_init(&sbi->s_orphan_lock);
2819         mutex_init(&sbi->s_resize_lock);
2820
2821         sb->s_root = NULL;
2822
2823         needs_recovery = (es->s_last_orphan != 0 ||
2824                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2825                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2826
2827         /*
2828          * The first inode we look at is the journal inode.  Don't try
2829          * root first: it may be modified in the journal!
2830          */
2831         if (!test_opt(sb, NOLOAD) &&
2832             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2833                 if (ext4_load_journal(sb, es, journal_devnum))
2834                         goto failed_mount3;
2835         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2836               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2837                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2838                        "suppressed and not mounted read-only");
2839                 goto failed_mount_wq;
2840         } else {
2841                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2842                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2843                 sbi->s_journal = NULL;
2844                 needs_recovery = 0;
2845                 goto no_journal;
2846         }
2847
2848         if (ext4_blocks_count(es) > 0xffffffffULL &&
2849             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2850                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2851                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2852                 goto failed_mount_wq;
2853         }
2854
2855         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2856                 jbd2_journal_set_features(sbi->s_journal,
2857                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2858                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2859         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2860                 jbd2_journal_set_features(sbi->s_journal,
2861                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2862                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2863                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2864         } else {
2865                 jbd2_journal_clear_features(sbi->s_journal,
2866                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2867                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2868         }
2869
2870         /* We have now updated the journal if required, so we can
2871          * validate the data journaling mode. */
2872         switch (test_opt(sb, DATA_FLAGS)) {
2873         case 0:
2874                 /* No mode set, assume a default based on the journal
2875                  * capabilities: ORDERED_DATA if the journal can
2876                  * cope, else JOURNAL_DATA
2877                  */
2878                 if (jbd2_journal_check_available_features
2879                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2880                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2881                 else
2882                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2883                 break;
2884
2885         case EXT4_MOUNT_ORDERED_DATA:
2886         case EXT4_MOUNT_WRITEBACK_DATA:
2887                 if (!jbd2_journal_check_available_features
2888                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2889                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2890                                "requested data journaling mode");
2891                         goto failed_mount_wq;
2892                 }
2893         default:
2894                 break;
2895         }
2896         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2897
2898 no_journal:
2899         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2900                                   ext4_count_free_blocks(sb));
2901         if (!err)
2902                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2903                                           ext4_count_free_inodes(sb));
2904         if (!err)
2905                 err = percpu_counter_init(&sbi->s_dirs_counter,
2906                                           ext4_count_dirs(sb));
2907         if (!err)
2908                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2909         if (err) {
2910                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2911                 goto failed_mount_wq;
2912         }
2913         if (test_opt(sb, NOBH)) {
2914                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2915                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2916                                 "its supported only with writeback mode");
2917                         clear_opt(sbi->s_mount_opt, NOBH);
2918                 }
2919                 if (test_opt(sb, DIOREAD_NOLOCK)) {
2920                         ext4_msg(sb, KERN_WARNING, "dioread_nolock option is "
2921                                 "not supported with nobh mode");
2922                         goto failed_mount_wq;
2923                 }
2924         }
2925         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2926         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2927                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2928                 goto failed_mount_wq;
2929         }
2930
2931         /*
2932          * The jbd2_journal_load will have done any necessary log recovery,
2933          * so we can safely mount the rest of the filesystem now.
2934          */
2935
2936         root = ext4_iget(sb, EXT4_ROOT_INO);
2937         if (IS_ERR(root)) {
2938                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2939                 ret = PTR_ERR(root);
2940                 goto failed_mount4;
2941         }
2942         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2943                 iput(root);
2944                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2945                 goto failed_mount4;
2946         }
2947         sb->s_root = d_alloc_root(root);
2948         if (!sb->s_root) {
2949                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2950                 iput(root);
2951                 ret = -ENOMEM;
2952                 goto failed_mount4;
2953         }
2954
2955         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2956
2957         /* determine the minimum size of new large inodes, if present */
2958         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2959                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2960                                                      EXT4_GOOD_OLD_INODE_SIZE;
2961                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2962                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2963                         if (sbi->s_want_extra_isize <
2964                             le16_to_cpu(es->s_want_extra_isize))
2965                                 sbi->s_want_extra_isize =
2966                                         le16_to_cpu(es->s_want_extra_isize);
2967                         if (sbi->s_want_extra_isize <
2968                             le16_to_cpu(es->s_min_extra_isize))
2969                                 sbi->s_want_extra_isize =
2970                                         le16_to_cpu(es->s_min_extra_isize);
2971                 }
2972         }
2973         /* Check if enough inode space is available */
2974         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2975                                                         sbi->s_inode_size) {
2976                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2977                                                        EXT4_GOOD_OLD_INODE_SIZE;
2978                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2979                          "available");
2980         }
2981
2982         if (test_opt(sb, DELALLOC) &&
2983             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2984                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2985                          "requested data journaling mode");
2986                 clear_opt(sbi->s_mount_opt, DELALLOC);
2987         }
2988         if (test_opt(sb, DIOREAD_NOLOCK)) {
2989                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2990                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2991                                 "option - requested data journaling mode");
2992                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2993                 }
2994                 if (sb->s_blocksize < PAGE_SIZE) {
2995                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2996                                 "option - block size is too small");
2997                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2998                 }
2999         }
3000
3001         err = ext4_setup_system_zone(sb);
3002         if (err) {
3003                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3004                          "zone (%d)", err);
3005                 goto failed_mount4;
3006         }
3007
3008         ext4_ext_init(sb);
3009         err = ext4_mb_init(sb, needs_recovery);
3010         if (err) {
3011                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
3012                          err);
3013                 goto failed_mount4;
3014         }
3015
3016         sbi->s_kobj.kset = ext4_kset;
3017         init_completion(&sbi->s_kobj_unregister);
3018         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3019                                    "%s", sb->s_id);
3020         if (err) {
3021                 ext4_mb_release(sb);
3022                 ext4_ext_release(sb);
3023                 goto failed_mount4;
3024         };
3025
3026         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3027         ext4_orphan_cleanup(sb, es);
3028         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3029         if (needs_recovery) {
3030                 ext4_msg(sb, KERN_INFO, "recovery complete");
3031                 ext4_mark_recovery_complete(sb, es);
3032         }
3033         if (EXT4_SB(sb)->s_journal) {
3034                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3035                         descr = " journalled data mode";
3036                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3037                         descr = " ordered data mode";
3038                 else
3039                         descr = " writeback data mode";
3040         } else
3041                 descr = "out journal";
3042
3043         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3044                 "Opts: %s", descr, orig_data);
3045
3046         lock_kernel();
3047         kfree(orig_data);
3048         return 0;
3049
3050 cantfind_ext4:
3051         if (!silent)
3052                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3053         goto failed_mount;
3054
3055 failed_mount4:
3056         ext4_msg(sb, KERN_ERR, "mount failed");
3057         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3058 failed_mount_wq:
3059         ext4_release_system_zone(sb);
3060         if (sbi->s_journal) {
3061                 jbd2_journal_destroy(sbi->s_journal);
3062                 sbi->s_journal = NULL;
3063         }
3064         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3065         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3066         percpu_counter_destroy(&sbi->s_dirs_counter);
3067         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3068 failed_mount3:
3069         if (sbi->s_flex_groups) {
3070                 if (is_vmalloc_addr(sbi->s_flex_groups))
3071                         vfree(sbi->s_flex_groups);
3072                 else
3073                         kfree(sbi->s_flex_groups);
3074         }
3075 failed_mount2:
3076         for (i = 0; i < db_count; i++)
3077                 brelse(sbi->s_group_desc[i]);
3078         kfree(sbi->s_group_desc);
3079 failed_mount:
3080         if (sbi->s_proc) {
3081                 remove_proc_entry(sb->s_id, ext4_proc_root);
3082         }
3083 #ifdef CONFIG_QUOTA
3084         for (i = 0; i < MAXQUOTAS; i++)
3085                 kfree(sbi->s_qf_names[i]);
3086 #endif
3087         ext4_blkdev_remove(sbi);
3088         brelse(bh);
3089 out_fail:
3090         sb->s_fs_info = NULL;
3091         kfree(sbi->s_blockgroup_lock);
3092         kfree(sbi);
3093         lock_kernel();
3094         kfree(orig_data);
3095         return ret;
3096 }
3097
3098 /*
3099  * Setup any per-fs journal parameters now.  We'll do this both on
3100  * initial mount, once the journal has been initialised but before we've
3101  * done any recovery; and again on any subsequent remount.
3102  */
3103 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3104 {
3105         struct ext4_sb_info *sbi = EXT4_SB(sb);
3106
3107         journal->j_commit_interval = sbi->s_commit_interval;
3108         journal->j_min_batch_time = sbi->s_min_batch_time;
3109         journal->j_max_batch_time = sbi->s_max_batch_time;
3110
3111         spin_lock(&journal->j_state_lock);
3112         if (test_opt(sb, BARRIER))
3113                 journal->j_flags |= JBD2_BARRIER;
3114         else
3115                 journal->j_flags &= ~JBD2_BARRIER;
3116         if (test_opt(sb, DATA_ERR_ABORT))
3117                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3118         else
3119                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3120         spin_unlock(&journal->j_state_lock);
3121 }
3122
3123 static journal_t *ext4_get_journal(struct super_block *sb,
3124                                    unsigned int journal_inum)
3125 {
3126         struct inode *journal_inode;
3127         journal_t *journal;
3128
3129         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3130
3131         /* First, test for the existence of a valid inode on disk.  Bad
3132          * things happen if we iget() an unused inode, as the subsequent
3133          * iput() will try to delete it. */
3134
3135         journal_inode = ext4_iget(sb, journal_inum);
3136         if (IS_ERR(journal_inode)) {
3137                 ext4_msg(sb, KERN_ERR, "no journal found");
3138                 return NULL;
3139         }
3140         if (!journal_inode->i_nlink) {
3141                 make_bad_inode(journal_inode);
3142                 iput(journal_inode);
3143                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3144                 return NULL;
3145         }
3146
3147         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3148                   journal_inode, journal_inode->i_size);
3149         if (!S_ISREG(journal_inode->i_mode)) {
3150                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3151                 iput(journal_inode);
3152                 return NULL;
3153         }
3154
3155         journal = jbd2_journal_init_inode(journal_inode);
3156         if (!journal) {
3157                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3158                 iput(journal_inode);
3159                 return NULL;
3160         }
3161         journal->j_private = sb;
3162         ext4_init_journal_params(sb, journal);
3163         return journal;
3164 }
3165
3166 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3167                                        dev_t j_dev)
3168 {
3169         struct buffer_head *bh;
3170         journal_t *journal;
3171         ext4_fsblk_t start;
3172         ext4_fsblk_t len;
3173         int hblock, blocksize;
3174         ext4_fsblk_t sb_block;
3175         unsigned long offset;
3176         struct ext4_super_block *es;
3177         struct block_device *bdev;
3178
3179         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3180
3181         bdev = ext4_blkdev_get(j_dev, sb);
3182         if (bdev == NULL)
3183                 return NULL;
3184
3185         if (bd_claim(bdev, sb)) {
3186                 ext4_msg(sb, KERN_ERR,
3187                         "failed to claim external journal device");
3188                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3189                 return NULL;
3190         }
3191
3192         blocksize = sb->s_blocksize;
3193         hblock = bdev_logical_block_size(bdev);
3194         if (blocksize < hblock) {
3195                 ext4_msg(sb, KERN_ERR,
3196                         "blocksize too small for journal device");
3197                 goto out_bdev;
3198         }
3199
3200         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3201         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3202         set_blocksize(bdev, blocksize);
3203         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3204                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3205                        "external journal");
3206                 goto out_bdev;
3207         }
3208
3209         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3210         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3211             !(le32_to_cpu(es->s_feature_incompat) &
3212               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3213                 ext4_msg(sb, KERN_ERR, "external journal has "
3214                                         "bad superblock");
3215                 brelse(bh);
3216                 goto out_bdev;
3217         }
3218
3219         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3220                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3221                 brelse(bh);
3222                 goto out_bdev;
3223         }
3224
3225         len = ext4_blocks_count(es);
3226         start = sb_block + 1;
3227         brelse(bh);     /* we're done with the superblock */
3228
3229         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3230                                         start, len, blocksize);
3231         if (!journal) {
3232                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3233                 goto out_bdev;
3234         }
3235         journal->j_private = sb;
3236         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3237         wait_on_buffer(journal->j_sb_buffer);
3238         if (!buffer_uptodate(journal->j_sb_buffer)) {
3239                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3240                 goto out_journal;
3241         }
3242         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3243                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3244                                         "user (unsupported) - %d",
3245                         be32_to_cpu(journal->j_superblock->s_nr_users));
3246                 goto out_journal;
3247         }
3248         EXT4_SB(sb)->journal_bdev = bdev;
3249         ext4_init_journal_params(sb, journal);
3250         return journal;
3251
3252 out_journal:
3253         jbd2_journal_destroy(journal);
3254 out_bdev:
3255         ext4_blkdev_put(bdev);
3256         return NULL;
3257 }
3258
3259 static int ext4_load_journal(struct super_block *sb,
3260                              struct ext4_super_block *es,
3261                              unsigned long journal_devnum)
3262 {
3263         journal_t *journal;
3264         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3265         dev_t journal_dev;
3266         int err = 0;
3267         int really_read_only;
3268
3269         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3270
3271         if (journal_devnum &&
3272             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3273                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3274                         "numbers have changed");
3275                 journal_dev = new_decode_dev(journal_devnum);
3276         } else
3277                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3278
3279         really_read_only = bdev_read_only(sb->s_bdev);
3280
3281         /*
3282          * Are we loading a blank journal or performing recovery after a
3283          * crash?  For recovery, we need to check in advance whether we
3284          * can get read-write access to the device.
3285          */
3286         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3287                 if (sb->s_flags & MS_RDONLY) {
3288                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3289                                         "required on readonly filesystem");
3290                         if (really_read_only) {
3291                                 ext4_msg(sb, KERN_ERR, "write access "
3292                                         "unavailable, cannot proceed");
3293                                 return -EROFS;
3294                         }
3295                         ext4_msg(sb, KERN_INFO, "write access will "
3296                                "be enabled during recovery");
3297                 }
3298         }
3299
3300         if (journal_inum && journal_dev) {
3301                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3302                        "and inode journals!");
3303                 return -EINVAL;
3304         }
3305
3306         if (journal_inum) {
3307                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3308                         return -EINVAL;
3309         } else {
3310                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3311                         return -EINVAL;
3312         }
3313
3314         if (!(journal->j_flags & JBD2_BARRIER))
3315                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3316
3317         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3318                 err = jbd2_journal_update_format(journal);
3319                 if (err)  {
3320                         ext4_msg(sb, KERN_ERR, "error updating journal");
3321                         jbd2_journal_destroy(journal);
3322                         return err;
3323                 }
3324         }
3325
3326         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3327                 err = jbd2_journal_wipe(journal, !really_read_only);
3328         if (!err)
3329                 err = jbd2_journal_load(journal);
3330
3331         if (err) {
3332                 ext4_msg(sb, KERN_ERR, "error loading journal");
3333                 jbd2_journal_destroy(journal);
3334                 return err;
3335         }
3336
3337         EXT4_SB(sb)->s_journal = journal;
3338         ext4_clear_journal_err(sb, es);
3339
3340         if (journal_devnum &&
3341             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3342                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3343
3344                 /* Make sure we flush the recovery flag to disk. */
3345                 ext4_commit_super(sb, 1);
3346         }
3347
3348         return 0;
3349 }
3350
3351 static int ext4_commit_super(struct super_block *sb, int sync)
3352 {
3353         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3354         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3355         int error = 0;
3356
3357         if (!sbh)
3358                 return error;
3359         if (buffer_write_io_error(sbh)) {
3360                 /*
3361                  * Oh, dear.  A previous attempt to write the
3362                  * superblock failed.  This could happen because the
3363                  * USB device was yanked out.  Or it could happen to
3364                  * be a transient write error and maybe the block will
3365                  * be remapped.  Nothing we can do but to retry the
3366                  * write and hope for the best.
3367                  */
3368                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3369                        "superblock detected");
3370                 clear_buffer_write_io_error(sbh);
3371                 set_buffer_uptodate(sbh);
3372         }
3373         /*
3374          * If the file system is mounted read-only, don't update the
3375          * superblock write time.  This avoids updating the superblock
3376          * write time when we are mounting the root file system
3377          * read/only but we need to replay the journal; at that point,
3378          * for people who are east of GMT and who make their clock
3379          * tick in localtime for Windows bug-for-bug compatibility,
3380          * the clock is set in the future, and this will cause e2fsck
3381          * to complain and force a full file system check.
3382          */
3383         if (!(sb->s_flags & MS_RDONLY))
3384                 es->s_wtime = cpu_to_le32(get_seconds());
3385         es->s_kbytes_written =
3386                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3387                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3388                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3389         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3390                                         &EXT4_SB(sb)->s_freeblocks_counter));
3391         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3392                                         &EXT4_SB(sb)->s_freeinodes_counter));
3393         sb->s_dirt = 0;
3394         BUFFER_TRACE(sbh, "marking dirty");
3395         mark_buffer_dirty(sbh);
3396         if (sync) {
3397                 error = sync_dirty_buffer(sbh);
3398                 if (error)
3399                         return error;
3400
3401                 error = buffer_write_io_error(sbh);
3402                 if (error) {
3403                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3404                                "superblock");
3405                         clear_buffer_write_io_error(sbh);
3406                         set_buffer_uptodate(sbh);
3407                 }
3408         }
3409         return error;
3410 }
3411
3412 /*
3413  * Have we just finished recovery?  If so, and if we are mounting (or
3414  * remounting) the filesystem readonly, then we will end up with a
3415  * consistent fs on disk.  Record that fact.
3416  */
3417 static void ext4_mark_recovery_complete(struct super_block *sb,
3418                                         struct ext4_super_block *es)
3419 {
3420         journal_t *journal = EXT4_SB(sb)->s_journal;
3421
3422         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3423                 BUG_ON(journal != NULL);
3424                 return;
3425         }
3426         jbd2_journal_lock_updates(journal);
3427         if (jbd2_journal_flush(journal) < 0)
3428                 goto out;
3429
3430         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3431             sb->s_flags & MS_RDONLY) {
3432                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3433                 ext4_commit_super(sb, 1);
3434         }
3435
3436 out:
3437         jbd2_journal_unlock_updates(journal);
3438 }
3439
3440 /*
3441  * If we are mounting (or read-write remounting) a filesystem whose journal
3442  * has recorded an error from a previous lifetime, move that error to the
3443  * main filesystem now.
3444  */
3445 static void ext4_clear_journal_err(struct super_block *sb,
3446                                    struct ext4_super_block *es)
3447 {
3448         journal_t *journal;
3449         int j_errno;
3450         const char *errstr;
3451
3452         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3453
3454         journal = EXT4_SB(sb)->s_journal;
3455
3456         /*
3457          * Now check for any error status which may have been recorded in the
3458          * journal by a prior ext4_error() or ext4_abort()
3459          */
3460
3461         j_errno = jbd2_journal_errno(journal);
3462         if (j_errno) {
3463                 char nbuf[16];
3464
3465                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3466                 ext4_warning(sb, "Filesystem error recorded "
3467                              "from previous mount: %s", errstr);
3468                 ext4_warning(sb, "Marking fs in need of filesystem check.");
3469
3470                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3471                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3472                 ext4_commit_super(sb, 1);
3473
3474                 jbd2_journal_clear_err(journal);
3475         }
3476 }
3477
3478 /*
3479  * Force the running and committing transactions to commit,
3480  * and wait on the commit.
3481  */
3482 int ext4_force_commit(struct super_block *sb)
3483 {
3484         journal_t *journal;
3485         int ret = 0;
3486
3487         if (sb->s_flags & MS_RDONLY)
3488                 return 0;
3489
3490         journal = EXT4_SB(sb)->s_journal;
3491         if (journal) {
3492                 vfs_check_frozen(sb, SB_FREEZE_WRITE);
3493                 ret = ext4_journal_force_commit(journal);
3494         }
3495
3496         return ret;
3497 }
3498
3499 static void ext4_write_super(struct super_block *sb)
3500 {
3501         lock_super(sb);
3502         ext4_commit_super(sb, 1);
3503         unlock_super(sb);
3504 }
3505
3506 static int ext4_sync_fs(struct super_block *sb, int wait)
3507 {
3508         int ret = 0;
3509         tid_t target;
3510         struct ext4_sb_info *sbi = EXT4_SB(sb);
3511
3512         trace_ext4_sync_fs(sb, wait);
3513         flush_workqueue(sbi->dio_unwritten_wq);
3514         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3515                 if (wait)
3516                         jbd2_log_wait_commit(sbi->s_journal, target);
3517         }
3518         return ret;
3519 }
3520
3521 /*
3522  * LVM calls this function before a (read-only) snapshot is created.  This
3523  * gives us a chance to flush the journal completely and mark the fs clean.
3524  */
3525 static int ext4_freeze(struct super_block *sb)
3526 {
3527         int error = 0;
3528         journal_t *journal;
3529
3530         if (sb->s_flags & MS_RDONLY)
3531                 return 0;
3532
3533         journal = EXT4_SB(sb)->s_journal;
3534
3535         /* Now we set up the journal barrier. */
3536         jbd2_journal_lock_updates(journal);
3537
3538         /*
3539          * Don't clear the needs_recovery flag if we failed to flush
3540          * the journal.
3541          */
3542         error = jbd2_journal_flush(journal);
3543         if (error < 0)
3544                 goto out;
3545
3546         /* Journal blocked and flushed, clear needs_recovery flag. */
3547         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3548         error = ext4_commit_super(sb, 1);
3549 out:
3550         /* we rely on s_frozen to stop further updates */
3551         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3552         return error;
3553 }
3554
3555 /*
3556  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3557  * flag here, even though the filesystem is not technically dirty yet.
3558  */
3559 static int ext4_unfreeze(struct super_block *sb)
3560 {
3561         if (sb->s_flags & MS_RDONLY)
3562                 return 0;
3563
3564         lock_super(sb);
3565         /* Reset the needs_recovery flag before the fs is unlocked. */
3566         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3567         ext4_commit_super(sb, 1);
3568         unlock_super(sb);
3569         return 0;
3570 }
3571
3572 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3573 {
3574         struct ext4_super_block *es;
3575         struct ext4_sb_info *sbi = EXT4_SB(sb);
3576         ext4_fsblk_t n_blocks_count = 0;
3577         unsigned long old_sb_flags;
3578         struct ext4_mount_options old_opts;
3579         ext4_group_t g;
3580         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3581         int err;
3582 #ifdef CONFIG_QUOTA
3583         int i;
3584 #endif
3585         char *orig_data = kstrdup(data, GFP_KERNEL);
3586
3587         lock_kernel();
3588
3589         /* Store the original options */
3590         lock_super(sb);
3591         old_sb_flags = sb->s_flags;
3592         old_opts.s_mount_opt = sbi->s_mount_opt;
3593         old_opts.s_resuid = sbi->s_resuid;
3594         old_opts.s_resgid = sbi->s_resgid;
3595         old_opts.s_commit_interval = sbi->s_commit_interval;
3596         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3597         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3598 #ifdef CONFIG_QUOTA
3599         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3600         for (i = 0; i < MAXQUOTAS; i++)
3601                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3602 #endif
3603         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3604                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3605
3606         /*
3607          * Allow the "check" option to be passed as a remount option.
3608          */
3609         if (!parse_options(data, sb, NULL, &journal_ioprio,
3610                            &n_blocks_count, 1)) {
3611                 err = -EINVAL;
3612                 goto restore_opts;
3613         }
3614
3615         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3616                 ext4_abort(sb, __func__, "Abort forced by user");
3617
3618         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3619                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3620
3621         es = sbi->s_es;
3622
3623         if (sbi->s_journal) {
3624                 ext4_init_journal_params(sb, sbi->s_journal);
3625                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3626         }
3627
3628         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3629                 n_blocks_count > ext4_blocks_count(es)) {
3630                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3631                         err = -EROFS;
3632                         goto restore_opts;
3633                 }
3634
3635                 if (*flags & MS_RDONLY) {
3636                         /*
3637                          * First of all, the unconditional stuff we have to do
3638                          * to disable replay of the journal when we next remount
3639                          */
3640                         sb->s_flags |= MS_RDONLY;
3641
3642                         /*
3643                          * OK, test if we are remounting a valid rw partition
3644                          * readonly, and if so set the rdonly flag and then
3645                          * mark the partition as valid again.
3646                          */
3647                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3648                             (sbi->s_mount_state & EXT4_VALID_FS))
3649                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3650
3651                         if (sbi->s_journal)
3652                                 ext4_mark_recovery_complete(sb, es);
3653                 } else {
3654                         /* Make sure we can mount this feature set readwrite */
3655                         if (!ext4_feature_set_ok(sb, 0)) {
3656                                 err = -EROFS;
3657                                 goto restore_opts;
3658                         }
3659                         /*
3660                          * Make sure the group descriptor checksums
3661                          * are sane.  If they aren't, refuse to remount r/w.
3662                          */
3663                         for (g = 0; g < sbi->s_groups_count; g++) {
3664                                 struct ext4_group_desc *gdp =
3665                                         ext4_get_group_desc(sb, g, NULL);
3666
3667                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3668                                         ext4_msg(sb, KERN_ERR,
3669                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3670                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3671                                                le16_to_cpu(gdp->bg_checksum));
3672                                         err = -EINVAL;
3673                                         goto restore_opts;
3674                                 }
3675                         }
3676
3677                         /*
3678                          * If we have an unprocessed orphan list hanging
3679                          * around from a previously readonly bdev mount,
3680                          * require a full umount/remount for now.
3681                          */
3682                         if (es->s_last_orphan) {
3683                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3684                                        "remount RDWR because of unprocessed "
3685                                        "orphan inode list.  Please "
3686                                        "umount/remount instead");
3687                                 err = -EINVAL;
3688                                 goto restore_opts;
3689                         }
3690
3691                         /*
3692                          * Mounting a RDONLY partition read-write, so reread
3693                          * and store the current valid flag.  (It may have
3694                          * been changed by e2fsck since we originally mounted
3695                          * the partition.)
3696                          */
3697                         if (sbi->s_journal)
3698                                 ext4_clear_journal_err(sb, es);
3699                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3700                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3701                                 goto restore_opts;
3702                         if (!ext4_setup_super(sb, es, 0))
3703                                 sb->s_flags &= ~MS_RDONLY;
3704                 }
3705         }
3706         ext4_setup_system_zone(sb);
3707         if (sbi->s_journal == NULL)
3708                 ext4_commit_super(sb, 1);
3709
3710 #ifdef CONFIG_QUOTA
3711         /* Release old quota file names */
3712         for (i = 0; i < MAXQUOTAS; i++)
3713                 if (old_opts.s_qf_names[i] &&
3714                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3715                         kfree(old_opts.s_qf_names[i]);
3716 #endif
3717         unlock_super(sb);
3718         unlock_kernel();
3719
3720         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
3721         kfree(orig_data);
3722         return 0;
3723
3724 restore_opts:
3725         sb->s_flags = old_sb_flags;
3726         sbi->s_mount_opt = old_opts.s_mount_opt;
3727         sbi->s_resuid = old_opts.s_resuid;
3728         sbi->s_resgid = old_opts.s_resgid;
3729         sbi->s_commit_interval = old_opts.s_commit_interval;
3730         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3731         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3732 #ifdef CONFIG_QUOTA
3733         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3734         for (i = 0; i < MAXQUOTAS; i++) {
3735                 if (sbi->s_qf_names[i] &&
3736                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3737                         kfree(sbi->s_qf_names[i]);
3738                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3739         }
3740 #endif
3741         unlock_super(sb);
3742         unlock_kernel();
3743         kfree(orig_data);
3744         return err;
3745 }
3746
3747 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3748 {
3749         struct super_block *sb = dentry->d_sb;
3750         struct ext4_sb_info *sbi = EXT4_SB(sb);
3751         struct ext4_super_block *es = sbi->s_es;
3752         u64 fsid;
3753
3754         if (test_opt(sb, MINIX_DF)) {
3755                 sbi->s_overhead_last = 0;
3756         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3757                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3758                 ext4_fsblk_t overhead = 0;
3759
3760                 /*
3761                  * Compute the overhead (FS structures).  This is constant
3762                  * for a given filesystem unless the number of block groups
3763                  * changes so we cache the previous value until it does.
3764                  */
3765
3766                 /*
3767                  * All of the blocks before first_data_block are
3768                  * overhead
3769                  */
3770                 overhead = le32_to_cpu(es->s_first_data_block);
3771
3772                 /*
3773                  * Add the overhead attributed to the superblock and
3774                  * block group descriptors.  If the sparse superblocks
3775                  * feature is turned on, then not all groups have this.
3776                  */
3777                 for (i = 0; i < ngroups; i++) {
3778                         overhead += ext4_bg_has_super(sb, i) +
3779                                 ext4_bg_num_gdb(sb, i);
3780                         cond_resched();
3781                 }
3782
3783                 /*
3784                  * Every block group has an inode bitmap, a block
3785                  * bitmap, and an inode table.
3786                  */
3787                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3788                 sbi->s_overhead_last = overhead;
3789                 smp_wmb();
3790                 sbi->s_blocks_last = ext4_blocks_count(es);
3791         }
3792
3793         buf->f_type = EXT4_SUPER_MAGIC;
3794         buf->f_bsize = sb->s_blocksize;
3795         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3796         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3797                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3798         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3799         if (buf->f_bfree < ext4_r_blocks_count(es))
3800                 buf->f_bavail = 0;
3801         buf->f_files = le32_to_cpu(es->s_inodes_count);
3802         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3803         buf->f_namelen = EXT4_NAME_LEN;
3804         fsid = le64_to_cpup((void *)es->s_uuid) ^
3805                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3806         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3807         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3808
3809         return 0;
3810 }
3811
3812 /* Helper function for writing quotas on sync - we need to start transaction
3813  * before quota file is locked for write. Otherwise the are possible deadlocks:
3814  * Process 1                         Process 2
3815  * ext4_create()                     quota_sync()
3816  *   jbd2_journal_start()                  write_dquot()
3817  *   dquot_initialize()                         down(dqio_mutex)
3818  *     down(dqio_mutex)                    jbd2_journal_start()
3819  *
3820  */
3821
3822 #ifdef CONFIG_QUOTA
3823
3824 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3825 {
3826         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3827 }
3828
3829 static int ext4_write_dquot(struct dquot *dquot)
3830 {
3831         int ret, err;
3832         handle_t *handle;
3833         struct inode *inode;
3834
3835         inode = dquot_to_inode(dquot);
3836         handle = ext4_journal_start(inode,
3837                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3838         if (IS_ERR(handle))
3839                 return PTR_ERR(handle);
3840         ret = dquot_commit(dquot);
3841         err = ext4_journal_stop(handle);
3842         if (!ret)
3843                 ret = err;
3844         return ret;
3845 }
3846
3847 static int ext4_acquire_dquot(struct dquot *dquot)
3848 {
3849         int ret, err;
3850         handle_t *handle;
3851
3852         handle = ext4_journal_start(dquot_to_inode(dquot),
3853                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3854         if (IS_ERR(handle))
3855                 return PTR_ERR(handle);
3856         ret = dquot_acquire(dquot);
3857         err = ext4_journal_stop(handle);
3858         if (!ret)
3859                 ret = err;
3860         return ret;
3861 }
3862
3863 static int ext4_release_dquot(struct dquot *dquot)
3864 {
3865         int ret, err;
3866         handle_t *handle;
3867
3868         handle = ext4_journal_start(dquot_to_inode(dquot),
3869                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3870         if (IS_ERR(handle)) {
3871                 /* Release dquot anyway to avoid endless cycle in dqput() */
3872                 dquot_release(dquot);
3873                 return PTR_ERR(handle);
3874         }
3875         ret = dquot_release(dquot);
3876         err = ext4_journal_stop(handle);
3877         if (!ret)
3878                 ret = err;
3879         return ret;
3880 }
3881
3882 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3883 {
3884         /* Are we journaling quotas? */
3885         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3886             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3887                 dquot_mark_dquot_dirty(dquot);
3888                 return ext4_write_dquot(dquot);
3889         } else {
3890                 return dquot_mark_dquot_dirty(dquot);
3891         }
3892 }
3893
3894 static int ext4_write_info(struct super_block *sb, int type)
3895 {
3896         int ret, err;
3897         handle_t *handle;
3898
3899         /* Data block + inode block */
3900         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3901         if (IS_ERR(handle))
3902                 return PTR_ERR(handle);
3903         ret = dquot_commit_info(sb, type);
3904         err = ext4_journal_stop(handle);
3905         if (!ret)
3906                 ret = err;
3907         return ret;
3908 }
3909
3910 /*
3911  * Turn on quotas during mount time - we need to find
3912  * the quota file and such...
3913  */
3914 static int ext4_quota_on_mount(struct super_block *sb, int type)
3915 {
3916         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3917                                   EXT4_SB(sb)->s_jquota_fmt, type);
3918 }
3919
3920 /*
3921  * Standard function to be called on quota_on
3922  */
3923 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3924                          char *name, int remount)
3925 {
3926         int err;
3927         struct path path;
3928
3929         if (!test_opt(sb, QUOTA))
3930                 return -EINVAL;
3931         /* When remounting, no checks are needed and in fact, name is NULL */
3932         if (remount)
3933                 return vfs_quota_on(sb, type, format_id, name, remount);
3934
3935         err = kern_path(name, LOOKUP_FOLLOW, &path);
3936         if (err)
3937                 return err;
3938
3939         /* Quotafile not on the same filesystem? */
3940         if (path.mnt->mnt_sb != sb) {
3941                 path_put(&path);
3942                 return -EXDEV;
3943         }
3944         /* Journaling quota? */
3945         if (EXT4_SB(sb)->s_qf_names[type]) {
3946                 /* Quotafile not in fs root? */
3947                 if (path.dentry->d_parent != sb->s_root)
3948                         ext4_msg(sb, KERN_WARNING,
3949                                 "Quota file not on filesystem root. "
3950                                 "Journaled quota will not work");
3951         }
3952
3953         /*
3954          * When we journal data on quota file, we have to flush journal to see
3955          * all updates to the file when we bypass pagecache...
3956          */
3957         if (EXT4_SB(sb)->s_journal &&
3958             ext4_should_journal_data(path.dentry->d_inode)) {
3959                 /*
3960                  * We don't need to lock updates but journal_flush() could
3961                  * otherwise be livelocked...
3962                  */
3963                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3964                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3965                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3966                 if (err) {
3967                         path_put(&path);
3968                         return err;
3969                 }
3970         }
3971
3972         err = vfs_quota_on_path(sb, type, format_id, &path);
3973         path_put(&path);
3974         return err;
3975 }
3976
3977 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3978  * acquiring the locks... As quota files are never truncated and quota code
3979  * itself serializes the operations (and noone else should touch the files)
3980  * we don't have to be afraid of races */
3981 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3982                                size_t len, loff_t off)
3983 {
3984         struct inode *inode = sb_dqopt(sb)->files[type];
3985         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3986         int err = 0;
3987         int offset = off & (sb->s_blocksize - 1);
3988         int tocopy;
3989         size_t toread;
3990         struct buffer_head *bh;
3991         loff_t i_size = i_size_read(inode);
3992
3993         if (off > i_size)
3994                 return 0;
3995         if (off+len > i_size)
3996                 len = i_size-off;
3997         toread = len;
3998         while (toread > 0) {
3999                 tocopy = sb->s_blocksize - offset < toread ?
4000                                 sb->s_blocksize - offset : toread;
4001                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4002                 if (err)
4003                         return err;
4004                 if (!bh)        /* A hole? */
4005                         memset(data, 0, tocopy);
4006                 else
4007                         memcpy(data, bh->b_data+offset, tocopy);
4008                 brelse(bh);
4009                 offset = 0;
4010                 toread -= tocopy;
4011                 data += tocopy;
4012                 blk++;
4013         }
4014         return len;
4015 }
4016
4017 /* Write to quotafile (we know the transaction is already started and has
4018  * enough credits) */
4019 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4020                                 const char *data, size_t len, loff_t off)
4021 {
4022         struct inode *inode = sb_dqopt(sb)->files[type];
4023         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4024         int err = 0;
4025         int offset = off & (sb->s_blocksize - 1);
4026         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
4027         struct buffer_head *bh;
4028         handle_t *handle = journal_current_handle();
4029
4030         if (EXT4_SB(sb)->s_journal && !handle) {
4031                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4032                         " cancelled because transaction is not started",
4033                         (unsigned long long)off, (unsigned long long)len);
4034                 return -EIO;
4035         }
4036         /*
4037          * Since we account only one data block in transaction credits,
4038          * then it is impossible to cross a block boundary.
4039          */
4040         if (sb->s_blocksize - offset < len) {
4041                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4042                         " cancelled because not block aligned",
4043                         (unsigned long long)off, (unsigned long long)len);
4044                 return -EIO;
4045         }
4046
4047         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4048         bh = ext4_bread(handle, inode, blk, 1, &err);
4049         if (!bh)
4050                 goto out;
4051         if (journal_quota) {
4052                 err = ext4_journal_get_write_access(handle, bh);
4053                 if (err) {
4054                         brelse(bh);
4055                         goto out;
4056                 }
4057         }
4058         lock_buffer(bh);
4059         memcpy(bh->b_data+offset, data, len);
4060         flush_dcache_page(bh->b_page);
4061         unlock_buffer(bh);
4062         if (journal_quota)
4063                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
4064         else {
4065                 /* Always do at least ordered writes for quotas */
4066                 err = ext4_jbd2_file_inode(handle, inode);
4067                 mark_buffer_dirty(bh);
4068         }
4069         brelse(bh);
4070 out:
4071         if (err) {
4072                 mutex_unlock(&inode->i_mutex);
4073                 return err;
4074         }
4075         if (inode->i_size < off + len) {
4076                 i_size_write(inode, off + len);
4077                 EXT4_I(inode)->i_disksize = inode->i_size;
4078         }
4079         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4080         ext4_mark_inode_dirty(handle, inode);
4081         mutex_unlock(&inode->i_mutex);
4082         return len;
4083 }
4084
4085 #endif
4086
4087 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
4088                        const char *dev_name, void *data, struct vfsmount *mnt)
4089 {
4090         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
4091 }
4092
4093 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4094 static struct file_system_type ext2_fs_type = {
4095         .owner          = THIS_MODULE,
4096         .name           = "ext2",
4097         .get_sb         = ext4_get_sb,
4098         .kill_sb        = kill_block_super,
4099         .fs_flags       = FS_REQUIRES_DEV,
4100 };
4101
4102 static inline void register_as_ext2(void)
4103 {
4104         int err = register_filesystem(&ext2_fs_type);
4105         if (err)
4106                 printk(KERN_WARNING
4107                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4108 }
4109
4110 static inline void unregister_as_ext2(void)
4111 {
4112         unregister_filesystem(&ext2_fs_type);
4113 }
4114 MODULE_ALIAS("ext2");
4115 #else
4116 static inline void register_as_ext2(void) { }
4117 static inline void unregister_as_ext2(void) { }
4118 #endif
4119
4120 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4121 static inline void register_as_ext3(void)
4122 {
4123         int err = register_filesystem(&ext3_fs_type);
4124         if (err)
4125                 printk(KERN_WARNING
4126                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4127 }
4128
4129 static inline void unregister_as_ext3(void)
4130 {
4131         unregister_filesystem(&ext3_fs_type);
4132 }
4133 MODULE_ALIAS("ext3");
4134 #else
4135 static inline void register_as_ext3(void) { }
4136 static inline void unregister_as_ext3(void) { }
4137 #endif
4138
4139 static struct file_system_type ext4_fs_type = {
4140         .owner          = THIS_MODULE,
4141         .name           = "ext4",
4142         .get_sb         = ext4_get_sb,
4143         .kill_sb        = kill_block_super,
4144         .fs_flags       = FS_REQUIRES_DEV,
4145 };
4146
4147 static int __init init_ext4_fs(void)
4148 {
4149         int err;
4150
4151         ext4_check_flag_values();
4152         err = init_ext4_system_zone();
4153         if (err)
4154                 return err;
4155         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4156         if (!ext4_kset)
4157                 goto out4;
4158         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4159         err = init_ext4_mballoc();
4160         if (err)
4161                 goto out3;
4162
4163         err = init_ext4_xattr();
4164         if (err)
4165                 goto out2;
4166         err = init_inodecache();
4167         if (err)
4168                 goto out1;
4169         register_as_ext2();
4170         register_as_ext3();
4171         err = register_filesystem(&ext4_fs_type);
4172         if (err)
4173                 goto out;
4174         return 0;
4175 out:
4176         unregister_as_ext2();
4177         unregister_as_ext3();
4178         destroy_inodecache();
4179 out1:
4180         exit_ext4_xattr();
4181 out2:
4182         exit_ext4_mballoc();
4183 out3:
4184         remove_proc_entry("fs/ext4", NULL);
4185         kset_unregister(ext4_kset);
4186 out4:
4187         exit_ext4_system_zone();
4188         return err;
4189 }
4190
4191 static void __exit exit_ext4_fs(void)
4192 {
4193         unregister_as_ext2();
4194         unregister_as_ext3();
4195         unregister_filesystem(&ext4_fs_type);
4196         destroy_inodecache();
4197         exit_ext4_xattr();
4198         exit_ext4_mballoc();
4199         remove_proc_entry("fs/ext4", NULL);
4200         kset_unregister(ext4_kset);
4201         exit_ext4_system_zone();
4202 }
4203
4204 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4205 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4206 MODULE_LICENSE("GPL");
4207 module_init(init_ext4_fs)
4208 module_exit(exit_ext4_fs)