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         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 650
 651         flush_workqueue(sbi->dio_unwritten_wq);
 652         destroy_workqueue(sbi->dio_unwritten_wq);
 653
 654         lock_super(sb);
 655         lock_kernel();
 656         if (sb->s_dirt)
 657                 ext4_commit_super(sb, 1);
 658
 659         if (sbi->s_journal) {
 660                 err = jbd2_journal_destroy(sbi->s_journal);
 661                 sbi->s_journal = NULL;
 662                 if (err < 0)
 663                         ext4_abort(sb, __func__,
 664                                    "Couldn't clean up the journal");
 665         }
 666
 667         ext4_release_system_zone(sb);
 668         ext4_mb_release(sb);
 669         ext4_ext_release(sb);
 670         ext4_xattr_put_super(sb);
 671
 672         if (!(sb->s_flags & MS_RDONLY)) {
 673                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 674                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 675                 ext4_commit_super(sb, 1);
 676         }
 677         if (sbi->s_proc) {
 678                 remove_proc_entry(sb->s_id, ext4_proc_root);
 679         }
 680         kobject_del(&sbi->s_kobj);
 681
 682         for (i = 0; i < sbi->s_gdb_count; i++)
 683                 brelse(sbi->s_group_desc[i]);
 684         kfree(sbi->s_group_desc);
 685         if (is_vmalloc_addr(sbi->s_flex_groups))
 686                 vfree(sbi->s_flex_groups);
 687         else
 688                 kfree(sbi->s_flex_groups);
 689         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 690         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 691         percpu_counter_destroy(&sbi->s_dirs_counter);
 692         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
 693         brelse(sbi->s_sbh);
 694 #ifdef CONFIG_QUOTA
 695         for (i = 0; i < MAXQUOTAS; i++)
 696                 kfree(sbi->s_qf_names[i]);
 697 #endif
 698
 699         /* Debugging code just in case the in-memory inode orphan list
 700          * isn't empty.  The on-disk one can be non-empty if we've
 701          * detected an error and taken the fs readonly, but the
 702          * in-memory list had better be clean by this point. */
 703         if (!list_empty(&sbi->s_orphan))
 704                 dump_orphan_list(sb, sbi);
 705         J_ASSERT(list_empty(&sbi->s_orphan));
 706
 707         invalidate_bdev(sb->s_bdev);
 708         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 709                 /*
 710                  * Invalidate the journal device's buffers.  We don't want them
 711                  * floating about in memory - the physical journal device may
 712                  * hotswapped, and it breaks the `ro-after' testing code.
 713                  */
 714                 sync_blockdev(sbi->journal_bdev);
 715                 invalidate_bdev(sbi->journal_bdev);
 716                 ext4_blkdev_remove(sbi);
 717         }
 718         sb->s_fs_info = NULL;
 719         /*
 720          * Now that we are completely done shutting down the
 721          * superblock, we need to actually destroy the kobject.
 722          */
 723         unlock_kernel();
 724         unlock_super(sb);
 725         kobject_put(&sbi->s_kobj);
 726         wait_for_completion(&sbi->s_kobj_unregister);
 727         kfree(sbi->s_blockgroup_lock);
 728         kfree(sbi);
 729 }
 730
 731 static struct kmem_cache *ext4_inode_cachep;
 732
 733 /*
 734  * Called inside transaction, so use GFP_NOFS
 735  */
 736 static struct inode *ext4_alloc_inode(struct super_block *sb)
 737 {
 738         struct ext4_inode_info *ei;
 739
 740         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 741         if (!ei)
 742                 return NULL;
 743
 744         ei->vfs_inode.i_version = 1;
 745         ei->vfs_inode.i_data.writeback_index = 0;
 746         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 747         INIT_LIST_HEAD(&ei->i_prealloc_list);
 748         spin_lock_init(&ei->i_prealloc_lock);
 749         /*
 750          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
 751          * therefore it can be null here.  Don't check it, just initialize
 752          * jinode.
 753          */
 754         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
 755         ei->i_reserved_data_blocks = 0;
 756         ei->i_reserved_meta_blocks = 0;
 757         ei->i_allocated_meta_blocks = 0;
 758         ei->i_da_metadata_calc_len = 0;
 759         ei->i_delalloc_reserved_flag = 0;
 760         spin_lock_init(&(ei->i_block_reservation_lock));
 761 #ifdef CONFIG_QUOTA
 762         ei->i_reserved_quota = 0;
 763 #endif
 764         INIT_LIST_HEAD(&ei->i_completed_io_list);
 765         spin_lock_init(&ei->i_completed_io_lock);
 766         ei->cur_aio_dio = NULL;
 767         ei->i_sync_tid = 0;
 768         ei->i_datasync_tid = 0;
 769
 770         return &ei->vfs_inode;
 771 }
 772
 773 static void ext4_destroy_inode(struct inode *inode)
 774 {
 775         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 776                 ext4_msg(inode->i_sb, KERN_ERR,
 777                          "Inode %lu (%p): orphan list check failed!",
 778                          inode->i_ino, EXT4_I(inode));
 779                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 780                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 781                                 true);
 782                 dump_stack();
 783         }
 784         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 785 }
 786
 787 static void init_once(void *foo)
 788 {
 789         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 790
 791         INIT_LIST_HEAD(&ei->i_orphan);
 792 #ifdef CONFIG_EXT4_FS_XATTR
 793         init_rwsem(&ei->xattr_sem);
 794 #endif
 795         init_rwsem(&ei->i_data_sem);
 796         inode_init_once(&ei->vfs_inode);
 797 }
 798
 799 static int init_inodecache(void)
 800 {
 801         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 802                                              sizeof(struct ext4_inode_info),
 803                                              0, (SLAB_RECLAIM_ACCOUNT|
 804                                                 SLAB_MEM_SPREAD),
 805                                              init_once);
 806         if (ext4_inode_cachep == NULL)
 807                 return -ENOMEM;
 808         return 0;
 809 }
 810
 811 static void destroy_inodecache(void)
 812 {
 813         kmem_cache_destroy(ext4_inode_cachep);
 814 }
 815
 816 static void ext4_clear_inode(struct inode *inode)
 817 {
 818         dquot_drop(inode);
 819         ext4_discard_preallocations(inode);
 820         if (EXT4_JOURNAL(inode))
 821                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
 822                                        &EXT4_I(inode)->jinode);
 823 }
 824
 825 static inline void ext4_show_quota_options(struct seq_file *seq,
 826                                            struct super_block *sb)
 827 {
 828 #if defined(CONFIG_QUOTA)
 829         struct ext4_sb_info *sbi = EXT4_SB(sb);
 830
 831         if (sbi->s_jquota_fmt) {
 832                 char *fmtname = "";
 833
 834                 switch (sbi->s_jquota_fmt) {
 835                 case QFMT_VFS_OLD:
 836                         fmtname = "vfsold";
 837                         break;
 838                 case QFMT_VFS_V0:
 839                         fmtname = "vfsv0";
 840                         break;
 841                 case QFMT_VFS_V1:
 842                         fmtname = "vfsv1";
 843                         break;
 844                 }
 845                 seq_printf(seq, ",jqfmt=%s", fmtname);
 846         }
 847
 848         if (sbi->s_qf_names[USRQUOTA])
 849                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
 850
 851         if (sbi->s_qf_names[GRPQUOTA])
 852                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
 853
 854         if (test_opt(sb, USRQUOTA))
 855                 seq_puts(seq, ",usrquota");
 856
 857         if (test_opt(sb, GRPQUOTA))
 858                 seq_puts(seq, ",grpquota");
 859 #endif
 860 }
 861
 862 /*
 863  * Show an option if
 864  *  - it's set to a non-default value OR
 865  *  - if the per-sb default is different from the global default
 866  */
 867 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
 868 {
 869         int def_errors;
 870         unsigned long def_mount_opts;
 871         struct super_block *sb = vfs->mnt_sb;
 872         struct ext4_sb_info *sbi = EXT4_SB(sb);
 873         struct ext4_super_block *es = sbi->s_es;
 874
 875         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
 876         def_errors     = le16_to_cpu(es->s_errors);
 877
 878         if (sbi->s_sb_block != 1)
 879                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
 880         if (test_opt(sb, MINIX_DF))
 881                 seq_puts(seq, ",minixdf");
 882         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
 883                 seq_puts(seq, ",grpid");
 884         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
 885                 seq_puts(seq, ",nogrpid");
 886         if (sbi->s_resuid != EXT4_DEF_RESUID ||
 887             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
 888                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
 889         }
 890         if (sbi->s_resgid != EXT4_DEF_RESGID ||
 891             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
 892                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
 893         }
 894         if (test_opt(sb, ERRORS_RO)) {
 895                 if (def_errors == EXT4_ERRORS_PANIC ||
 896                     def_errors == EXT4_ERRORS_CONTINUE) {
 897                         seq_puts(seq, ",errors=remount-ro");
 898                 }
 899         }
 900         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
 901                 seq_puts(seq, ",errors=continue");
 902         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
 903                 seq_puts(seq, ",errors=panic");
 904         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
 905                 seq_puts(seq, ",nouid32");
 906         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
 907                 seq_puts(seq, ",debug");
 908         if (test_opt(sb, OLDALLOC))
 909                 seq_puts(seq, ",oldalloc");
 910 #ifdef CONFIG_EXT4_FS_XATTR
 911         if (test_opt(sb, XATTR_USER) &&
 912                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
 913                 seq_puts(seq, ",user_xattr");
 914         if (!test_opt(sb, XATTR_USER) &&
 915             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
 916                 seq_puts(seq, ",nouser_xattr");
 917         }
 918 #endif
 919 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 920         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
 921                 seq_puts(seq, ",acl");
 922         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
 923                 seq_puts(seq, ",noacl");
 924 #endif
 925         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
 926                 seq_printf(seq, ",commit=%u",
 927                            (unsigned) (sbi->s_commit_interval / HZ));
 928         }
 929         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
 930                 seq_printf(seq, ",min_batch_time=%u",
 931                            (unsigned) sbi->s_min_batch_time);
 932         }
 933         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
 934                 seq_printf(seq, ",max_batch_time=%u",
 935                            (unsigned) sbi->s_min_batch_time);
 936         }
 937
 938         /*
 939          * We're changing the default of barrier mount option, so
 940          * let's always display its mount state so it's clear what its
 941          * status is.
 942          */
 943         seq_puts(seq, ",barrier=");
 944         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
 945         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
 946                 seq_puts(seq, ",journal_async_commit");
 947         else if (test_opt(sb, JOURNAL_CHECKSUM))
 948                 seq_puts(seq, ",journal_checksum");
 949         if (test_opt(sb, NOBH))
 950                 seq_puts(seq, ",nobh");
 951         if (test_opt(sb, I_VERSION))
 952                 seq_puts(seq, ",i_version");
 953         if (!test_opt(sb, DELALLOC))
 954                 seq_puts(seq, ",nodelalloc");
 955
 956
 957         if (sbi->s_stripe)
 958                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
 959         /*
 960          * journal mode get enabled in different ways
 961          * So just print the value even if we didn't specify it
 962          */
 963         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
 964                 seq_puts(seq, ",data=journal");
 965         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
 966                 seq_puts(seq, ",data=ordered");
 967         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
 968                 seq_puts(seq, ",data=writeback");
 969
 970         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
 971                 seq_printf(seq, ",inode_readahead_blks=%u",
 972                            sbi->s_inode_readahead_blks);
 973
 974         if (test_opt(sb, DATA_ERR_ABORT))
 975                 seq_puts(seq, ",data_err=abort");
 976
 977         if (test_opt(sb, NO_AUTO_DA_ALLOC))
 978                 seq_puts(seq, ",noauto_da_alloc");
 979
 980         if (test_opt(sb, DISCARD))
 981                 seq_puts(seq, ",discard");
 982
 983         if (test_opt(sb, NOLOAD))
 984                 seq_puts(seq, ",norecovery");
 985
 986         if (test_opt(sb, DIOREAD_NOLOCK))
 987                 seq_puts(seq, ",dioread_nolock");
 988
 989         ext4_show_quota_options(seq, sb);
 990
 991         return 0;
 992 }
 993
 994 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 995                                         u64 ino, u32 generation)
 996 {
 997         struct inode *inode;
 998
 999         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1000                 return ERR_PTR(-ESTALE);
1001         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1002                 return ERR_PTR(-ESTALE);
1003
1004         /* iget isn't really right if the inode is currently unallocated!!
1005          *
1006          * ext4_read_inode will return a bad_inode if the inode had been
1007          * deleted, so we should be safe.
1008          *
1009          * Currently we don't know the generation for parent directory, so
1010          * a generation of 0 means "accept any"
1011          */
1012         inode = ext4_iget(sb, ino);
1013         if (IS_ERR(inode))
1014                 return ERR_CAST(inode);
1015         if (generation && inode->i_generation != generation) {
1016                 iput(inode);
1017                 return ERR_PTR(-ESTALE);
1018         }
1019
1020         return inode;
1021 }
1022
1023 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1024                                         int fh_len, int fh_type)
1025 {
1026         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1027                                     ext4_nfs_get_inode);
1028 }
1029
1030 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1031                                         int fh_len, int fh_type)
1032 {
1033         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1034                                     ext4_nfs_get_inode);
1035 }
1036
1037 /*
1038  * Try to release metadata pages (indirect blocks, directories) which are
1039  * mapped via the block device.  Since these pages could have journal heads
1040  * which would prevent try_to_free_buffers() from freeing them, we must use
1041  * jbd2 layer's try_to_free_buffers() function to release them.
1042  */
1043 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1044                                  gfp_t wait)
1045 {
1046         journal_t *journal = EXT4_SB(sb)->s_journal;
1047
1048         WARN_ON(PageChecked(page));
1049         if (!page_has_buffers(page))
1050                 return 0;
1051         if (journal)
1052                 return jbd2_journal_try_to_free_buffers(journal, page,
1053                                                         wait & ~__GFP_WAIT);
1054         return try_to_free_buffers(page);
1055 }
1056
1057 #ifdef CONFIG_QUOTA
1058 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1059 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1060
1061 static int ext4_write_dquot(struct dquot *dquot);
1062 static int ext4_acquire_dquot(struct dquot *dquot);
1063 static int ext4_release_dquot(struct dquot *dquot);
1064 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1065 static int ext4_write_info(struct super_block *sb, int type);
1066 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1067                                 char *path);
1068 static int ext4_quota_on_mount(struct super_block *sb, int type);
1069 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1070                                size_t len, loff_t off);
1071 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1072                                 const char *data, size_t len, loff_t off);
1073
1074 static const struct dquot_operations ext4_quota_operations = {
1075 #ifdef CONFIG_QUOTA
1076         .get_reserved_space = ext4_get_reserved_space,
1077 #endif
1078         .write_dquot    = ext4_write_dquot,
1079         .acquire_dquot  = ext4_acquire_dquot,
1080         .release_dquot  = ext4_release_dquot,
1081         .mark_dirty     = ext4_mark_dquot_dirty,
1082         .write_info     = ext4_write_info,
1083         .alloc_dquot    = dquot_alloc,
1084         .destroy_dquot  = dquot_destroy,
1085 };
1086
1087 static const struct quotactl_ops ext4_qctl_operations = {
1088         .quota_on       = ext4_quota_on,
1089         .quota_off      = dquot_quota_off,
1090         .quota_sync     = dquot_quota_sync,
1091         .get_info       = dquot_get_dqinfo,
1092         .set_info       = dquot_set_dqinfo,
1093         .get_dqblk      = dquot_get_dqblk,
1094         .set_dqblk      = dquot_set_dqblk
1095 };
1096 #endif
1097
1098 static const struct super_operations ext4_sops = {
1099         .alloc_inode    = ext4_alloc_inode,
1100         .destroy_inode  = ext4_destroy_inode,
1101         .write_inode    = ext4_write_inode,
1102         .dirty_inode    = ext4_dirty_inode,
1103         .delete_inode   = ext4_delete_inode,
1104         .put_super      = ext4_put_super,
1105         .sync_fs        = ext4_sync_fs,
1106         .freeze_fs      = ext4_freeze,
1107         .unfreeze_fs    = ext4_unfreeze,
1108         .statfs         = ext4_statfs,
1109         .remount_fs     = ext4_remount,
1110         .clear_inode    = ext4_clear_inode,
1111         .show_options   = ext4_show_options,
1112 #ifdef CONFIG_QUOTA
1113         .quota_read     = ext4_quota_read,
1114         .quota_write    = ext4_quota_write,
1115 #endif
1116         .bdev_try_to_free_page = bdev_try_to_free_page,
1117 };
1118
1119 static const struct super_operations ext4_nojournal_sops = {
1120         .alloc_inode    = ext4_alloc_inode,
1121         .destroy_inode  = ext4_destroy_inode,
1122         .write_inode    = ext4_write_inode,
1123         .dirty_inode    = ext4_dirty_inode,
1124         .delete_inode   = ext4_delete_inode,
1125         .write_super    = ext4_write_super,
1126         .put_super      = ext4_put_super,
1127         .statfs         = ext4_statfs,
1128         .remount_fs     = ext4_remount,
1129         .clear_inode    = ext4_clear_inode,
1130         .show_options   = ext4_show_options,
1131 #ifdef CONFIG_QUOTA
1132         .quota_read     = ext4_quota_read,
1133         .quota_write    = ext4_quota_write,
1134 #endif
1135         .bdev_try_to_free_page = bdev_try_to_free_page,
1136 };
1137
1138 static const struct export_operations ext4_export_ops = {
1139         .fh_to_dentry = ext4_fh_to_dentry,
1140         .fh_to_parent = ext4_fh_to_parent,
1141         .get_parent = ext4_get_parent,
1142 };
1143
1144 enum {
1145         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1146         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1147         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1148         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1149         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1150         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1151         Opt_journal_update, Opt_journal_dev,
1152         Opt_journal_checksum, Opt_journal_async_commit,
1153         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1154         Opt_data_err_abort, Opt_data_err_ignore,
1155         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1156         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1157         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1158         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1159         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1160         Opt_block_validity, Opt_noblock_validity,
1161         Opt_inode_readahead_blks, Opt_journal_ioprio,
1162         Opt_dioread_nolock, Opt_dioread_lock,
1163         Opt_discard, Opt_nodiscard,
1164 };
1165
1166 static const match_table_t tokens = {
1167         {Opt_bsd_df, "bsddf"},
1168         {Opt_minix_df, "minixdf"},
1169         {Opt_grpid, "grpid"},
1170         {Opt_grpid, "bsdgroups"},
1171         {Opt_nogrpid, "nogrpid"},
1172         {Opt_nogrpid, "sysvgroups"},
1173         {Opt_resgid, "resgid=%u"},
1174         {Opt_resuid, "resuid=%u"},
1175         {Opt_sb, "sb=%u"},
1176         {Opt_err_cont, "errors=continue"},
1177         {Opt_err_panic, "errors=panic"},
1178         {Opt_err_ro, "errors=remount-ro"},
1179         {Opt_nouid32, "nouid32"},
1180         {Opt_debug, "debug"},
1181         {Opt_oldalloc, "oldalloc"},
1182         {Opt_orlov, "orlov"},
1183         {Opt_user_xattr, "user_xattr"},
1184         {Opt_nouser_xattr, "nouser_xattr"},
1185         {Opt_acl, "acl"},
1186         {Opt_noacl, "noacl"},
1187         {Opt_noload, "noload"},
1188         {Opt_noload, "norecovery"},
1189         {Opt_nobh, "nobh"},
1190         {Opt_bh, "bh"},
1191         {Opt_commit, "commit=%u"},
1192         {Opt_min_batch_time, "min_batch_time=%u"},
1193         {Opt_max_batch_time, "max_batch_time=%u"},
1194         {Opt_journal_update, "journal=update"},
1195         {Opt_journal_dev, "journal_dev=%u"},
1196         {Opt_journal_checksum, "journal_checksum"},
1197         {Opt_journal_async_commit, "journal_async_commit"},
1198         {Opt_abort, "abort"},
1199         {Opt_data_journal, "data=journal"},
1200         {Opt_data_ordered, "data=ordered"},
1201         {Opt_data_writeback, "data=writeback"},
1202         {Opt_data_err_abort, "data_err=abort"},
1203         {Opt_data_err_ignore, "data_err=ignore"},
1204         {Opt_offusrjquota, "usrjquota="},
1205         {Opt_usrjquota, "usrjquota=%s"},
1206         {Opt_offgrpjquota, "grpjquota="},
1207         {Opt_grpjquota, "grpjquota=%s"},
1208         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1209         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1210         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1211         {Opt_grpquota, "grpquota"},
1212         {Opt_noquota, "noquota"},
1213         {Opt_quota, "quota"},
1214         {Opt_usrquota, "usrquota"},
1215         {Opt_barrier, "barrier=%u"},
1216         {Opt_barrier, "barrier"},
1217         {Opt_nobarrier, "nobarrier"},
1218         {Opt_i_version, "i_version"},
1219         {Opt_stripe, "stripe=%u"},
1220         {Opt_resize, "resize"},
1221         {Opt_delalloc, "delalloc"},
1222         {Opt_nodelalloc, "nodelalloc"},
1223         {Opt_block_validity, "block_validity"},
1224         {Opt_noblock_validity, "noblock_validity"},
1225         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1226         {Opt_journal_ioprio, "journal_ioprio=%u"},
1227         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1228         {Opt_auto_da_alloc, "auto_da_alloc"},
1229         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1230         {Opt_dioread_nolock, "dioread_nolock"},
1231         {Opt_dioread_lock, "dioread_lock"},
1232         {Opt_discard, "discard"},
1233         {Opt_nodiscard, "nodiscard"},
1234         {Opt_err, NULL},
1235 };
1236
1237 static ext4_fsblk_t get_sb_block(void **data)
1238 {
1239         ext4_fsblk_t    sb_block;
1240         char            *options = (char *) *data;
1241
1242         if (!options || strncmp(options, "sb=", 3) != 0)
1243                 return 1;       /* Default location */
1244
1245         options += 3;
1246         /* TODO: use simple_strtoll with >32bit ext4 */
1247         sb_block = simple_strtoul(options, &options, 0);
1248         if (*options && *options != ',') {
1249                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1250                        (char *) *data);
1251                 return 1;
1252         }
1253         if (*options == ',')
1254                 options++;
1255         *data = (void *) options;
1256
1257         return sb_block;
1258 }
1259
1260 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1261 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1262         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1263
1264 #ifdef CONFIG_QUOTA
1265 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1266 {
1267         struct ext4_sb_info *sbi = EXT4_SB(sb);
1268         char *qname;
1269
1270         if (sb_any_quota_loaded(sb) &&
1271                 !sbi->s_qf_names[qtype]) {
1272                 ext4_msg(sb, KERN_ERR,
1273                         "Cannot change journaled "
1274                         "quota options when quota turned on");
1275                 return 0;
1276         }
1277         qname = match_strdup(args);
1278         if (!qname) {
1279                 ext4_msg(sb, KERN_ERR,
1280                         "Not enough memory for storing quotafile name");
1281                 return 0;
1282         }
1283         if (sbi->s_qf_names[qtype] &&
1284                 strcmp(sbi->s_qf_names[qtype], qname)) {
1285                 ext4_msg(sb, KERN_ERR,
1286                         "%s quota file already specified", QTYPE2NAME(qtype));
1287                 kfree(qname);
1288                 return 0;
1289         }
1290         sbi->s_qf_names[qtype] = qname;
1291         if (strchr(sbi->s_qf_names[qtype], '/')) {
1292                 ext4_msg(sb, KERN_ERR,
1293                         "quotafile must be on filesystem root");
1294                 kfree(sbi->s_qf_names[qtype]);
1295                 sbi->s_qf_names[qtype] = NULL;
1296                 return 0;
1297         }
1298         set_opt(sbi->s_mount_opt, QUOTA);
1299         return 1;
1300 }
1301
1302 static int clear_qf_name(struct super_block *sb, int qtype)
1303 {
1304
1305         struct ext4_sb_info *sbi = EXT4_SB(sb);
1306
1307         if (sb_any_quota_loaded(sb) &&
1308                 sbi->s_qf_names[qtype]) {
1309                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1310                         " when quota turned on");
1311                 return 0;
1312         }
1313         /*
1314          * The space will be released later when all options are confirmed
1315          * to be correct
1316          */
1317         sbi->s_qf_names[qtype] = NULL;
1318         return 1;
1319 }
1320 #endif
1321
1322 static int parse_options(char *options, struct super_block *sb,
1323                          unsigned long *journal_devnum,
1324                          unsigned int *journal_ioprio,
1325                          ext4_fsblk_t *n_blocks_count, int is_remount)
1326 {
1327         struct ext4_sb_info *sbi = EXT4_SB(sb);
1328         char *p;
1329         substring_t args[MAX_OPT_ARGS];
1330         int data_opt = 0;
1331         int option;
1332 #ifdef CONFIG_QUOTA
1333         int qfmt;
1334 #endif
1335
1336         if (!options)
1337                 return 1;
1338
1339         while ((p = strsep(&options, ",")) != NULL) {
1340                 int token;
1341                 if (!*p)
1342                         continue;
1343
1344                 /*
1345                  * Initialize args struct so we know whether arg was
1346                  * found; some options take optional arguments.
1347                  */
1348                 args[0].to = args[0].from = 0;
1349                 token = match_token(p, tokens, args);
1350                 switch (token) {
1351                 case Opt_bsd_df:
1352                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1353                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1354                         break;
1355                 case Opt_minix_df:
1356                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1357                         set_opt(sbi->s_mount_opt, MINIX_DF);
1358
1359                         break;
1360                 case Opt_grpid:
1361                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1362                         set_opt(sbi->s_mount_opt, GRPID);
1363
1364                         break;
1365                 case Opt_nogrpid:
1366                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1367                         clear_opt(sbi->s_mount_opt, GRPID);
1368
1369                         break;
1370                 case Opt_resuid:
1371                         if (match_int(&args[0], &option))
1372                                 return 0;
1373                         sbi->s_resuid = option;
1374                         break;
1375                 case Opt_resgid:
1376                         if (match_int(&args[0], &option))
1377                                 return 0;
1378                         sbi->s_resgid = option;
1379                         break;
1380                 case Opt_sb:
1381                         /* handled by get_sb_block() instead of here */
1382                         /* *sb_block = match_int(&args[0]); */
1383                         break;
1384                 case Opt_err_panic:
1385                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1386                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1387                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1388                         break;
1389                 case Opt_err_ro:
1390                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1391                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1392                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1393                         break;
1394                 case Opt_err_cont:
1395                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1396                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1397                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1398                         break;
1399                 case Opt_nouid32:
1400                         set_opt(sbi->s_mount_opt, NO_UID32);
1401                         break;
1402                 case Opt_debug:
1403                         set_opt(sbi->s_mount_opt, DEBUG);
1404                         break;
1405                 case Opt_oldalloc:
1406                         set_opt(sbi->s_mount_opt, OLDALLOC);
1407                         break;
1408                 case Opt_orlov:
1409                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1410                         break;
1411 #ifdef CONFIG_EXT4_FS_XATTR
1412                 case Opt_user_xattr:
1413                         set_opt(sbi->s_mount_opt, XATTR_USER);
1414                         break;
1415                 case Opt_nouser_xattr:
1416                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1417                         break;
1418 #else
1419                 case Opt_user_xattr:
1420                 case Opt_nouser_xattr:
1421                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1422                         break;
1423 #endif
1424 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1425                 case Opt_acl:
1426                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1427                         break;
1428                 case Opt_noacl:
1429                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1430                         break;
1431 #else
1432                 case Opt_acl:
1433                 case Opt_noacl:
1434                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1435                         break;
1436 #endif
1437                 case Opt_journal_update:
1438                         /* @@@ FIXME */
1439                         /* Eventually we will want to be able to create
1440                            a journal file here.  For now, only allow the
1441                            user to specify an existing inode to be the
1442                            journal file. */
1443                         if (is_remount) {
1444                                 ext4_msg(sb, KERN_ERR,
1445                                          "Cannot specify journal on remount");
1446                                 return 0;
1447                         }
1448                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1449                         break;
1450                 case Opt_journal_dev:
1451                         if (is_remount) {
1452                                 ext4_msg(sb, KERN_ERR,
1453                                         "Cannot specify journal on remount");
1454                                 return 0;
1455                         }
1456                         if (match_int(&args[0], &option))
1457                                 return 0;
1458                         *journal_devnum = option;
1459                         break;
1460                 case Opt_journal_checksum:
1461                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1462                         break;
1463                 case Opt_journal_async_commit:
1464                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1465                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1466                         break;
1467                 case Opt_noload:
1468                         set_opt(sbi->s_mount_opt, NOLOAD);
1469                         break;
1470                 case Opt_commit:
1471                         if (match_int(&args[0], &option))
1472                                 return 0;
1473                         if (option < 0)
1474                                 return 0;
1475                         if (option == 0)
1476                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1477                         sbi->s_commit_interval = HZ * option;
1478                         break;
1479                 case Opt_max_batch_time:
1480                         if (match_int(&args[0], &option))
1481                                 return 0;
1482                         if (option < 0)
1483                                 return 0;
1484                         if (option == 0)
1485                                 option = EXT4_DEF_MAX_BATCH_TIME;
1486                         sbi->s_max_batch_time = option;
1487                         break;
1488                 case Opt_min_batch_time:
1489                         if (match_int(&args[0], &option))
1490                                 return 0;
1491                         if (option < 0)
1492                                 return 0;
1493                         sbi->s_min_batch_time = option;
1494                         break;
1495                 case Opt_data_journal:
1496                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1497                         goto datacheck;
1498                 case Opt_data_ordered:
1499                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1500                         goto datacheck;
1501                 case Opt_data_writeback:
1502                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1503                 datacheck:
1504                         if (is_remount) {
1505                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1506                                         ext4_msg(sb, KERN_ERR,
1507                                                 "Cannot change data mode on remount");
1508                                         return 0;
1509                                 }
1510                         } else {
1511                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1512                                 sbi->s_mount_opt |= data_opt;
1513                         }
1514                         break;
1515                 case Opt_data_err_abort:
1516                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1517                         break;
1518                 case Opt_data_err_ignore:
1519                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1520                         break;
1521 #ifdef CONFIG_QUOTA
1522                 case Opt_usrjquota:
1523                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1524                                 return 0;
1525                         break;
1526                 case Opt_grpjquota:
1527                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1528                                 return 0;
1529                         break;
1530                 case Opt_offusrjquota:
1531                         if (!clear_qf_name(sb, USRQUOTA))
1532                                 return 0;
1533                         break;
1534                 case Opt_offgrpjquota:
1535                         if (!clear_qf_name(sb, GRPQUOTA))
1536                                 return 0;
1537                         break;
1538
1539                 case Opt_jqfmt_vfsold:
1540                         qfmt = QFMT_VFS_OLD;
1541                         goto set_qf_format;
1542                 case Opt_jqfmt_vfsv0:
1543                         qfmt = QFMT_VFS_V0;
1544                         goto set_qf_format;
1545                 case Opt_jqfmt_vfsv1:
1546                         qfmt = QFMT_VFS_V1;
1547 set_qf_format:
1548                         if (sb_any_quota_loaded(sb) &&
1549                             sbi->s_jquota_fmt != qfmt) {
1550                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1551                                         "journaled quota options when "
1552                                         "quota turned on");
1553                                 return 0;
1554                         }
1555                         sbi->s_jquota_fmt = qfmt;
1556                         break;
1557                 case Opt_quota:
1558                 case Opt_usrquota:
1559                         set_opt(sbi->s_mount_opt, QUOTA);
1560                         set_opt(sbi->s_mount_opt, USRQUOTA);
1561                         break;
1562                 case Opt_grpquota:
1563                         set_opt(sbi->s_mount_opt, QUOTA);
1564                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1565                         break;
1566                 case Opt_noquota:
1567                         if (sb_any_quota_loaded(sb)) {
1568                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1569                                         "options when quota turned on");
1570                                 return 0;
1571                         }
1572                         clear_opt(sbi->s_mount_opt, QUOTA);
1573                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1574                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1575                         break;
1576 #else
1577                 case Opt_quota:
1578                 case Opt_usrquota:
1579                 case Opt_grpquota:
1580                         ext4_msg(sb, KERN_ERR,
1581                                 "quota options not supported");
1582                         break;
1583                 case Opt_usrjquota:
1584                 case Opt_grpjquota:
1585                 case Opt_offusrjquota:
1586                 case Opt_offgrpjquota:
1587                 case Opt_jqfmt_vfsold:
1588                 case Opt_jqfmt_vfsv0:
1589                 case Opt_jqfmt_vfsv1:
1590                         ext4_msg(sb, KERN_ERR,
1591                                 "journaled quota options not supported");
1592                         break;
1593                 case Opt_noquota:
1594                         break;
1595 #endif
1596                 case Opt_abort:
1597                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1598                         break;
1599                 case Opt_nobarrier:
1600                         clear_opt(sbi->s_mount_opt, BARRIER);
1601                         break;
1602                 case Opt_barrier:
1603                         if (args[0].from) {
1604                                 if (match_int(&args[0], &option))
1605                                         return 0;
1606                         } else
1607                                 option = 1;     /* No argument, default to 1 */
1608                         if (option)
1609                                 set_opt(sbi->s_mount_opt, BARRIER);
1610                         else
1611                                 clear_opt(sbi->s_mount_opt, BARRIER);
1612                         break;
1613                 case Opt_ignore:
1614                         break;
1615                 case Opt_resize:
1616                         if (!is_remount) {
1617                                 ext4_msg(sb, KERN_ERR,
1618                                         "resize option only available "
1619                                         "for remount");
1620                                 return 0;
1621                         }
1622                         if (match_int(&args[0], &option) != 0)
1623                                 return 0;
1624                         *n_blocks_count = option;
1625                         break;
1626                 case Opt_nobh:
1627                         set_opt(sbi->s_mount_opt, NOBH);
1628                         break;
1629                 case Opt_bh:
1630                         clear_opt(sbi->s_mount_opt, NOBH);
1631                         break;
1632                 case Opt_i_version:
1633                         set_opt(sbi->s_mount_opt, I_VERSION);
1634                         sb->s_flags |= MS_I_VERSION;
1635                         break;
1636                 case Opt_nodelalloc:
1637                         clear_opt(sbi->s_mount_opt, DELALLOC);
1638                         break;
1639                 case Opt_stripe:
1640                         if (match_int(&args[0], &option))
1641                                 return 0;
1642                         if (option < 0)
1643                                 return 0;
1644                         sbi->s_stripe = option;
1645                         break;
1646                 case Opt_delalloc:
1647                         set_opt(sbi->s_mount_opt, DELALLOC);
1648                         break;
1649                 case Opt_block_validity:
1650                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1651                         break;
1652                 case Opt_noblock_validity:
1653                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1654                         break;
1655                 case Opt_inode_readahead_blks:
1656                         if (match_int(&args[0], &option))
1657                                 return 0;
1658                         if (option < 0 || option > (1 << 30))
1659                                 return 0;
1660                         if (!is_power_of_2(option)) {
1661                                 ext4_msg(sb, KERN_ERR,
1662                                          "EXT4-fs: inode_readahead_blks"
1663                                          " must be a power of 2");
1664                                 return 0;
1665                         }
1666                         sbi->s_inode_readahead_blks = option;
1667                         break;
1668                 case Opt_journal_ioprio:
1669                         if (match_int(&args[0], &option))
1670                                 return 0;
1671                         if (option < 0 || option > 7)
1672                                 break;
1673                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1674                                                             option);
1675                         break;
1676                 case Opt_noauto_da_alloc:
1677                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1678                         break;
1679                 case Opt_auto_da_alloc:
1680                         if (args[0].from) {
1681                                 if (match_int(&args[0], &option))
1682                                         return 0;
1683                         } else
1684                                 option = 1;     /* No argument, default to 1 */
1685                         if (option)
1686                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1687                         else
1688                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1689                         break;
1690                 case Opt_discard:
1691                         set_opt(sbi->s_mount_opt, DISCARD);
1692                         break;
1693                 case Opt_nodiscard:
1694                         clear_opt(sbi->s_mount_opt, DISCARD);
1695                         break;
1696                 case Opt_dioread_nolock:
1697                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1698                         break;
1699                 case Opt_dioread_lock:
1700                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1701                         break;
1702                 default:
1703                         ext4_msg(sb, KERN_ERR,
1704                                "Unrecognized mount option \"%s\" "
1705                                "or missing value", p);
1706                         return 0;
1707                 }
1708         }
1709 #ifdef CONFIG_QUOTA
1710         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1711                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1712                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1713
1714                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1715                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1716
1717                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1718                         ext4_msg(sb, KERN_ERR, "old and new quota "
1719                                         "format mixing");
1720                         return 0;
1721                 }
1722
1723                 if (!sbi->s_jquota_fmt) {
1724                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1725                                         "not specified");
1726                         return 0;
1727                 }
1728         } else {
1729                 if (sbi->s_jquota_fmt) {
1730                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1731                                         "specified with no journaling "
1732                                         "enabled");
1733                         return 0;
1734                 }
1735         }
1736 #endif
1737         return 1;
1738 }
1739
1740 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1741                             int read_only)
1742 {
1743         struct ext4_sb_info *sbi = EXT4_SB(sb);
1744         int res = 0;
1745
1746         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1747                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1748                          "forcing read-only mode");
1749                 res = MS_RDONLY;
1750         }
1751         if (read_only)
1752                 return res;
1753         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1754                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1755                          "running e2fsck is recommended");
1756         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1757                 ext4_msg(sb, KERN_WARNING,
1758                          "warning: mounting fs with errors, "
1759                          "running e2fsck is recommended");
1760         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1761                  le16_to_cpu(es->s_mnt_count) >=
1762                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1763                 ext4_msg(sb, KERN_WARNING,
1764                          "warning: maximal mount count reached, "
1765                          "running e2fsck is recommended");
1766         else if (le32_to_cpu(es->s_checkinterval) &&
1767                 (le32_to_cpu(es->s_lastcheck) +
1768                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1769                 ext4_msg(sb, KERN_WARNING,
1770                          "warning: checktime reached, "
1771                          "running e2fsck is recommended");
1772         if (!sbi->s_journal)
1773                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1774         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1775                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1776         le16_add_cpu(&es->s_mnt_count, 1);
1777         es->s_mtime = cpu_to_le32(get_seconds());
1778         ext4_update_dynamic_rev(sb);
1779         if (sbi->s_journal)
1780                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1781
1782         ext4_commit_super(sb, 1);
1783         if (test_opt(sb, DEBUG))
1784                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1785                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1786                         sb->s_blocksize,
1787                         sbi->s_groups_count,
1788                         EXT4_BLOCKS_PER_GROUP(sb),
1789                         EXT4_INODES_PER_GROUP(sb),
1790                         sbi->s_mount_opt);
1791
1792         return res;
1793 }
1794
1795 static int ext4_fill_flex_info(struct super_block *sb)
1796 {
1797         struct ext4_sb_info *sbi = EXT4_SB(sb);
1798         struct ext4_group_desc *gdp = NULL;
1799         ext4_group_t flex_group_count;
1800         ext4_group_t flex_group;
1801         int groups_per_flex = 0;
1802         size_t size;
1803         int i;
1804
1805         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1806         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1807
1808         if (groups_per_flex < 2) {
1809                 sbi->s_log_groups_per_flex = 0;
1810                 return 1;
1811         }
1812
1813         /* We allocate both existing and potentially added groups */
1814         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1815                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1816                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1817         size = flex_group_count * sizeof(struct flex_groups);
1818         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1819         if (sbi->s_flex_groups == NULL) {
1820                 sbi->s_flex_groups = vmalloc(size);
1821                 if (sbi->s_flex_groups)
1822                         memset(sbi->s_flex_groups, 0, size);
1823         }
1824         if (sbi->s_flex_groups == NULL) {
1825                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1826                                 "%u flex groups", flex_group_count);
1827                 goto failed;
1828         }
1829
1830         for (i = 0; i < sbi->s_groups_count; i++) {
1831                 gdp = ext4_get_group_desc(sb, i, NULL);
1832
1833                 flex_group = ext4_flex_group(sbi, i);
1834                 atomic_add(ext4_free_inodes_count(sb, gdp),
1835                            &sbi->s_flex_groups[flex_group].free_inodes);
1836                 atomic_add(ext4_free_blks_count(sb, gdp),
1837                            &sbi->s_flex_groups[flex_group].free_blocks);
1838                 atomic_add(ext4_used_dirs_count(sb, gdp),
1839                            &sbi->s_flex_groups[flex_group].used_dirs);
1840         }
1841
1842         return 1;
1843 failed:
1844         return 0;
1845 }
1846
1847 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1848                             struct ext4_group_desc *gdp)
1849 {
1850         __u16 crc = 0;
1851
1852         if (sbi->s_es->s_feature_ro_compat &
1853             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1854                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1855                 __le32 le_group = cpu_to_le32(block_group);
1856
1857                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1858                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1859                 crc = crc16(crc, (__u8 *)gdp, offset);
1860                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1861                 /* for checksum of struct ext4_group_desc do the rest...*/
1862                 if ((sbi->s_es->s_feature_incompat &
1863                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1864                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1865                         crc = crc16(crc, (__u8 *)gdp + offset,
1866                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1867                                         offset);
1868         }
1869
1870         return cpu_to_le16(crc);
1871 }
1872
1873 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1874                                 struct ext4_group_desc *gdp)
1875 {
1876         if ((sbi->s_es->s_feature_ro_compat &
1877              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1878             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1879                 return 0;
1880
1881         return 1;
1882 }
1883
1884 /* Called at mount-time, super-block is locked */
1885 static int ext4_check_descriptors(struct super_block *sb)
1886 {
1887         struct ext4_sb_info *sbi = EXT4_SB(sb);
1888         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1889         ext4_fsblk_t last_block;
1890         ext4_fsblk_t block_bitmap;
1891         ext4_fsblk_t inode_bitmap;
1892         ext4_fsblk_t inode_table;
1893         int flexbg_flag = 0;
1894         ext4_group_t i;
1895
1896         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1897                 flexbg_flag = 1;
1898
1899         ext4_debug("Checking group descriptors");
1900
1901         for (i = 0; i < sbi->s_groups_count; i++) {
1902                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1903
1904                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1905                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1906                 else
1907                         last_block = first_block +
1908                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1909
1910                 block_bitmap = ext4_block_bitmap(sb, gdp);
1911                 if (block_bitmap < first_block || block_bitmap > last_block) {
1912                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1913                                "Block bitmap for group %u not in group "
1914                                "(block %llu)!", i, block_bitmap);
1915                         return 0;
1916                 }
1917                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1918                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1919                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1920                                "Inode bitmap for group %u not in group "
1921                                "(block %llu)!", i, inode_bitmap);
1922                         return 0;
1923                 }
1924                 inode_table = ext4_inode_table(sb, gdp);
1925                 if (inode_table < first_block ||
1926                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1927                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1928                                "Inode table for group %u not in group "
1929                                "(block %llu)!", i, inode_table);
1930                         return 0;
1931                 }
1932                 ext4_lock_group(sb, i);
1933                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1934                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1935                                  "Checksum for group %u failed (%u!=%u)",
1936                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1937                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1938                         if (!(sb->s_flags & MS_RDONLY)) {
1939                                 ext4_unlock_group(sb, i);
1940                                 return 0;
1941                         }
1942                 }
1943                 ext4_unlock_group(sb, i);
1944                 if (!flexbg_flag)
1945                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1946         }
1947
1948         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1949         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1950         return 1;
1951 }
1952
1953 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1954  * the superblock) which were deleted from all directories, but held open by
1955  * a process at the time of a crash.  We walk the list and try to delete these
1956  * inodes at recovery time (only with a read-write filesystem).
1957  *
1958  * In order to keep the orphan inode chain consistent during traversal (in
1959  * case of crash during recovery), we link each inode into the superblock
1960  * orphan list_head and handle it the same way as an inode deletion during
1961  * normal operation (which journals the operations for us).
1962  *
1963  * We only do an iget() and an iput() on each inode, which is very safe if we
1964  * accidentally point at an in-use or already deleted inode.  The worst that
1965  * can happen in this case is that we get a "bit already cleared" message from
1966  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1967  * e2fsck was run on this filesystem, and it must have already done the orphan
1968  * inode cleanup for us, so we can safely abort without any further action.
1969  */
1970 static void ext4_orphan_cleanup(struct super_block *sb,
1971                                 struct ext4_super_block *es)
1972 {
1973         unsigned int s_flags = sb->s_flags;
1974         int nr_orphans = 0, nr_truncates = 0;
1975 #ifdef CONFIG_QUOTA
1976         int i;
1977 #endif
1978         if (!es->s_last_orphan) {
1979                 jbd_debug(4, "no orphan inodes to clean up\n");
1980                 return;
1981         }
1982
1983         if (bdev_read_only(sb->s_bdev)) {
1984                 ext4_msg(sb, KERN_ERR, "write access "
1985                         "unavailable, skipping orphan cleanup");
1986                 return;
1987         }
1988
1989         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1990                 if (es->s_last_orphan)
1991                         jbd_debug(1, "Errors on filesystem, "
1992                                   "clearing orphan list.\n");
1993                 es->s_last_orphan = 0;
1994                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1995                 return;
1996         }
1997
1998         if (s_flags & MS_RDONLY) {
1999                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2000                 sb->s_flags &= ~MS_RDONLY;
2001         }
2002 #ifdef CONFIG_QUOTA
2003         /* Needed for iput() to work correctly and not trash data */
2004         sb->s_flags |= MS_ACTIVE;
2005         /* Turn on quotas so that they are updated correctly */
2006         for (i = 0; i < MAXQUOTAS; i++) {
2007                 if (EXT4_SB(sb)->s_qf_names[i]) {
2008                         int ret = ext4_quota_on_mount(sb, i);
2009                         if (ret < 0)
2010                                 ext4_msg(sb, KERN_ERR,
2011                                         "Cannot turn on journaled "
2012                                         "quota: error %d", ret);
2013                 }
2014         }
2015 #endif
2016
2017         while (es->s_last_orphan) {
2018                 struct inode *inode;
2019
2020                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2021                 if (IS_ERR(inode)) {
2022                         es->s_last_orphan = 0;
2023                         break;
2024                 }
2025
2026                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2027                 dquot_initialize(inode);
2028                 if (inode->i_nlink) {
2029                         ext4_msg(sb, KERN_DEBUG,
2030                                 "%s: truncating inode %lu to %lld bytes",
2031                                 __func__, inode->i_ino, inode->i_size);
2032                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2033                                   inode->i_ino, inode->i_size);
2034                         ext4_truncate(inode);
2035                         nr_truncates++;
2036                 } else {
2037                         ext4_msg(sb, KERN_DEBUG,
2038                                 "%s: deleting unreferenced inode %lu",
2039                                 __func__, inode->i_ino);
2040                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2041                                   inode->i_ino);
2042                         nr_orphans++;
2043                 }
2044                 iput(inode);  /* The delete magic happens here! */
2045         }
2046
2047 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2048
2049         if (nr_orphans)
2050                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2051                        PLURAL(nr_orphans));
2052         if (nr_truncates)
2053                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2054                        PLURAL(nr_truncates));
2055 #ifdef CONFIG_QUOTA
2056         /* Turn quotas off */
2057         for (i = 0; i < MAXQUOTAS; i++) {
2058                 if (sb_dqopt(sb)->files[i])
2059                         dquot_quota_off(sb, i);
2060         }
2061 #endif
2062         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2063 }
2064
2065 /*
2066  * Maximal extent format file size.
2067  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2068  * extent format containers, within a sector_t, and within i_blocks
2069  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2070  * so that won't be a limiting factor.
2071  *
2072  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2073  */
2074 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2075 {
2076         loff_t res;
2077         loff_t upper_limit = MAX_LFS_FILESIZE;
2078
2079         /* small i_blocks in vfs inode? */
2080         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2081                 /*
2082                  * CONFIG_LBDAF is not enabled implies the inode
2083                  * i_block represent total blocks in 512 bytes
2084                  * 32 == size of vfs inode i_blocks * 8
2085                  */
2086                 upper_limit = (1LL << 32) - 1;
2087
2088                 /* total blocks in file system block size */
2089                 upper_limit >>= (blkbits - 9);
2090                 upper_limit <<= blkbits;
2091         }
2092
2093         /* 32-bit extent-start container, ee_block */
2094         res = 1LL << 32;
2095         res <<= blkbits;
2096         res -= 1;
2097
2098         /* Sanity check against vm- & vfs- imposed limits */
2099         if (res > upper_limit)
2100                 res = upper_limit;
2101
2102         return res;
2103 }
2104
2105 /*
2106  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2107  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2108  * We need to be 1 filesystem block less than the 2^48 sector limit.
2109  */
2110 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2111 {
2112         loff_t res = EXT4_NDIR_BLOCKS;
2113         int meta_blocks;
2114         loff_t upper_limit;
2115         /* This is calculated to be the largest file size for a dense, block
2116          * mapped file such that the file's total number of 512-byte sectors,
2117          * including data and all indirect blocks, does not exceed (2^48 - 1).
2118          *
2119          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2120          * number of 512-byte sectors of the file.
2121          */
2122
2123         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2124                 /*
2125                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2126                  * the inode i_block field represents total file blocks in
2127                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2128                  */
2129                 upper_limit = (1LL << 32) - 1;
2130
2131                 /* total blocks in file system block size */
2132                 upper_limit >>= (bits - 9);
2133
2134         } else {
2135                 /*
2136                  * We use 48 bit ext4_inode i_blocks
2137                  * With EXT4_HUGE_FILE_FL set the i_blocks
2138                  * represent total number of blocks in
2139                  * file system block size
2140                  */
2141                 upper_limit = (1LL << 48) - 1;
2142
2143         }
2144
2145         /* indirect blocks */
2146         meta_blocks = 1;
2147         /* double indirect blocks */
2148         meta_blocks += 1 + (1LL << (bits-2));
2149         /* tripple indirect blocks */
2150         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2151
2152         upper_limit -= meta_blocks;
2153         upper_limit <<= bits;
2154
2155         res += 1LL << (bits-2);
2156         res += 1LL << (2*(bits-2));
2157         res += 1LL << (3*(bits-2));
2158         res <<= bits;
2159         if (res > upper_limit)
2160                 res = upper_limit;
2161
2162         if (res > MAX_LFS_FILESIZE)
2163                 res = MAX_LFS_FILESIZE;
2164
2165         return res;
2166 }
2167
2168 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2169                                    ext4_fsblk_t logical_sb_block, int nr)
2170 {
2171         struct ext4_sb_info *sbi = EXT4_SB(sb);
2172         ext4_group_t bg, first_meta_bg;
2173         int has_super = 0;
2174
2175         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2176
2177         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2178             nr < first_meta_bg)
2179                 return logical_sb_block + nr + 1;
2180         bg = sbi->s_desc_per_block * nr;
2181         if (ext4_bg_has_super(sb, bg))
2182                 has_super = 1;
2183
2184         return (has_super + ext4_group_first_block_no(sb, bg));
2185 }
2186
2187 /**
2188  * ext4_get_stripe_size: Get the stripe size.
2189  * @sbi: In memory super block info
2190  *
2191  * If we have specified it via mount option, then
2192  * use the mount option value. If the value specified at mount time is
2193  * greater than the blocks per group use the super block value.
2194  * If the super block value is greater than blocks per group return 0.
2195  * Allocator needs it be less than blocks per group.
2196  *
2197  */
2198 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2199 {
2200         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2201         unsigned long stripe_width =
2202                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2203
2204         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2205                 return sbi->s_stripe;
2206
2207         if (stripe_width <= sbi->s_blocks_per_group)
2208                 return stripe_width;
2209
2210         if (stride <= sbi->s_blocks_per_group)
2211                 return stride;
2212
2213         return 0;
2214 }
2215
2216 /* sysfs supprt */
2217
2218 struct ext4_attr {
2219         struct attribute attr;
2220         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2221         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2222                          const char *, size_t);
2223         int offset;
2224 };
2225
2226 static int parse_strtoul(const char *buf,
2227                 unsigned long max, unsigned long *value)
2228 {
2229         char *endp;
2230
2231         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2232         endp = skip_spaces(endp);
2233         if (*endp || *value > max)
2234                 return -EINVAL;
2235
2236         return 0;
2237 }
2238
2239 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2240                                               struct ext4_sb_info *sbi,
2241                                               char *buf)
2242 {
2243         return snprintf(buf, PAGE_SIZE, "%llu\n",
2244                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2245 }
2246
2247 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2248                                          struct ext4_sb_info *sbi, char *buf)
2249 {
2250         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2251
2252         return snprintf(buf, PAGE_SIZE, "%lu\n",
2253                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2254                          sbi->s_sectors_written_start) >> 1);
2255 }
2256
2257 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2258                                           struct ext4_sb_info *sbi, char *buf)
2259 {
2260         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2261
2262         return snprintf(buf, PAGE_SIZE, "%llu\n",
2263                         (unsigned long long)(sbi->s_kbytes_written +
2264                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2265                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2266 }
2267
2268 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2269                                           struct ext4_sb_info *sbi,
2270                                           const char *buf, size_t count)
2271 {
2272         unsigned long t;
2273
2274         if (parse_strtoul(buf, 0x40000000, &t))
2275                 return -EINVAL;
2276
2277         if (!is_power_of_2(t))
2278                 return -EINVAL;
2279
2280         sbi->s_inode_readahead_blks = t;
2281         return count;
2282 }
2283
2284 static ssize_t sbi_ui_show(struct ext4_attr *a,
2285                            struct ext4_sb_info *sbi, char *buf)
2286 {
2287         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2288
2289         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2290 }
2291
2292 static ssize_t sbi_ui_store(struct ext4_attr *a,
2293                             struct ext4_sb_info *sbi,
2294                             const char *buf, size_t count)
2295 {
2296         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2297         unsigned long t;
2298
2299         if (parse_strtoul(buf, 0xffffffff, &t))
2300                 return -EINVAL;
2301         *ui = t;
2302         return count;
2303 }
2304
2305 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2306 static struct ext4_attr ext4_attr_##_name = {                   \
2307         .attr = {.name = __stringify(_name), .mode = _mode },   \
2308         .show   = _show,                                        \
2309         .store  = _store,                                       \
2310         .offset = offsetof(struct ext4_sb_info, _elname),       \
2311 }
2312 #define EXT4_ATTR(name, mode, show, store) \
2313 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2314
2315 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2316 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2317 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2318         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2319 #define ATTR_LIST(name) &ext4_attr_##name.attr
2320
2321 EXT4_RO_ATTR(delayed_allocation_blocks);
2322 EXT4_RO_ATTR(session_write_kbytes);
2323 EXT4_RO_ATTR(lifetime_write_kbytes);
2324 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2325                  inode_readahead_blks_store, s_inode_readahead_blks);
2326 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2327 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2328 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2329 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2330 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2331 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2332 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2333 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2334
2335 static struct attribute *ext4_attrs[] = {
2336         ATTR_LIST(delayed_allocation_blocks),
2337         ATTR_LIST(session_write_kbytes),
2338         ATTR_LIST(lifetime_write_kbytes),
2339         ATTR_LIST(inode_readahead_blks),
2340         ATTR_LIST(inode_goal),
2341         ATTR_LIST(mb_stats),
2342         ATTR_LIST(mb_max_to_scan),
2343         ATTR_LIST(mb_min_to_scan),
2344         ATTR_LIST(mb_order2_req),
2345         ATTR_LIST(mb_stream_req),
2346         ATTR_LIST(mb_group_prealloc),
2347         ATTR_LIST(max_writeback_mb_bump),
2348         NULL,
2349 };
2350
2351 static ssize_t ext4_attr_show(struct kobject *kobj,
2352                               struct attribute *attr, char *buf)
2353 {
2354         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2355                                                 s_kobj);
2356         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2357
2358         return a->show ? a->show(a, sbi, buf) : 0;
2359 }
2360
2361 static ssize_t ext4_attr_store(struct kobject *kobj,
2362                                struct attribute *attr,
2363                                const char *buf, size_t len)
2364 {
2365         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2366                                                 s_kobj);
2367         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2368
2369         return a->store ? a->store(a, sbi, buf, len) : 0;
2370 }
2371
2372 static void ext4_sb_release(struct kobject *kobj)
2373 {
2374         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2375                                                 s_kobj);
2376         complete(&sbi->s_kobj_unregister);
2377 }
2378
2379
2380 static const struct sysfs_ops ext4_attr_ops = {
2381         .show   = ext4_attr_show,
2382         .store  = ext4_attr_store,
2383 };
2384
2385 static struct kobj_type ext4_ktype = {
2386         .default_attrs  = ext4_attrs,
2387         .sysfs_ops      = &ext4_attr_ops,
2388         .release        = ext4_sb_release,
2389 };
2390
2391 /*
2392  * Check whether this filesystem can be mounted based on
2393  * the features present and the RDONLY/RDWR mount requested.
2394  * Returns 1 if this filesystem can be mounted as requested,
2395  * 0 if it cannot be.
2396  */
2397 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2398 {
2399         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2400                 ext4_msg(sb, KERN_ERR,
2401                         "Couldn't mount because of "
2402                         "unsupported optional features (%x)",
2403                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2404                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2405                 return 0;
2406         }
2407
2408         if (readonly)
2409                 return 1;
2410
2411         /* Check that feature set is OK for a read-write mount */
2412         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2413                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2414                          "unsupported optional features (%x)",
2415                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2416                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2417                 return 0;
2418         }
2419         /*
2420          * Large file size enabled file system can only be mounted
2421          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2422          */
2423         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2424                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2425                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2426                                  "cannot be mounted RDWR without "
2427                                  "CONFIG_LBDAF");
2428                         return 0;
2429                 }
2430         }
2431         return 1;
2432 }
2433
2434 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2435                                 __releases(kernel_lock)
2436                                 __acquires(kernel_lock)
2437 {
2438         char *orig_data = kstrdup(data, GFP_KERNEL);
2439         struct buffer_head *bh;
2440         struct ext4_super_block *es = NULL;
2441         struct ext4_sb_info *sbi;
2442         ext4_fsblk_t block;
2443         ext4_fsblk_t sb_block = get_sb_block(&data);
2444         ext4_fsblk_t logical_sb_block;
2445         unsigned long offset = 0;
2446         unsigned long journal_devnum = 0;
2447         unsigned long def_mount_opts;
2448         struct inode *root;
2449         char *cp;
2450         const char *descr;
2451         int ret = -EINVAL;
2452         int blocksize;
2453         unsigned int db_count;
2454         unsigned int i;
2455         int needs_recovery, has_huge_files;
2456         __u64 blocks_count;
2457         int err;
2458         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2459
2460         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2461         if (!sbi)
2462                 return -ENOMEM;
2463
2464         sbi->s_blockgroup_lock =
2465                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2466         if (!sbi->s_blockgroup_lock) {
2467                 kfree(sbi);
2468                 return -ENOMEM;
2469         }
2470         sb->s_fs_info = sbi;
2471         sbi->s_mount_opt = 0;
2472         sbi->s_resuid = EXT4_DEF_RESUID;
2473         sbi->s_resgid = EXT4_DEF_RESGID;
2474         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2475         sbi->s_sb_block = sb_block;
2476         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2477                                                       sectors[1]);
2478
2479         unlock_kernel();
2480
2481         /* Cleanup superblock name */
2482         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2483                 *cp = '!';
2484
2485         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2486         if (!blocksize) {
2487                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2488                 goto out_fail;
2489         }
2490
2491         /*
2492          * The ext4 superblock will not be buffer aligned for other than 1kB
2493          * block sizes.  We need to calculate the offset from buffer start.
2494          */
2495         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2496                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2497                 offset = do_div(logical_sb_block, blocksize);
2498         } else {
2499                 logical_sb_block = sb_block;
2500         }
2501
2502         if (!(bh = sb_bread(sb, logical_sb_block))) {
2503                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2504                 goto out_fail;
2505         }
2506         /*
2507          * Note: s_es must be initialized as soon as possible because
2508          *       some ext4 macro-instructions depend on its value
2509          */
2510         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2511         sbi->s_es = es;
2512         sb->s_magic = le16_to_cpu(es->s_magic);
2513         if (sb->s_magic != EXT4_SUPER_MAGIC)
2514                 goto cantfind_ext4;
2515         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2516
2517         /* Set defaults before we parse the mount options */
2518         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2519         if (def_mount_opts & EXT4_DEFM_DEBUG)
2520                 set_opt(sbi->s_mount_opt, DEBUG);
2521         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
2522                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
2523                         "2.6.38");
2524                 set_opt(sbi->s_mount_opt, GRPID);
2525         }
2526         if (def_mount_opts & EXT4_DEFM_UID16)
2527                 set_opt(sbi->s_mount_opt, NO_UID32);
2528 #ifdef CONFIG_EXT4_FS_XATTR
2529         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2530                 set_opt(sbi->s_mount_opt, XATTR_USER);
2531 #endif
2532 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2533         if (def_mount_opts & EXT4_DEFM_ACL)
2534                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2535 #endif
2536         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2537                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2538         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2539                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
2540         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2541                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2542
2543         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2544                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2545         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2546                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2547         else
2548                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2549
2550         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2551         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2552         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2553         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2554         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2555
2556         set_opt(sbi->s_mount_opt, BARRIER);
2557
2558         /*
2559          * enable delayed allocation by default
2560          * Use -o nodelalloc to turn it off
2561          */
2562         if (!IS_EXT3_SB(sb))
2563                 set_opt(sbi->s_mount_opt, DELALLOC);
2564
2565         if (!parse_options((char *) data, sb, &journal_devnum,
2566                            &journal_ioprio, NULL, 0))
2567                 goto failed_mount;
2568
2569         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2570                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2571
2572         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2573             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2574              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2575              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2576                 ext4_msg(sb, KERN_WARNING,
2577                        "feature flags set on rev 0 fs, "
2578                        "running e2fsck is recommended");
2579
2580         /*
2581          * Check feature flags regardless of the revision level, since we
2582          * previously didn't change the revision level when setting the flags,
2583          * so there is a chance incompat flags are set on a rev 0 filesystem.
2584          */
2585         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2586                 goto failed_mount;
2587
2588         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2589
2590         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2591             blocksize > EXT4_MAX_BLOCK_SIZE) {
2592                 ext4_msg(sb, KERN_ERR,
2593                        "Unsupported filesystem blocksize %d", blocksize);
2594                 goto failed_mount;
2595         }
2596
2597         if (sb->s_blocksize != blocksize) {
2598                 /* Validate the filesystem blocksize */
2599                 if (!sb_set_blocksize(sb, blocksize)) {
2600                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2601                                         blocksize);
2602                         goto failed_mount;
2603                 }
2604
2605                 brelse(bh);
2606                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2607                 offset = do_div(logical_sb_block, blocksize);
2608                 bh = sb_bread(sb, logical_sb_block);
2609                 if (!bh) {
2610                         ext4_msg(sb, KERN_ERR,
2611                                "Can't read superblock on 2nd try");
2612                         goto failed_mount;
2613                 }
2614                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2615                 sbi->s_es = es;
2616                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2617                         ext4_msg(sb, KERN_ERR,
2618                                "Magic mismatch, very weird!");
2619                         goto failed_mount;
2620                 }
2621         }
2622
2623         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2624                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2625         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2626                                                       has_huge_files);
2627         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2628
2629         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2630                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2631                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2632         } else {
2633                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2634                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2635                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2636                     (!is_power_of_2(sbi->s_inode_size)) ||
2637                     (sbi->s_inode_size > blocksize)) {
2638                         ext4_msg(sb, KERN_ERR,
2639                                "unsupported inode size: %d",
2640                                sbi->s_inode_size);
2641                         goto failed_mount;
2642                 }
2643                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2644                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2645         }
2646
2647         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2648         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2649                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2650                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2651                     !is_power_of_2(sbi->s_desc_size)) {
2652                         ext4_msg(sb, KERN_ERR,
2653                                "unsupported descriptor size %lu",
2654                                sbi->s_desc_size);
2655                         goto failed_mount;
2656                 }
2657         } else
2658                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2659
2660         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2661         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2662         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2663                 goto cantfind_ext4;
2664
2665         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2666         if (sbi->s_inodes_per_block == 0)
2667                 goto cantfind_ext4;
2668         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2669                                         sbi->s_inodes_per_block;
2670         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2671         sbi->s_sbh = bh;
2672         sbi->s_mount_state = le16_to_cpu(es->s_state);
2673         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2674         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2675
2676         for (i = 0; i < 4; i++)
2677                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2678         sbi->s_def_hash_version = es->s_def_hash_version;
2679         i = le32_to_cpu(es->s_flags);
2680         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2681                 sbi->s_hash_unsigned = 3;
2682         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2683 #ifdef __CHAR_UNSIGNED__
2684                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2685                 sbi->s_hash_unsigned = 3;
2686 #else
2687                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2688 #endif
2689                 sb->s_dirt = 1;
2690         }
2691
2692         if (sbi->s_blocks_per_group > blocksize * 8) {
2693                 ext4_msg(sb, KERN_ERR,
2694                        "#blocks per group too big: %lu",
2695                        sbi->s_blocks_per_group);
2696                 goto failed_mount;
2697         }
2698         if (sbi->s_inodes_per_group > blocksize * 8) {
2699                 ext4_msg(sb, KERN_ERR,
2700                        "#inodes per group too big: %lu",
2701                        sbi->s_inodes_per_group);
2702                 goto failed_mount;
2703         }
2704
2705         /*
2706          * Test whether we have more sectors than will fit in sector_t,
2707          * and whether the max offset is addressable by the page cache.
2708          */
2709         if ((ext4_blocks_count(es) >
2710              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2711             (ext4_blocks_count(es) >
2712              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2713                 ext4_msg(sb, KERN_ERR, "filesystem"
2714                          " too large to mount safely on this system");
2715                 if (sizeof(sector_t) < 8)
2716                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2717                 ret = -EFBIG;
2718                 goto failed_mount;
2719         }
2720
2721         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2722                 goto cantfind_ext4;
2723
2724         /* check blocks count against device size */
2725         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2726         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2727                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2728                        "exceeds size of device (%llu blocks)",
2729                        ext4_blocks_count(es), blocks_count);
2730                 goto failed_mount;
2731         }
2732
2733         /*
2734          * It makes no sense for the first data block to be beyond the end
2735          * of the filesystem.
2736          */
2737         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2738                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2739                          "block %u is beyond end of filesystem (%llu)",
2740                          le32_to_cpu(es->s_first_data_block),
2741                          ext4_blocks_count(es));
2742                 goto failed_mount;
2743         }
2744         blocks_count = (ext4_blocks_count(es) -
2745                         le32_to_cpu(es->s_first_data_block) +
2746                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2747         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2748         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2749                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2750                        "(block count %llu, first data block %u, "
2751                        "blocks per group %lu)", sbi->s_groups_count,
2752                        ext4_blocks_count(es),
2753                        le32_to_cpu(es->s_first_data_block),
2754                        EXT4_BLOCKS_PER_GROUP(sb));
2755                 goto failed_mount;
2756         }
2757         sbi->s_groups_count = blocks_count;
2758         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2759                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2760         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2761                    EXT4_DESC_PER_BLOCK(sb);
2762         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2763                                     GFP_KERNEL);
2764         if (sbi->s_group_desc == NULL) {
2765                 ext4_msg(sb, KERN_ERR, "not enough memory");
2766                 goto failed_mount;
2767         }
2768
2769 #ifdef CONFIG_PROC_FS
2770         if (ext4_proc_root)
2771                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2772 #endif
2773
2774         bgl_lock_init(sbi->s_blockgroup_lock);
2775
2776         for (i = 0; i < db_count; i++) {
2777                 block = descriptor_loc(sb, logical_sb_block, i);
2778                 sbi->s_group_desc[i] = sb_bread(sb, block);
2779                 if (!sbi->s_group_desc[i]) {
2780                         ext4_msg(sb, KERN_ERR,
2781                                "can't read group descriptor %d", i);
2782                         db_count = i;
2783                         goto failed_mount2;
2784                 }
2785         }
2786         if (!ext4_check_descriptors(sb)) {
2787                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2788                 goto failed_mount2;
2789         }
2790         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2791                 if (!ext4_fill_flex_info(sb)) {
2792                         ext4_msg(sb, KERN_ERR,
2793                                "unable to initialize "
2794                                "flex_bg meta info!");
2795                         goto failed_mount2;
2796                 }
2797
2798         sbi->s_gdb_count = db_count;
2799         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2800         spin_lock_init(&sbi->s_next_gen_lock);
2801
2802         sbi->s_stripe = ext4_get_stripe_size(sbi);
2803         sbi->s_max_writeback_mb_bump = 128;
2804
2805         /*
2806          * set up enough so that it can read an inode
2807          */
2808         if (!test_opt(sb, NOLOAD) &&
2809             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2810                 sb->s_op = &ext4_sops;
2811         else
2812                 sb->s_op = &ext4_nojournal_sops;
2813         sb->s_export_op = &ext4_export_ops;
2814         sb->s_xattr = ext4_xattr_handlers;
2815 #ifdef CONFIG_QUOTA
2816         sb->s_qcop = &ext4_qctl_operations;
2817         sb->dq_op = &ext4_quota_operations;
2818 #endif
2819         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2820         mutex_init(&sbi->s_orphan_lock);
2821         mutex_init(&sbi->s_resize_lock);
2822
2823         sb->s_root = NULL;
2824
2825         needs_recovery = (es->s_last_orphan != 0 ||
2826                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2827                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2828
2829         /*
2830          * The first inode we look at is the journal inode.  Don't try
2831          * root first: it may be modified in the journal!
2832          */
2833         if (!test_opt(sb, NOLOAD) &&
2834             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2835                 if (ext4_load_journal(sb, es, journal_devnum))
2836                         goto failed_mount3;
2837         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2838               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2839                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2840                        "suppressed and not mounted read-only");
2841                 goto failed_mount_wq;
2842         } else {
2843                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2844                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2845                 sbi->s_journal = NULL;
2846                 needs_recovery = 0;
2847                 goto no_journal;
2848         }
2849
2850         if (ext4_blocks_count(es) > 0xffffffffULL &&
2851             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2852                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2853                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2854                 goto failed_mount_wq;
2855         }
2856
2857         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2858                 jbd2_journal_set_features(sbi->s_journal,
2859                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2860                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2861         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2862                 jbd2_journal_set_features(sbi->s_journal,
2863                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2864                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2865                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2866         } else {
2867                 jbd2_journal_clear_features(sbi->s_journal,
2868                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2869                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2870         }
2871
2872         /* We have now updated the journal if required, so we can
2873          * validate the data journaling mode. */
2874         switch (test_opt(sb, DATA_FLAGS)) {
2875         case 0:
2876                 /* No mode set, assume a default based on the journal
2877                  * capabilities: ORDERED_DATA if the journal can
2878                  * cope, else JOURNAL_DATA
2879                  */
2880                 if (jbd2_journal_check_available_features
2881                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2882                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2883                 else
2884                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2885                 break;
2886
2887         case EXT4_MOUNT_ORDERED_DATA:
2888         case EXT4_MOUNT_WRITEBACK_DATA:
2889                 if (!jbd2_journal_check_available_features
2890                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2891                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2892                                "requested data journaling mode");
2893                         goto failed_mount_wq;
2894                 }
2895         default:
2896                 break;
2897         }
2898         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2899
2900 no_journal:
2901         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2902                                   ext4_count_free_blocks(sb));
2903         if (!err)
2904                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2905                                           ext4_count_free_inodes(sb));
2906         if (!err)
2907                 err = percpu_counter_init(&sbi->s_dirs_counter,
2908                                           ext4_count_dirs(sb));
2909         if (!err)
2910                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2911         if (err) {
2912                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2913                 goto failed_mount_wq;
2914         }
2915         if (test_opt(sb, NOBH)) {
2916                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2917                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2918                                 "its supported only with writeback mode");
2919                         clear_opt(sbi->s_mount_opt, NOBH);
2920                 }
2921                 if (test_opt(sb, DIOREAD_NOLOCK)) {
2922                         ext4_msg(sb, KERN_WARNING, "dioread_nolock option is "
2923                                 "not supported with nobh mode");
2924                         goto failed_mount_wq;
2925                 }
2926         }
2927         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2928         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2929                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2930                 goto failed_mount_wq;
2931         }
2932
2933         /*
2934          * The jbd2_journal_load will have done any necessary log recovery,
2935          * so we can safely mount the rest of the filesystem now.
2936          */
2937
2938         root = ext4_iget(sb, EXT4_ROOT_INO);
2939         if (IS_ERR(root)) {
2940                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2941                 ret = PTR_ERR(root);
2942                 goto failed_mount4;
2943         }
2944         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2945                 iput(root);
2946                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2947                 goto failed_mount4;
2948         }
2949         sb->s_root = d_alloc_root(root);
2950         if (!sb->s_root) {
2951                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2952                 iput(root);
2953                 ret = -ENOMEM;
2954                 goto failed_mount4;
2955         }
2956
2957         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2958
2959         /* determine the minimum size of new large inodes, if present */
2960         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2961                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2962                                                      EXT4_GOOD_OLD_INODE_SIZE;
2963                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2964                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2965                         if (sbi->s_want_extra_isize <
2966                             le16_to_cpu(es->s_want_extra_isize))
2967                                 sbi->s_want_extra_isize =
2968                                         le16_to_cpu(es->s_want_extra_isize);
2969                         if (sbi->s_want_extra_isize <
2970                             le16_to_cpu(es->s_min_extra_isize))
2971                                 sbi->s_want_extra_isize =
2972                                         le16_to_cpu(es->s_min_extra_isize);
2973                 }
2974         }
2975         /* Check if enough inode space is available */
2976         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2977                                                         sbi->s_inode_size) {
2978                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2979                                                        EXT4_GOOD_OLD_INODE_SIZE;
2980                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2981                          "available");
2982         }
2983
2984         if (test_opt(sb, DELALLOC) &&
2985             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2986                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2987                          "requested data journaling mode");
2988                 clear_opt(sbi->s_mount_opt, DELALLOC);
2989         }
2990         if (test_opt(sb, DIOREAD_NOLOCK)) {
2991                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2992                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2993                                 "option - requested data journaling mode");
2994                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2995                 }
2996                 if (sb->s_blocksize < PAGE_SIZE) {
2997                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2998                                 "option - block size is too small");
2999                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
3000                 }
3001         }
3002
3003         err = ext4_setup_system_zone(sb);
3004         if (err) {
3005                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3006                          "zone (%d)", err);
3007                 goto failed_mount4;
3008         }
3009
3010         ext4_ext_init(sb);
3011         err = ext4_mb_init(sb, needs_recovery);
3012         if (err) {
3013                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
3014                          err);
3015                 goto failed_mount4;
3016         }
3017
3018         sbi->s_kobj.kset = ext4_kset;
3019         init_completion(&sbi->s_kobj_unregister);
3020         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3021                                    "%s", sb->s_id);
3022         if (err) {
3023                 ext4_mb_release(sb);
3024                 ext4_ext_release(sb);
3025                 goto failed_mount4;
3026         };
3027
3028         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3029         ext4_orphan_cleanup(sb, es);
3030         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3031         if (needs_recovery) {
3032                 ext4_msg(sb, KERN_INFO, "recovery complete");
3033                 ext4_mark_recovery_complete(sb, es);
3034         }
3035         if (EXT4_SB(sb)->s_journal) {
3036                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3037                         descr = " journalled data mode";
3038                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3039                         descr = " ordered data mode";
3040                 else
3041                         descr = " writeback data mode";
3042         } else
3043                 descr = "out journal";
3044
3045         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3046                 "Opts: %s", descr, orig_data);
3047
3048         lock_kernel();
3049         kfree(orig_data);
3050         return 0;
3051
3052 cantfind_ext4:
3053         if (!silent)
3054                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3055         goto failed_mount;
3056
3057 failed_mount4:
3058         ext4_msg(sb, KERN_ERR, "mount failed");
3059         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3060 failed_mount_wq:
3061         ext4_release_system_zone(sb);
3062         if (sbi->s_journal) {
3063                 jbd2_journal_destroy(sbi->s_journal);
3064                 sbi->s_journal = NULL;
3065         }
3066         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3067         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3068         percpu_counter_destroy(&sbi->s_dirs_counter);
3069         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3070 failed_mount3:
3071         if (sbi->s_flex_groups) {
3072                 if (is_vmalloc_addr(sbi->s_flex_groups))
3073                         vfree(sbi->s_flex_groups);
3074                 else
3075                         kfree(sbi->s_flex_groups);
3076         }
3077 failed_mount2:
3078         for (i = 0; i < db_count; i++)
3079                 brelse(sbi->s_group_desc[i]);
3080         kfree(sbi->s_group_desc);
3081 failed_mount:
3082         if (sbi->s_proc) {
3083                 remove_proc_entry(sb->s_id, ext4_proc_root);
3084         }
3085 #ifdef CONFIG_QUOTA
3086         for (i = 0; i < MAXQUOTAS; i++)
3087                 kfree(sbi->s_qf_names[i]);
3088 #endif
3089         ext4_blkdev_remove(sbi);
3090         brelse(bh);
3091 out_fail:
3092         sb->s_fs_info = NULL;
3093         kfree(sbi->s_blockgroup_lock);
3094         kfree(sbi);
3095         lock_kernel();
3096         kfree(orig_data);
3097         return ret;
3098 }
3099
3100 /*
3101  * Setup any per-fs journal parameters now.  We'll do this both on
3102  * initial mount, once the journal has been initialised but before we've
3103  * done any recovery; and again on any subsequent remount.
3104  */
3105 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3106 {
3107         struct ext4_sb_info *sbi = EXT4_SB(sb);
3108
3109         journal->j_commit_interval = sbi->s_commit_interval;
3110         journal->j_min_batch_time = sbi->s_min_batch_time;
3111         journal->j_max_batch_time = sbi->s_max_batch_time;
3112
3113         spin_lock(&journal->j_state_lock);
3114         if (test_opt(sb, BARRIER))
3115                 journal->j_flags |= JBD2_BARRIER;
3116         else
3117                 journal->j_flags &= ~JBD2_BARRIER;
3118         if (test_opt(sb, DATA_ERR_ABORT))
3119                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3120         else
3121                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3122         spin_unlock(&journal->j_state_lock);
3123 }
3124
3125 static journal_t *ext4_get_journal(struct super_block *sb,
3126                                    unsigned int journal_inum)
3127 {
3128         struct inode *journal_inode;
3129         journal_t *journal;
3130
3131         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3132
3133         /* First, test for the existence of a valid inode on disk.  Bad
3134          * things happen if we iget() an unused inode, as the subsequent
3135          * iput() will try to delete it. */
3136
3137         journal_inode = ext4_iget(sb, journal_inum);
3138         if (IS_ERR(journal_inode)) {
3139                 ext4_msg(sb, KERN_ERR, "no journal found");
3140                 return NULL;
3141         }
3142         if (!journal_inode->i_nlink) {
3143                 make_bad_inode(journal_inode);
3144                 iput(journal_inode);
3145                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3146                 return NULL;
3147         }
3148
3149         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3150                   journal_inode, journal_inode->i_size);
3151         if (!S_ISREG(journal_inode->i_mode)) {
3152                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3153                 iput(journal_inode);
3154                 return NULL;
3155         }
3156
3157         journal = jbd2_journal_init_inode(journal_inode);
3158         if (!journal) {
3159                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3160                 iput(journal_inode);
3161                 return NULL;
3162         }
3163         journal->j_private = sb;
3164         ext4_init_journal_params(sb, journal);
3165         return journal;
3166 }
3167
3168 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3169                                        dev_t j_dev)
3170 {
3171         struct buffer_head *bh;
3172         journal_t *journal;
3173         ext4_fsblk_t start;
3174         ext4_fsblk_t len;
3175         int hblock, blocksize;
3176         ext4_fsblk_t sb_block;
3177         unsigned long offset;
3178         struct ext4_super_block *es;
3179         struct block_device *bdev;
3180
3181         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3182
3183         bdev = ext4_blkdev_get(j_dev, sb);
3184         if (bdev == NULL)
3185                 return NULL;
3186
3187         if (bd_claim(bdev, sb)) {
3188                 ext4_msg(sb, KERN_ERR,
3189                         "failed to claim external journal device");
3190                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3191                 return NULL;
3192         }
3193
3194         blocksize = sb->s_blocksize;
3195         hblock = bdev_logical_block_size(bdev);
3196         if (blocksize < hblock) {
3197                 ext4_msg(sb, KERN_ERR,
3198                         "blocksize too small for journal device");
3199                 goto out_bdev;
3200         }
3201
3202         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3203         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3204         set_blocksize(bdev, blocksize);
3205         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3206                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3207                        "external journal");
3208                 goto out_bdev;
3209         }
3210
3211         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3212         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3213             !(le32_to_cpu(es->s_feature_incompat) &
3214               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3215                 ext4_msg(sb, KERN_ERR, "external journal has "
3216                                         "bad superblock");
3217                 brelse(bh);
3218                 goto out_bdev;
3219         }
3220
3221         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3222                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3223                 brelse(bh);
3224                 goto out_bdev;
3225         }
3226
3227         len = ext4_blocks_count(es);
3228         start = sb_block + 1;
3229         brelse(bh);     /* we're done with the superblock */
3230
3231         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3232                                         start, len, blocksize);
3233         if (!journal) {
3234                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3235                 goto out_bdev;
3236         }
3237         journal->j_private = sb;
3238         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3239         wait_on_buffer(journal->j_sb_buffer);
3240         if (!buffer_uptodate(journal->j_sb_buffer)) {
3241                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3242                 goto out_journal;
3243         }
3244         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3245                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3246                                         "user (unsupported) - %d",
3247                         be32_to_cpu(journal->j_superblock->s_nr_users));
3248                 goto out_journal;
3249         }
3250         EXT4_SB(sb)->journal_bdev = bdev;
3251         ext4_init_journal_params(sb, journal);
3252         return journal;
3253
3254 out_journal:
3255         jbd2_journal_destroy(journal);
3256 out_bdev:
3257         ext4_blkdev_put(bdev);
3258         return NULL;
3259 }
3260
3261 static int ext4_load_journal(struct super_block *sb,
3262                              struct ext4_super_block *es,
3263                              unsigned long journal_devnum)
3264 {
3265         journal_t *journal;
3266         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3267         dev_t journal_dev;
3268         int err = 0;
3269         int really_read_only;
3270
3271         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3272
3273         if (journal_devnum &&
3274             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3275                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3276                         "numbers have changed");
3277                 journal_dev = new_decode_dev(journal_devnum);
3278         } else
3279                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3280
3281         really_read_only = bdev_read_only(sb->s_bdev);
3282
3283         /*
3284          * Are we loading a blank journal or performing recovery after a
3285          * crash?  For recovery, we need to check in advance whether we
3286          * can get read-write access to the device.
3287          */
3288         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3289                 if (sb->s_flags & MS_RDONLY) {
3290                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3291                                         "required on readonly filesystem");
3292                         if (really_read_only) {
3293                                 ext4_msg(sb, KERN_ERR, "write access "
3294                                         "unavailable, cannot proceed");
3295                                 return -EROFS;
3296                         }
3297                         ext4_msg(sb, KERN_INFO, "write access will "
3298                                "be enabled during recovery");
3299                 }
3300         }
3301
3302         if (journal_inum && journal_dev) {
3303                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3304                        "and inode journals!");
3305                 return -EINVAL;
3306         }
3307
3308         if (journal_inum) {
3309                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3310                         return -EINVAL;
3311         } else {
3312                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3313                         return -EINVAL;
3314         }
3315
3316         if (!(journal->j_flags & JBD2_BARRIER))
3317                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3318
3319         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3320                 err = jbd2_journal_update_format(journal);
3321                 if (err)  {
3322                         ext4_msg(sb, KERN_ERR, "error updating journal");
3323                         jbd2_journal_destroy(journal);
3324                         return err;
3325                 }
3326         }
3327
3328         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3329                 err = jbd2_journal_wipe(journal, !really_read_only);
3330         if (!err)
3331                 err = jbd2_journal_load(journal);
3332
3333         if (err) {
3334                 ext4_msg(sb, KERN_ERR, "error loading journal");
3335                 jbd2_journal_destroy(journal);
3336                 return err;
3337         }
3338
3339         EXT4_SB(sb)->s_journal = journal;
3340         ext4_clear_journal_err(sb, es);
3341
3342         if (journal_devnum &&
3343             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3344                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3345
3346                 /* Make sure we flush the recovery flag to disk. */
3347                 ext4_commit_super(sb, 1);
3348         }
3349
3350         return 0;
3351 }
3352
3353 static int ext4_commit_super(struct super_block *sb, int sync)
3354 {
3355         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3356         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3357         int error = 0;
3358
3359         if (!sbh)
3360                 return error;
3361         if (buffer_write_io_error(sbh)) {
3362                 /*
3363                  * Oh, dear.  A previous attempt to write the
3364                  * superblock failed.  This could happen because the
3365                  * USB device was yanked out.  Or it could happen to
3366                  * be a transient write error and maybe the block will
3367                  * be remapped.  Nothing we can do but to retry the
3368                  * write and hope for the best.
3369                  */
3370                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3371                        "superblock detected");
3372                 clear_buffer_write_io_error(sbh);
3373                 set_buffer_uptodate(sbh);
3374         }
3375         /*
3376          * If the file system is mounted read-only, don't update the
3377          * superblock write time.  This avoids updating the superblock
3378          * write time when we are mounting the root file system
3379          * read/only but we need to replay the journal; at that point,
3380          * for people who are east of GMT and who make their clock
3381          * tick in localtime for Windows bug-for-bug compatibility,
3382          * the clock is set in the future, and this will cause e2fsck
3383          * to complain and force a full file system check.
3384          */
3385         if (!(sb->s_flags & MS_RDONLY))
3386                 es->s_wtime = cpu_to_le32(get_seconds());
3387         es->s_kbytes_written =
3388                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
3389                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3390                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3391         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3392                                         &EXT4_SB(sb)->s_freeblocks_counter));
3393         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3394                                         &EXT4_SB(sb)->s_freeinodes_counter));
3395         sb->s_dirt = 0;
3396         BUFFER_TRACE(sbh, "marking dirty");
3397         mark_buffer_dirty(sbh);
3398         if (sync) {
3399                 error = sync_dirty_buffer(sbh);
3400                 if (error)
3401                         return error;
3402
3403                 error = buffer_write_io_error(sbh);
3404                 if (error) {
3405                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3406                                "superblock");
3407                         clear_buffer_write_io_error(sbh);
3408                         set_buffer_uptodate(sbh);
3409                 }
3410         }
3411         return error;
3412 }
3413
3414 /*
3415  * Have we just finished recovery?  If so, and if we are mounting (or
3416  * remounting) the filesystem readonly, then we will end up with a
3417  * consistent fs on disk.  Record that fact.
3418  */
3419 static void ext4_mark_recovery_complete(struct super_block *sb,
3420                                         struct ext4_super_block *es)
3421 {
3422         journal_t *journal = EXT4_SB(sb)->s_journal;
3423
3424         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3425                 BUG_ON(journal != NULL);
3426                 return;
3427         }
3428         jbd2_journal_lock_updates(journal);
3429         if (jbd2_journal_flush(journal) < 0)
3430                 goto out;
3431
3432         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3433             sb->s_flags & MS_RDONLY) {
3434                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3435                 ext4_commit_super(sb, 1);
3436         }
3437
3438 out:
3439         jbd2_journal_unlock_updates(journal);
3440 }
3441
3442 /*
3443  * If we are mounting (or read-write remounting) a filesystem whose journal
3444  * has recorded an error from a previous lifetime, move that error to the
3445  * main filesystem now.
3446  */
3447 static void ext4_clear_journal_err(struct super_block *sb,
3448                                    struct ext4_super_block *es)
3449 {
3450         journal_t *journal;
3451         int j_errno;
3452         const char *errstr;
3453
3454         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3455
3456         journal = EXT4_SB(sb)->s_journal;
3457
3458         /*
3459          * Now check for any error status which may have been recorded in the
3460          * journal by a prior ext4_error() or ext4_abort()
3461          */
3462
3463         j_errno = jbd2_journal_errno(journal);
3464         if (j_errno) {
3465                 char nbuf[16];
3466
3467                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3468                 ext4_warning(sb, "Filesystem error recorded "
3469                              "from previous mount: %s", errstr);
3470                 ext4_warning(sb, "Marking fs in need of filesystem check.");
3471
3472                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3473                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3474                 ext4_commit_super(sb, 1);
3475
3476                 jbd2_journal_clear_err(journal);
3477         }
3478 }
3479
3480 /*
3481  * Force the running and committing transactions to commit,
3482  * and wait on the commit.
3483  */
3484 int ext4_force_commit(struct super_block *sb)
3485 {
3486         journal_t *journal;
3487         int ret = 0;
3488
3489         if (sb->s_flags & MS_RDONLY)
3490                 return 0;
3491
3492         journal = EXT4_SB(sb)->s_journal;
3493         if (journal) {
3494                 vfs_check_frozen(sb, SB_FREEZE_WRITE);
3495                 ret = ext4_journal_force_commit(journal);
3496         }
3497
3498         return ret;
3499 }
3500
3501 static void ext4_write_super(struct super_block *sb)
3502 {
3503         lock_super(sb);
3504         ext4_commit_super(sb, 1);
3505         unlock_super(sb);
3506 }
3507
3508 static int ext4_sync_fs(struct super_block *sb, int wait)
3509 {
3510         int ret = 0;
3511         tid_t target;
3512         struct ext4_sb_info *sbi = EXT4_SB(sb);
3513
3514         trace_ext4_sync_fs(sb, wait);
3515         flush_workqueue(sbi->dio_unwritten_wq);
3516         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3517                 if (wait)
3518                         jbd2_log_wait_commit(sbi->s_journal, target);
3519         }
3520         return ret;
3521 }
3522
3523 /*
3524  * LVM calls this function before a (read-only) snapshot is created.  This
3525  * gives us a chance to flush the journal completely and mark the fs clean.
3526  */
3527 static int ext4_freeze(struct super_block *sb)
3528 {
3529         int error = 0;
3530         journal_t *journal;
3531
3532         if (sb->s_flags & MS_RDONLY)
3533                 return 0;
3534
3535         journal = EXT4_SB(sb)->s_journal;
3536
3537         /* Now we set up the journal barrier. */
3538         jbd2_journal_lock_updates(journal);
3539
3540         /*
3541          * Don't clear the needs_recovery flag if we failed to flush
3542          * the journal.
3543          */
3544         error = jbd2_journal_flush(journal);
3545         if (error < 0)
3546                 goto out;
3547
3548         /* Journal blocked and flushed, clear needs_recovery flag. */
3549         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3550         error = ext4_commit_super(sb, 1);
3551 out:
3552         /* we rely on s_frozen to stop further updates */
3553         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3554         return error;
3555 }
3556
3557 /*
3558  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3559  * flag here, even though the filesystem is not technically dirty yet.
3560  */
3561 static int ext4_unfreeze(struct super_block *sb)
3562 {
3563         if (sb->s_flags & MS_RDONLY)
3564                 return 0;
3565
3566         lock_super(sb);
3567         /* Reset the needs_recovery flag before the fs is unlocked. */
3568         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3569         ext4_commit_super(sb, 1);
3570         unlock_super(sb);
3571         return 0;
3572 }
3573
3574 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3575 {
3576         struct ext4_super_block *es;
3577         struct ext4_sb_info *sbi = EXT4_SB(sb);
3578         ext4_fsblk_t n_blocks_count = 0;
3579         unsigned long old_sb_flags;
3580         struct ext4_mount_options old_opts;
3581         int enable_quota = 0;
3582         ext4_group_t g;
3583         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3584         int err;
3585 #ifdef CONFIG_QUOTA
3586         int i;
3587 #endif
3588         char *orig_data = kstrdup(data, GFP_KERNEL);
3589
3590         lock_kernel();
3591
3592         /* Store the original options */
3593         lock_super(sb);
3594         old_sb_flags = sb->s_flags;
3595         old_opts.s_mount_opt = sbi->s_mount_opt;
3596         old_opts.s_resuid = sbi->s_resuid;
3597         old_opts.s_resgid = sbi->s_resgid;
3598         old_opts.s_commit_interval = sbi->s_commit_interval;
3599         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3600         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3601 #ifdef CONFIG_QUOTA
3602         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3603         for (i = 0; i < MAXQUOTAS; i++)
3604                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3605 #endif
3606         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3607                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3608
3609         /*
3610          * Allow the "check" option to be passed as a remount option.
3611          */
3612         if (!parse_options(data, sb, NULL, &journal_ioprio,
3613                            &n_blocks_count, 1)) {
3614                 err = -EINVAL;
3615                 goto restore_opts;
3616         }
3617
3618         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3619                 ext4_abort(sb, __func__, "Abort forced by user");
3620
3621         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3622                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3623
3624         es = sbi->s_es;
3625
3626         if (sbi->s_journal) {
3627                 ext4_init_journal_params(sb, sbi->s_journal);
3628                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3629         }
3630
3631         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3632                 n_blocks_count > ext4_blocks_count(es)) {
3633                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3634                         err = -EROFS;
3635                         goto restore_opts;
3636                 }
3637
3638                 if (*flags & MS_RDONLY) {
3639                         err = dquot_suspend(sb, -1);
3640                         if (err < 0)
3641                                 goto restore_opts;
3642
3643                         /*
3644                          * First of all, the unconditional stuff we have to do
3645                          * to disable replay of the journal when we next remount
3646                          */
3647                         sb->s_flags |= MS_RDONLY;
3648
3649                         /*
3650                          * OK, test if we are remounting a valid rw partition
3651                          * readonly, and if so set the rdonly flag and then
3652                          * mark the partition as valid again.
3653                          */
3654                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3655                             (sbi->s_mount_state & EXT4_VALID_FS))
3656                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3657
3658                         if (sbi->s_journal)
3659                                 ext4_mark_recovery_complete(sb, es);
3660                 } else {
3661                         /* Make sure we can mount this feature set readwrite */
3662                         if (!ext4_feature_set_ok(sb, 0)) {
3663                                 err = -EROFS;
3664                                 goto restore_opts;
3665                         }
3666                         /*
3667                          * Make sure the group descriptor checksums
3668                          * are sane.  If they aren't, refuse to remount r/w.
3669                          */
3670                         for (g = 0; g < sbi->s_groups_count; g++) {
3671                                 struct ext4_group_desc *gdp =
3672                                         ext4_get_group_desc(sb, g, NULL);
3673
3674                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3675                                         ext4_msg(sb, KERN_ERR,
3676                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3677                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3678                                                le16_to_cpu(gdp->bg_checksum));
3679                                         err = -EINVAL;
3680                                         goto restore_opts;
3681                                 }
3682                         }
3683
3684                         /*
3685                          * If we have an unprocessed orphan list hanging
3686                          * around from a previously readonly bdev mount,
3687                          * require a full umount/remount for now.
3688                          */
3689                         if (es->s_last_orphan) {
3690                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3691                                        "remount RDWR because of unprocessed "
3692                                        "orphan inode list.  Please "
3693                                        "umount/remount instead");
3694                                 err = -EINVAL;
3695                                 goto restore_opts;
3696                         }
3697
3698                         /*
3699                          * Mounting a RDONLY partition read-write, so reread
3700                          * and store the current valid flag.  (It may have
3701                          * been changed by e2fsck since we originally mounted
3702                          * the partition.)
3703                          */
3704                         if (sbi->s_journal)
3705                                 ext4_clear_journal_err(sb, es);
3706                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3707                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3708                                 goto restore_opts;
3709                         if (!ext4_setup_super(sb, es, 0))
3710                                 sb->s_flags &= ~MS_RDONLY;
3711                         enable_quota = 1;
3712                 }
3713         }
3714         ext4_setup_system_zone(sb);
3715         if (sbi->s_journal == NULL)
3716                 ext4_commit_super(sb, 1);
3717
3718 #ifdef CONFIG_QUOTA
3719         /* Release old quota file names */
3720         for (i = 0; i < MAXQUOTAS; i++)
3721                 if (old_opts.s_qf_names[i] &&
3722                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3723                         kfree(old_opts.s_qf_names[i]);
3724 #endif
3725         unlock_super(sb);
3726         unlock_kernel();
3727         if (enable_quota)
3728                 dquot_resume(sb, -1);
3729
3730         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
3731         kfree(orig_data);
3732         return 0;
3733
3734 restore_opts:
3735         sb->s_flags = old_sb_flags;
3736         sbi->s_mount_opt = old_opts.s_mount_opt;
3737         sbi->s_resuid = old_opts.s_resuid;
3738         sbi->s_resgid = old_opts.s_resgid;
3739         sbi->s_commit_interval = old_opts.s_commit_interval;
3740         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3741         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3742 #ifdef CONFIG_QUOTA
3743         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3744         for (i = 0; i < MAXQUOTAS; i++) {
3745                 if (sbi->s_qf_names[i] &&
3746                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3747                         kfree(sbi->s_qf_names[i]);
3748                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3749         }
3750 #endif
3751         unlock_super(sb);
3752         unlock_kernel();
3753         kfree(orig_data);
3754         return err;
3755 }
3756
3757 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3758 {
3759         struct super_block *sb = dentry->d_sb;
3760         struct ext4_sb_info *sbi = EXT4_SB(sb);
3761         struct ext4_super_block *es = sbi->s_es;
3762         u64 fsid;
3763
3764         if (test_opt(sb, MINIX_DF)) {
3765                 sbi->s_overhead_last = 0;
3766         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3767                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3768                 ext4_fsblk_t overhead = 0;
3769
3770                 /*
3771                  * Compute the overhead (FS structures).  This is constant
3772                  * for a given filesystem unless the number of block groups
3773                  * changes so we cache the previous value until it does.
3774                  */
3775
3776                 /*
3777                  * All of the blocks before first_data_block are
3778                  * overhead
3779                  */
3780                 overhead = le32_to_cpu(es->s_first_data_block);
3781
3782                 /*
3783                  * Add the overhead attributed to the superblock and
3784                  * block group descriptors.  If the sparse superblocks
3785                  * feature is turned on, then not all groups have this.
3786                  */
3787                 for (i = 0; i < ngroups; i++) {
3788                         overhead += ext4_bg_has_super(sb, i) +
3789                                 ext4_bg_num_gdb(sb, i);
3790                         cond_resched();
3791                 }
3792
3793                 /*
3794                  * Every block group has an inode bitmap, a block
3795                  * bitmap, and an inode table.
3796                  */
3797                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3798                 sbi->s_overhead_last = overhead;
3799                 smp_wmb();
3800                 sbi->s_blocks_last = ext4_blocks_count(es);
3801         }
3802
3803         buf->f_type = EXT4_SUPER_MAGIC;
3804         buf->f_bsize = sb->s_blocksize;
3805         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3806         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3807                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3808         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3809         if (buf->f_bfree < ext4_r_blocks_count(es))
3810                 buf->f_bavail = 0;
3811         buf->f_files = le32_to_cpu(es->s_inodes_count);
3812         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3813         buf->f_namelen = EXT4_NAME_LEN;
3814         fsid = le64_to_cpup((void *)es->s_uuid) ^
3815                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3816         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3817         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3818
3819         return 0;
3820 }
3821
3822 /* Helper function for writing quotas on sync - we need to start transaction
3823  * before quota file is locked for write. Otherwise the are possible deadlocks:
3824  * Process 1                         Process 2
3825  * ext4_create()                     quota_sync()
3826  *   jbd2_journal_start()                  write_dquot()
3827  *   dquot_initialize()                         down(dqio_mutex)
3828  *     down(dqio_mutex)                    jbd2_journal_start()
3829  *
3830  */
3831
3832 #ifdef CONFIG_QUOTA
3833
3834 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3835 {
3836         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3837 }
3838
3839 static int ext4_write_dquot(struct dquot *dquot)
3840 {
3841         int ret, err;
3842         handle_t *handle;
3843         struct inode *inode;
3844
3845         inode = dquot_to_inode(dquot);
3846         handle = ext4_journal_start(inode,
3847                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3848         if (IS_ERR(handle))
3849                 return PTR_ERR(handle);
3850         ret = dquot_commit(dquot);
3851         err = ext4_journal_stop(handle);
3852         if (!ret)
3853                 ret = err;
3854         return ret;
3855 }
3856
3857 static int ext4_acquire_dquot(struct dquot *dquot)
3858 {
3859         int ret, err;
3860         handle_t *handle;
3861
3862         handle = ext4_journal_start(dquot_to_inode(dquot),
3863                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3864         if (IS_ERR(handle))
3865                 return PTR_ERR(handle);
3866         ret = dquot_acquire(dquot);
3867         err = ext4_journal_stop(handle);
3868         if (!ret)
3869                 ret = err;
3870         return ret;
3871 }
3872
3873 static int ext4_release_dquot(struct dquot *dquot)
3874 {
3875         int ret, err;
3876         handle_t *handle;
3877
3878         handle = ext4_journal_start(dquot_to_inode(dquot),
3879                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3880         if (IS_ERR(handle)) {
3881                 /* Release dquot anyway to avoid endless cycle in dqput() */
3882                 dquot_release(dquot);
3883                 return PTR_ERR(handle);
3884         }
3885         ret = dquot_release(dquot);
3886         err = ext4_journal_stop(handle);
3887         if (!ret)
3888                 ret = err;
3889         return ret;
3890 }
3891
3892 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3893 {
3894         /* Are we journaling quotas? */
3895         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3896             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3897                 dquot_mark_dquot_dirty(dquot);
3898                 return ext4_write_dquot(dquot);
3899         } else {
3900                 return dquot_mark_dquot_dirty(dquot);
3901         }
3902 }
3903
3904 static int ext4_write_info(struct super_block *sb, int type)
3905 {
3906         int ret, err;
3907         handle_t *handle;
3908
3909         /* Data block + inode block */
3910         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3911         if (IS_ERR(handle))
3912                 return PTR_ERR(handle);
3913         ret = dquot_commit_info(sb, type);
3914         err = ext4_journal_stop(handle);
3915         if (!ret)
3916                 ret = err;
3917         return ret;
3918 }
3919
3920 /*
3921  * Turn on quotas during mount time - we need to find
3922  * the quota file and such...
3923  */
3924 static int ext4_quota_on_mount(struct super_block *sb, int type)
3925 {
3926         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3927                                         EXT4_SB(sb)->s_jquota_fmt, type);
3928 }
3929
3930 /*
3931  * Standard function to be called on quota_on
3932  */
3933 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3934                          char *name)
3935 {
3936         int err;
3937         struct path path;
3938
3939         if (!test_opt(sb, QUOTA))
3940                 return -EINVAL;
3941
3942         err = kern_path(name, LOOKUP_FOLLOW, &path);
3943         if (err)
3944                 return err;
3945
3946         /* Quotafile not on the same filesystem? */
3947         if (path.mnt->mnt_sb != sb) {
3948                 path_put(&path);
3949                 return -EXDEV;
3950         }
3951         /* Journaling quota? */
3952         if (EXT4_SB(sb)->s_qf_names[type]) {
3953                 /* Quotafile not in fs root? */
3954                 if (path.dentry->d_parent != sb->s_root)
3955                         ext4_msg(sb, KERN_WARNING,
3956                                 "Quota file not on filesystem root. "
3957                                 "Journaled quota will not work");
3958         }
3959
3960         /*
3961          * When we journal data on quota file, we have to flush journal to see
3962          * all updates to the file when we bypass pagecache...
3963          */
3964         if (EXT4_SB(sb)->s_journal &&
3965             ext4_should_journal_data(path.dentry->d_inode)) {
3966                 /*
3967                  * We don't need to lock updates but journal_flush() could
3968                  * otherwise be livelocked...
3969                  */
3970                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3971                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3972                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3973                 if (err) {
3974                         path_put(&path);
3975                         return err;
3976                 }
3977         }
3978
3979         err = dquot_quota_on_path(sb, type, format_id, &path);
3980         path_put(&path);
3981         return err;
3982 }
3983
3984 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3985  * acquiring the locks... As quota files are never truncated and quota code
3986  * itself serializes the operations (and noone else should touch the files)
3987  * we don't have to be afraid of races */
3988 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3989                                size_t len, loff_t off)
3990 {
3991         struct inode *inode = sb_dqopt(sb)->files[type];
3992         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3993         int err = 0;
3994         int offset = off & (sb->s_blocksize - 1);
3995         int tocopy;
3996         size_t toread;
3997         struct buffer_head *bh;
3998         loff_t i_size = i_size_read(inode);
3999
4000         if (off > i_size)
4001                 return 0;
4002         if (off+len > i_size)
4003                 len = i_size-off;
4004         toread = len;
4005         while (toread > 0) {
4006                 tocopy = sb->s_blocksize - offset < toread ?
4007                                 sb->s_blocksize - offset : toread;
4008                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4009                 if (err)
4010                         return err;
4011                 if (!bh)        /* A hole? */
4012                         memset(data, 0, tocopy);
4013                 else
4014                         memcpy(data, bh->b_data+offset, tocopy);
4015                 brelse(bh);
4016                 offset = 0;
4017                 toread -= tocopy;
4018                 data += tocopy;
4019                 blk++;
4020         }
4021         return len;
4022 }
4023
4024 /* Write to quotafile (we know the transaction is already started and has
4025  * enough credits) */
4026 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4027                                 const char *data, size_t len, loff_t off)
4028 {
4029         struct inode *inode = sb_dqopt(sb)->files[type];
4030         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4031         int err = 0;
4032         int offset = off & (sb->s_blocksize - 1);
4033         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
4034         struct buffer_head *bh;
4035         handle_t *handle = journal_current_handle();
4036
4037         if (EXT4_SB(sb)->s_journal && !handle) {
4038                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4039                         " cancelled because transaction is not started",
4040                         (unsigned long long)off, (unsigned long long)len);
4041                 return -EIO;
4042         }
4043         /*
4044          * Since we account only one data block in transaction credits,
4045          * then it is impossible to cross a block boundary.
4046          */
4047         if (sb->s_blocksize - offset < len) {
4048                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4049                         " cancelled because not block aligned",
4050                         (unsigned long long)off, (unsigned long long)len);
4051                 return -EIO;
4052         }
4053
4054         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4055         bh = ext4_bread(handle, inode, blk, 1, &err);
4056         if (!bh)
4057                 goto out;
4058         if (journal_quota) {
4059                 err = ext4_journal_get_write_access(handle, bh);
4060                 if (err) {
4061                         brelse(bh);
4062                         goto out;
4063                 }
4064         }
4065         lock_buffer(bh);
4066         memcpy(bh->b_data+offset, data, len);
4067         flush_dcache_page(bh->b_page);
4068         unlock_buffer(bh);
4069         if (journal_quota)
4070                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
4071         else {
4072                 /* Always do at least ordered writes for quotas */
4073                 err = ext4_jbd2_file_inode(handle, inode);
4074                 mark_buffer_dirty(bh);
4075         }
4076         brelse(bh);
4077 out:
4078         if (err) {
4079                 mutex_unlock(&inode->i_mutex);
4080                 return err;
4081         }
4082         if (inode->i_size < off + len) {
4083                 i_size_write(inode, off + len);
4084                 EXT4_I(inode)->i_disksize = inode->i_size;
4085         }
4086         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4087         ext4_mark_inode_dirty(handle, inode);
4088         mutex_unlock(&inode->i_mutex);
4089         return len;
4090 }
4091
4092 #endif
4093
4094 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
4095                        const char *dev_name, void *data, struct vfsmount *mnt)
4096 {
4097         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
4098 }
4099
4100 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4101 static struct file_system_type ext2_fs_type = {
4102         .owner          = THIS_MODULE,
4103         .name           = "ext2",
4104         .get_sb         = ext4_get_sb,
4105         .kill_sb        = kill_block_super,
4106         .fs_flags       = FS_REQUIRES_DEV,
4107 };
4108
4109 static inline void register_as_ext2(void)
4110 {
4111         int err = register_filesystem(&ext2_fs_type);
4112         if (err)
4113                 printk(KERN_WARNING
4114                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4115 }
4116
4117 static inline void unregister_as_ext2(void)
4118 {
4119         unregister_filesystem(&ext2_fs_type);
4120 }
4121 MODULE_ALIAS("ext2");
4122 #else
4123 static inline void register_as_ext2(void) { }
4124 static inline void unregister_as_ext2(void) { }
4125 #endif
4126
4127 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4128 static inline void register_as_ext3(void)
4129 {
4130         int err = register_filesystem(&ext3_fs_type);
4131         if (err)
4132                 printk(KERN_WARNING
4133                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4134 }
4135
4136 static inline void unregister_as_ext3(void)
4137 {
4138         unregister_filesystem(&ext3_fs_type);
4139 }
4140 MODULE_ALIAS("ext3");
4141 #else
4142 static inline void register_as_ext3(void) { }
4143 static inline void unregister_as_ext3(void) { }
4144 #endif
4145
4146 static struct file_system_type ext4_fs_type = {
4147         .owner          = THIS_MODULE,
4148         .name           = "ext4",
4149         .get_sb         = ext4_get_sb,
4150         .kill_sb        = kill_block_super,
4151         .fs_flags       = FS_REQUIRES_DEV,
4152 };
4153
4154 static int __init init_ext4_fs(void)
4155 {
4156         int err;
4157
4158         ext4_check_flag_values();
4159         err = init_ext4_system_zone();
4160         if (err)
4161                 return err;
4162         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4163         if (!ext4_kset)
4164                 goto out4;
4165         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4166         err = init_ext4_mballoc();
4167         if (err)
4168                 goto out3;
4169
4170         err = init_ext4_xattr();
4171         if (err)
4172                 goto out2;
4173         err = init_inodecache();
4174         if (err)
4175                 goto out1;
4176         register_as_ext2();
4177         register_as_ext3();
4178         err = register_filesystem(&ext4_fs_type);
4179         if (err)
4180                 goto out;
4181         return 0;
4182 out:
4183         unregister_as_ext2();
4184         unregister_as_ext3();
4185         destroy_inodecache();
4186 out1:
4187         exit_ext4_xattr();
4188 out2:
4189         exit_ext4_mballoc();
4190 out3:
4191         remove_proc_entry("fs/ext4", NULL);
4192         kset_unregister(ext4_kset);
4193 out4:
4194         exit_ext4_system_zone();
4195         return err;
4196 }
4197
4198 static void __exit exit_ext4_fs(void)
4199 {
4200         unregister_as_ext2();
4201         unregister_as_ext3();
4202         unregister_filesystem(&ext4_fs_type);
4203         destroy_inodecache();
4204         exit_ext4_xattr();
4205         exit_ext4_mballoc();
4206         remove_proc_entry("fs/ext4", NULL);
4207         kset_unregister(ext4_kset);
4208         exit_ext4_system_zone();
4209 }
4210
4211 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4212 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4213 MODULE_LICENSE("GPL");
4214 module_init(init_ext4_fs)
4215 module_exit(exit_ext4_fs)