fs/super.c

   1 /*
   2  *  linux/fs/super.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  super.c contains code to handle: - mount structures
   7  *                                   - super-block tables
   8  *                                   - filesystem drivers list
   9  *                                   - mount system call
  10  *                                   - umount system call
  11  *                                   - ustat system call
  12  *
  13  * GK 2/5/95  -  Changed to support mounting the root fs via NFS
  14  *
  15  *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
  16  *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
  17  *  Added options to /proc/mounts:
  18  *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
  19  *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
  20  *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/slab.h>
  25 #include <linux/acct.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/mount.h>
  29 #include <linux/security.h>
  30 #include <linux/writeback.h>            /* for the emergency remount stuff */
  31 #include <linux/idr.h>
  32 #include <linux/mutex.h>
  33 #include <linux/backing-dev.h>
  34 #include "internal.h"
  35
  36
  37 LIST_HEAD(super_blocks);
  38 DEFINE_SPINLOCK(sb_lock);
  39
  40 /**
  41  *      alloc_super     -       create new superblock
  42  *      @type:  filesystem type superblock should belong to
  43  *
  44  *      Allocates and initializes a new &struct super_block.  alloc_super()
  45  *      returns a pointer new superblock or %NULL if allocation had failed.
  46  */
  47 static struct super_block *alloc_super(struct file_system_type *type)
  48 {
  49         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
  50         static const struct super_operations default_op;
  51
  52         if (s) {
  53                 if (security_sb_alloc(s)) {
  54                         kfree(s);
  55                         s = NULL;
  56                         goto out;
  57                 }
  58                 INIT_LIST_HEAD(&s->s_files);
  59                 INIT_LIST_HEAD(&s->s_instances);
  60                 INIT_HLIST_HEAD(&s->s_anon);
  61                 INIT_LIST_HEAD(&s->s_inodes);
  62                 INIT_LIST_HEAD(&s->s_dentry_lru);
  63                 init_rwsem(&s->s_umount);
  64                 mutex_init(&s->s_lock);
  65                 lockdep_set_class(&s->s_umount, &type->s_umount_key);
  66                 /*
  67                  * The locking rules for s_lock are up to the
  68                  * filesystem. For example ext3fs has different
  69                  * lock ordering than usbfs:
  70                  */
  71                 lockdep_set_class(&s->s_lock, &type->s_lock_key);
  72                 /*
  73                  * sget() can have s_umount recursion.
  74                  *
  75                  * When it cannot find a suitable sb, it allocates a new
  76                  * one (this one), and tries again to find a suitable old
  77                  * one.
  78                  *
  79                  * In case that succeeds, it will acquire the s_umount
  80                  * lock of the old one. Since these are clearly distrinct
  81                  * locks, and this object isn't exposed yet, there's no
  82                  * risk of deadlocks.
  83                  *
  84                  * Annotate this by putting this lock in a different
  85                  * subclass.
  86                  */
  87                 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
  88                 s->s_count = 1;
  89                 atomic_set(&s->s_active, 1);
  90                 mutex_init(&s->s_vfs_rename_mutex);
  91                 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
  92                 mutex_init(&s->s_dquot.dqio_mutex);
  93                 mutex_init(&s->s_dquot.dqonoff_mutex);
  94                 init_rwsem(&s->s_dquot.dqptr_sem);
  95                 init_waitqueue_head(&s->s_wait_unfrozen);
  96                 s->s_maxbytes = MAX_NON_LFS;
  97                 s->dq_op = sb_dquot_ops;
  98                 s->s_qcop = sb_quotactl_ops;
  99                 s->s_op = &default_op;
 100                 s->s_time_gran = 1000000000;
 101         }
 102 out:
 103         return s;
 104 }
 105
 106 /**
 107  *      destroy_super   -       frees a superblock
 108  *      @s: superblock to free
 109  *
 110  *      Frees a superblock.
 111  */
 112 static inline void destroy_super(struct super_block *s)
 113 {
 114         security_sb_free(s);
 115         kfree(s->s_subtype);
 116         kfree(s->s_options);
 117         kfree(s);
 118 }
 119
 120 /* Superblock refcounting  */
 121
 122 /*
 123  * Drop a superblock's refcount.  The caller must hold sb_lock.
 124  */
 125 void __put_super(struct super_block *sb)
 126 {
 127         if (!--sb->s_count) {
 128                 list_del_init(&sb->s_list);
 129                 destroy_super(sb);
 130         }
 131 }
 132
 133 /**
 134  *      put_super       -       drop a temporary reference to superblock
 135  *      @sb: superblock in question
 136  *
 137  *      Drops a temporary reference, frees superblock if there's no
 138  *      references left.
 139  */
 140 void put_super(struct super_block *sb)
 141 {
 142         spin_lock(&sb_lock);
 143         __put_super(sb);
 144         spin_unlock(&sb_lock);
 145 }
 146
 147
 148 /**
 149  *      deactivate_locked_super -       drop an active reference to superblock
 150  *      @s: superblock to deactivate
 151  *
 152  *      Drops an active reference to superblock, converting it into a temprory
 153  *      one if there is no other active references left.  In that case we
 154  *      tell fs driver to shut it down and drop the temporary reference we
 155  *      had just acquired.
 156  *
 157  *      Caller holds exclusive lock on superblock; that lock is released.
 158  */
 159 void deactivate_locked_super(struct super_block *s)
 160 {
 161         struct file_system_type *fs = s->s_type;
 162         if (atomic_dec_and_test(&s->s_active)) {
 163                 vfs_dq_off(s, 0);
 164                 fs->kill_sb(s);
 165                 put_filesystem(fs);
 166                 put_super(s);
 167         } else {
 168                 up_write(&s->s_umount);
 169         }
 170 }
 171
 172 EXPORT_SYMBOL(deactivate_locked_super);
 173
 174 /**
 175  *      deactivate_super        -       drop an active reference to superblock
 176  *      @s: superblock to deactivate
 177  *
 178  *      Variant of deactivate_locked_super(), except that superblock is *not*
 179  *      locked by caller.  If we are going to drop the final active reference,
 180  *      lock will be acquired prior to that.
 181  */
 182 void deactivate_super(struct super_block *s)
 183 {
 184         if (!atomic_add_unless(&s->s_active, -1, 1)) {
 185                 down_write(&s->s_umount);
 186                 deactivate_locked_super(s);
 187         }
 188 }
 189
 190 EXPORT_SYMBOL(deactivate_super);
 191
 192 /**
 193  *      grab_super - acquire an active reference
 194  *      @s: reference we are trying to make active
 195  *
 196  *      Tries to acquire an active reference.  grab_super() is used when we
 197  *      had just found a superblock in super_blocks or fs_type->fs_supers
 198  *      and want to turn it into a full-blown active reference.  grab_super()
 199  *      is called with sb_lock held and drops it.  Returns 1 in case of
 200  *      success, 0 if we had failed (superblock contents was already dead or
 201  *      dying when grab_super() had been called).
 202  */
 203 static int grab_super(struct super_block *s) __releases(sb_lock)
 204 {
 205         if (atomic_inc_not_zero(&s->s_active)) {
 206                 spin_unlock(&sb_lock);
 207                 return 1;
 208         }
 209         /* it's going away */
 210         s->s_count++;
 211         spin_unlock(&sb_lock);
 212         /* wait for it to die */
 213         down_write(&s->s_umount);
 214         up_write(&s->s_umount);
 215         put_super(s);
 216         return 0;
 217 }
 218
 219 /*
 220  * Superblock locking.  We really ought to get rid of these two.
 221  */
 222 void lock_super(struct super_block * sb)
 223 {
 224         get_fs_excl();
 225         mutex_lock(&sb->s_lock);
 226 }
 227
 228 void unlock_super(struct super_block * sb)
 229 {
 230         put_fs_excl();
 231         mutex_unlock(&sb->s_lock);
 232 }
 233
 234 EXPORT_SYMBOL(lock_super);
 235 EXPORT_SYMBOL(unlock_super);
 236
 237 /**
 238  *      generic_shutdown_super  -       common helper for ->kill_sb()
 239  *      @sb: superblock to kill
 240  *
 241  *      generic_shutdown_super() does all fs-independent work on superblock
 242  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
 243  *      that need destruction out of superblock, call generic_shutdown_super()
 244  *      and release aforementioned objects.  Note: dentries and inodes _are_
 245  *      taken care of and do not need specific handling.
 246  *
 247  *      Upon calling this function, the filesystem may no longer alter or
 248  *      rearrange the set of dentries belonging to this super_block, nor may it
 249  *      change the attachments of dentries to inodes.
 250  */
 251 void generic_shutdown_super(struct super_block *sb)
 252 {
 253         const struct super_operations *sop = sb->s_op;
 254
 255
 256         if (sb->s_root) {
 257                 shrink_dcache_for_umount(sb);
 258                 sync_filesystem(sb);
 259                 get_fs_excl();
 260                 sb->s_flags &= ~MS_ACTIVE;
 261
 262                 /* bad name - it should be evict_inodes() */
 263                 invalidate_inodes(sb);
 264
 265                 if (sop->put_super)
 266                         sop->put_super(sb);
 267
 268                 /* Forget any remaining inodes */
 269                 if (invalidate_inodes(sb)) {
 270                         printk("VFS: Busy inodes after unmount of %s. "
 271                            "Self-destruct in 5 seconds.  Have a nice day...\n",
 272                            sb->s_id);
 273                 }
 274                 put_fs_excl();
 275         }
 276         spin_lock(&sb_lock);
 277         /* should be initialized for __put_super_and_need_restart() */
 278         list_del_init(&sb->s_instances);
 279         spin_unlock(&sb_lock);
 280         up_write(&sb->s_umount);
 281 }
 282
 283 EXPORT_SYMBOL(generic_shutdown_super);
 284
 285 /**
 286  *      sget    -       find or create a superblock
 287  *      @type:  filesystem type superblock should belong to
 288  *      @test:  comparison callback
 289  *      @set:   setup callback
 290  *      @data:  argument to each of them
 291  */
 292 struct super_block *sget(struct file_system_type *type,
 293                         int (*test)(struct super_block *,void *),
 294                         int (*set)(struct super_block *,void *),
 295                         void *data)
 296 {
 297         struct super_block *s = NULL;
 298         struct super_block *old;
 299         int err;
 300
 301 retry:
 302         spin_lock(&sb_lock);
 303         if (test) {
 304                 list_for_each_entry(old, &type->fs_supers, s_instances) {
 305                         if (!test(old, data))
 306                                 continue;
 307                         if (!grab_super(old))
 308                                 goto retry;
 309                         if (s) {
 310                                 up_write(&s->s_umount);
 311                                 destroy_super(s);
 312                         }
 313                         down_write(&old->s_umount);
 314                         return old;
 315                 }
 316         }
 317         if (!s) {
 318                 spin_unlock(&sb_lock);
 319                 s = alloc_super(type);
 320                 if (!s)
 321                         return ERR_PTR(-ENOMEM);
 322                 goto retry;
 323         }
 324
 325         err = set(s, data);
 326         if (err) {
 327                 spin_unlock(&sb_lock);
 328                 up_write(&s->s_umount);
 329                 destroy_super(s);
 330                 return ERR_PTR(err);
 331         }
 332         s->s_type = type;
 333         strlcpy(s->s_id, type->name, sizeof(s->s_id));
 334         list_add_tail(&s->s_list, &super_blocks);
 335         list_add(&s->s_instances, &type->fs_supers);
 336         spin_unlock(&sb_lock);
 337         get_filesystem(type);
 338         return s;
 339 }
 340
 341 EXPORT_SYMBOL(sget);
 342
 343 void drop_super(struct super_block *sb)
 344 {
 345         up_read(&sb->s_umount);
 346         put_super(sb);
 347 }
 348
 349 EXPORT_SYMBOL(drop_super);
 350
 351 /**
 352  * sync_supers - helper for periodic superblock writeback
 353  *
 354  * Call the write_super method if present on all dirty superblocks in
 355  * the system.  This is for the periodic writeback used by most older
 356  * filesystems.  For data integrity superblock writeback use
 357  * sync_filesystems() instead.
 358  *
 359  * Note: check the dirty flag before waiting, so we don't
 360  * hold up the sync while mounting a device. (The newly
 361  * mounted device won't need syncing.)
 362  */
 363 void sync_supers(void)
 364 {
 365         struct super_block *sb, *n;
 366
 367         spin_lock(&sb_lock);
 368         list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
 369                 if (list_empty(&sb->s_instances))
 370                         continue;
 371                 if (sb->s_op->write_super && sb->s_dirt) {
 372                         sb->s_count++;
 373                         spin_unlock(&sb_lock);
 374
 375                         down_read(&sb->s_umount);
 376                         if (sb->s_root && sb->s_dirt)
 377                                 sb->s_op->write_super(sb);
 378                         up_read(&sb->s_umount);
 379
 380                         spin_lock(&sb_lock);
 381                         __put_super(sb);
 382                 }
 383         }
 384         spin_unlock(&sb_lock);
 385 }
 386
 387 /**
 388  *      iterate_supers - call function for all active superblocks
 389  *      @f: function to call
 390  *      @arg: argument to pass to it
 391  *
 392  *      Scans the superblock list and calls given function, passing it
 393  *      locked superblock and given argument.
 394  */
 395 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
 396 {
 397         struct super_block *sb, *n;
 398
 399         spin_lock(&sb_lock);
 400         list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
 401                 if (list_empty(&sb->s_instances))
 402                         continue;
 403                 sb->s_count++;
 404                 spin_unlock(&sb_lock);
 405
 406                 down_read(&sb->s_umount);
 407                 if (sb->s_root)
 408                         f(sb, arg);
 409                 up_read(&sb->s_umount);
 410
 411                 spin_lock(&sb_lock);
 412                 __put_super(sb);
 413         }
 414         spin_unlock(&sb_lock);
 415 }
 416
 417 /**
 418  *      get_super - get the superblock of a device
 419  *      @bdev: device to get the superblock for
 420  *
 421  *      Scans the superblock list and finds the superblock of the file system
 422  *      mounted on the device given. %NULL is returned if no match is found.
 423  */
 424
 425 struct super_block *get_super(struct block_device *bdev)
 426 {
 427         struct super_block *sb;
 428
 429         if (!bdev)
 430                 return NULL;
 431
 432         spin_lock(&sb_lock);
 433 rescan:
 434         list_for_each_entry(sb, &super_blocks, s_list) {
 435                 if (list_empty(&sb->s_instances))
 436                         continue;
 437                 if (sb->s_bdev == bdev) {
 438                         sb->s_count++;
 439                         spin_unlock(&sb_lock);
 440                         down_read(&sb->s_umount);
 441                         /* still alive? */
 442                         if (sb->s_root)
 443                                 return sb;
 444                         up_read(&sb->s_umount);
 445                         /* nope, got unmounted */
 446                         spin_lock(&sb_lock);
 447                         __put_super(sb);
 448                         goto rescan;
 449                 }
 450         }
 451         spin_unlock(&sb_lock);
 452         return NULL;
 453 }
 454
 455 EXPORT_SYMBOL(get_super);
 456
 457 /**
 458  * get_active_super - get an active reference to the superblock of a device
 459  * @bdev: device to get the superblock for
 460  *
 461  * Scans the superblock list and finds the superblock of the file system
 462  * mounted on the device given.  Returns the superblock with an active
 463  * reference or %NULL if none was found.
 464  */
 465 struct super_block *get_active_super(struct block_device *bdev)
 466 {
 467         struct super_block *sb;
 468
 469         if (!bdev)
 470                 return NULL;
 471
 472 restart:
 473         spin_lock(&sb_lock);
 474         list_for_each_entry(sb, &super_blocks, s_list) {
 475                 if (list_empty(&sb->s_instances))
 476                         continue;
 477                 if (sb->s_bdev == bdev) {
 478                         if (grab_super(sb)) /* drops sb_lock */
 479                                 return sb;
 480                         else
 481                                 goto restart;
 482                 }
 483         }
 484         spin_unlock(&sb_lock);
 485         return NULL;
 486 }
 487
 488 struct super_block *user_get_super(dev_t dev)
 489 {
 490         struct super_block *sb;
 491
 492         spin_lock(&sb_lock);
 493 rescan:
 494         list_for_each_entry(sb, &super_blocks, s_list) {
 495                 if (list_empty(&sb->s_instances))
 496                         continue;
 497                 if (sb->s_dev ==  dev) {
 498                         sb->s_count++;
 499                         spin_unlock(&sb_lock);
 500                         down_read(&sb->s_umount);
 501                         /* still alive? */
 502                         if (sb->s_root)
 503                                 return sb;
 504                         up_read(&sb->s_umount);
 505                         /* nope, got unmounted */
 506                         spin_lock(&sb_lock);
 507                         __put_super(sb);
 508                         goto rescan;
 509                 }
 510         }
 511         spin_unlock(&sb_lock);
 512         return NULL;
 513 }
 514
 515 /**
 516  *      do_remount_sb - asks filesystem to change mount options.
 517  *      @sb:    superblock in question
 518  *      @flags: numeric part of options
 519  *      @data:  the rest of options
 520  *      @force: whether or not to force the change
 521  *
 522  *      Alters the mount options of a mounted file system.
 523  */
 524 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 525 {
 526         int retval;
 527         int remount_rw, remount_ro;
 528
 529         if (sb->s_frozen != SB_UNFROZEN)
 530                 return -EBUSY;
 531
 532 #ifdef CONFIG_BLOCK
 533         if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
 534                 return -EACCES;
 535 #endif
 536
 537         if (flags & MS_RDONLY)
 538                 acct_auto_close(sb);
 539         shrink_dcache_sb(sb);
 540         sync_filesystem(sb);
 541
 542         remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
 543         remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
 544
 545         /* If we are remounting RDONLY and current sb is read/write,
 546            make sure there are no rw files opened */
 547         if (remount_ro) {
 548                 if (force)
 549                         mark_files_ro(sb);
 550                 else if (!fs_may_remount_ro(sb))
 551                         return -EBUSY;
 552                 retval = vfs_dq_off(sb, 1);
 553                 if (retval < 0 && retval != -ENOSYS)
 554                         return -EBUSY;
 555         }
 556
 557         if (sb->s_op->remount_fs) {
 558                 retval = sb->s_op->remount_fs(sb, &flags, data);
 559                 if (retval)
 560                         return retval;
 561         }
 562         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
 563         if (remount_rw)
 564                 vfs_dq_quota_on_remount(sb);
 565         /*
 566          * Some filesystems modify their metadata via some other path than the
 567          * bdev buffer cache (eg. use a private mapping, or directories in
 568          * pagecache, etc). Also file data modifications go via their own
 569          * mappings. So If we try to mount readonly then copy the filesystem
 570          * from bdev, we could get stale data, so invalidate it to give a best
 571          * effort at coherency.
 572          */
 573         if (remount_ro && sb->s_bdev)
 574                 invalidate_bdev(sb->s_bdev);
 575         return 0;
 576 }
 577
 578 static void do_emergency_remount(struct work_struct *work)
 579 {
 580         struct super_block *sb, *n;
 581
 582         spin_lock(&sb_lock);
 583         list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
 584                 if (list_empty(&sb->s_instances))
 585                         continue;
 586                 sb->s_count++;
 587                 spin_unlock(&sb_lock);
 588                 down_write(&sb->s_umount);
 589                 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
 590                         /*
 591                          * What lock protects sb->s_flags??
 592                          */
 593                         do_remount_sb(sb, MS_RDONLY, NULL, 1);
 594                 }
 595                 up_write(&sb->s_umount);
 596                 spin_lock(&sb_lock);
 597                 __put_super(sb);
 598         }
 599         spin_unlock(&sb_lock);
 600         kfree(work);
 601         printk("Emergency Remount complete\n");
 602 }
 603
 604 void emergency_remount(void)
 605 {
 606         struct work_struct *work;
 607
 608         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 609         if (work) {
 610                 INIT_WORK(work, do_emergency_remount);
 611                 schedule_work(work);
 612         }
 613 }
 614
 615 /*
 616  * Unnamed block devices are dummy devices used by virtual
 617  * filesystems which don't use real block-devices.  -- jrs
 618  */
 619
 620 static DEFINE_IDA(unnamed_dev_ida);
 621 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
 622 static int unnamed_dev_start = 0; /* don't bother trying below it */
 623
 624 int set_anon_super(struct super_block *s, void *data)
 625 {
 626         int dev;
 627         int error;
 628
 629  retry:
 630         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
 631                 return -ENOMEM;
 632         spin_lock(&unnamed_dev_lock);
 633         error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
 634         if (!error)
 635                 unnamed_dev_start = dev + 1;
 636         spin_unlock(&unnamed_dev_lock);
 637         if (error == -EAGAIN)
 638                 /* We raced and lost with another CPU. */
 639                 goto retry;
 640         else if (error)
 641                 return -EAGAIN;
 642
 643         if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
 644                 spin_lock(&unnamed_dev_lock);
 645                 ida_remove(&unnamed_dev_ida, dev);
 646                 if (unnamed_dev_start > dev)
 647                         unnamed_dev_start = dev;
 648                 spin_unlock(&unnamed_dev_lock);
 649                 return -EMFILE;
 650         }
 651         s->s_dev = MKDEV(0, dev & MINORMASK);
 652         s->s_bdi = &noop_backing_dev_info;
 653         return 0;
 654 }
 655
 656 EXPORT_SYMBOL(set_anon_super);
 657
 658 void kill_anon_super(struct super_block *sb)
 659 {
 660         int slot = MINOR(sb->s_dev);
 661
 662         generic_shutdown_super(sb);
 663         spin_lock(&unnamed_dev_lock);
 664         ida_remove(&unnamed_dev_ida, slot);
 665         if (slot < unnamed_dev_start)
 666                 unnamed_dev_start = slot;
 667         spin_unlock(&unnamed_dev_lock);
 668 }
 669
 670 EXPORT_SYMBOL(kill_anon_super);
 671
 672 void kill_litter_super(struct super_block *sb)
 673 {
 674         if (sb->s_root)
 675                 d_genocide(sb->s_root);
 676         kill_anon_super(sb);
 677 }
 678
 679 EXPORT_SYMBOL(kill_litter_super);
 680
 681 static int ns_test_super(struct super_block *sb, void *data)
 682 {
 683         return sb->s_fs_info == data;
 684 }
 685
 686 static int ns_set_super(struct super_block *sb, void *data)
 687 {
 688         sb->s_fs_info = data;
 689         return set_anon_super(sb, NULL);
 690 }
 691
 692 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
 693         int (*fill_super)(struct super_block *, void *, int),
 694         struct vfsmount *mnt)
 695 {
 696         struct super_block *sb;
 697
 698         sb = sget(fs_type, ns_test_super, ns_set_super, data);
 699         if (IS_ERR(sb))
 700                 return PTR_ERR(sb);
 701
 702         if (!sb->s_root) {
 703                 int err;
 704                 sb->s_flags = flags;
 705                 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
 706                 if (err) {
 707                         deactivate_locked_super(sb);
 708                         return err;
 709                 }
 710
 711                 sb->s_flags |= MS_ACTIVE;
 712         }
 713
 714         simple_set_mnt(mnt, sb);
 715         return 0;
 716 }
 717
 718 EXPORT_SYMBOL(get_sb_ns);
 719
 720 #ifdef CONFIG_BLOCK
 721 static int set_bdev_super(struct super_block *s, void *data)
 722 {
 723         s->s_bdev = data;
 724         s->s_dev = s->s_bdev->bd_dev;
 725
 726         /*
 727          * We set the bdi here to the queue backing, file systems can
 728          * overwrite this in ->fill_super()
 729          */
 730         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
 731         return 0;
 732 }
 733
 734 static int test_bdev_super(struct super_block *s, void *data)
 735 {
 736         return (void *)s->s_bdev == data;
 737 }
 738
 739 int get_sb_bdev(struct file_system_type *fs_type,
 740         int flags, const char *dev_name, void *data,
 741         int (*fill_super)(struct super_block *, void *, int),
 742         struct vfsmount *mnt)
 743 {
 744         struct block_device *bdev;
 745         struct super_block *s;
 746         fmode_t mode = FMODE_READ;
 747         int error = 0;
 748
 749         if (!(flags & MS_RDONLY))
 750                 mode |= FMODE_WRITE;
 751
 752         bdev = open_bdev_exclusive(dev_name, mode, fs_type);
 753         if (IS_ERR(bdev))
 754                 return PTR_ERR(bdev);
 755
 756         /*
 757          * once the super is inserted into the list by sget, s_umount
 758          * will protect the lockfs code from trying to start a snapshot
 759          * while we are mounting
 760          */
 761         mutex_lock(&bdev->bd_fsfreeze_mutex);
 762         if (bdev->bd_fsfreeze_count > 0) {
 763                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
 764                 error = -EBUSY;
 765                 goto error_bdev;
 766         }
 767         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
 768         mutex_unlock(&bdev->bd_fsfreeze_mutex);
 769         if (IS_ERR(s))
 770                 goto error_s;
 771
 772         if (s->s_root) {
 773                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 774                         deactivate_locked_super(s);
 775                         error = -EBUSY;
 776                         goto error_bdev;
 777                 }
 778
 779                 close_bdev_exclusive(bdev, mode);
 780         } else {
 781                 char b[BDEVNAME_SIZE];
 782
 783                 s->s_flags = flags;
 784                 s->s_mode = mode;
 785                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 786                 sb_set_blocksize(s, block_size(bdev));
 787                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 788                 if (error) {
 789                         deactivate_locked_super(s);
 790                         goto error;
 791                 }
 792
 793                 s->s_flags |= MS_ACTIVE;
 794                 bdev->bd_super = s;
 795         }
 796
 797         simple_set_mnt(mnt, s);
 798         return 0;
 799
 800 error_s:
 801         error = PTR_ERR(s);
 802 error_bdev:
 803         close_bdev_exclusive(bdev, mode);
 804 error:
 805         return error;
 806 }
 807
 808 EXPORT_SYMBOL(get_sb_bdev);
 809
 810 void kill_block_super(struct super_block *sb)
 811 {
 812         struct block_device *bdev = sb->s_bdev;
 813         fmode_t mode = sb->s_mode;
 814
 815         bdev->bd_super = NULL;
 816         generic_shutdown_super(sb);
 817         sync_blockdev(bdev);
 818         close_bdev_exclusive(bdev, mode);
 819 }
 820
 821 EXPORT_SYMBOL(kill_block_super);
 822 #endif
 823
 824 int get_sb_nodev(struct file_system_type *fs_type,
 825         int flags, void *data,
 826         int (*fill_super)(struct super_block *, void *, int),
 827         struct vfsmount *mnt)
 828 {
 829         int error;
 830         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
 831
 832         if (IS_ERR(s))
 833                 return PTR_ERR(s);
 834
 835         s->s_flags = flags;
 836
 837         error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 838         if (error) {
 839                 deactivate_locked_super(s);
 840                 return error;
 841         }
 842         s->s_flags |= MS_ACTIVE;
 843         simple_set_mnt(mnt, s);
 844         return 0;
 845 }
 846
 847 EXPORT_SYMBOL(get_sb_nodev);
 848
 849 static int compare_single(struct super_block *s, void *p)
 850 {
 851         return 1;
 852 }
 853
 854 int get_sb_single(struct file_system_type *fs_type,
 855         int flags, void *data,
 856         int (*fill_super)(struct super_block *, void *, int),
 857         struct vfsmount *mnt)
 858 {
 859         struct super_block *s;
 860         int error;
 861
 862         s = sget(fs_type, compare_single, set_anon_super, NULL);
 863         if (IS_ERR(s))
 864                 return PTR_ERR(s);
 865         if (!s->s_root) {
 866                 s->s_flags = flags;
 867                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 868                 if (error) {
 869                         deactivate_locked_super(s);
 870                         return error;
 871                 }
 872                 s->s_flags |= MS_ACTIVE;
 873         } else {
 874                 do_remount_sb(s, flags, data, 0);
 875         }
 876         simple_set_mnt(mnt, s);
 877         return 0;
 878 }
 879
 880 EXPORT_SYMBOL(get_sb_single);
 881
 882 struct vfsmount *
 883 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 884 {
 885         struct vfsmount *mnt;
 886         char *secdata = NULL;
 887         int error;
 888
 889         if (!type)
 890                 return ERR_PTR(-ENODEV);
 891
 892         error = -ENOMEM;
 893         mnt = alloc_vfsmnt(name);
 894         if (!mnt)
 895                 goto out;
 896
 897         if (flags & MS_KERNMOUNT)
 898                 mnt->mnt_flags = MNT_INTERNAL;
 899
 900         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
 901                 secdata = alloc_secdata();
 902                 if (!secdata)
 903                         goto out_mnt;
 904
 905                 error = security_sb_copy_data(data, secdata);
 906                 if (error)
 907                         goto out_free_secdata;
 908         }
 909
 910         error = type->get_sb(type, flags, name, data, mnt);
 911         if (error < 0)
 912                 goto out_free_secdata;
 913         BUG_ON(!mnt->mnt_sb);
 914         WARN_ON(!mnt->mnt_sb->s_bdi);
 915
 916         error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
 917         if (error)
 918                 goto out_sb;
 919
 920         /*
 921          * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
 922          * but s_maxbytes was an unsigned long long for many releases. Throw
 923          * this warning for a little while to try and catch filesystems that
 924          * violate this rule. This warning should be either removed or
 925          * converted to a BUG() in 2.6.34.
 926          */
 927         WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
 928                 "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
 929
 930         mnt->mnt_mountpoint = mnt->mnt_root;
 931         mnt->mnt_parent = mnt;
 932         up_write(&mnt->mnt_sb->s_umount);
 933         free_secdata(secdata);
 934         return mnt;
 935 out_sb:
 936         dput(mnt->mnt_root);
 937         deactivate_locked_super(mnt->mnt_sb);
 938 out_free_secdata:
 939         free_secdata(secdata);
 940 out_mnt:
 941         free_vfsmnt(mnt);
 942 out:
 943         return ERR_PTR(error);
 944 }
 945
 946 EXPORT_SYMBOL_GPL(vfs_kern_mount);
 947
 948 /**
 949  * freeze_super -- lock the filesystem and force it into a consistent state
 950  * @super: the super to lock
 951  *
 952  * Syncs the super to make sure the filesystem is consistent and calls the fs's
 953  * freeze_fs.  Subsequent calls to this without first thawing the fs will return
 954  * -EBUSY.
 955  */
 956 int freeze_super(struct super_block *sb)
 957 {
 958         int ret;
 959
 960         atomic_inc(&sb->s_active);
 961         down_write(&sb->s_umount);
 962         if (sb->s_frozen) {
 963                 deactivate_locked_super(sb);
 964                 return -EBUSY;
 965         }
 966
 967         if (sb->s_flags & MS_RDONLY) {
 968                 sb->s_frozen = SB_FREEZE_TRANS;
 969                 smp_wmb();
 970                 up_write(&sb->s_umount);
 971                 return 0;
 972         }
 973
 974         sb->s_frozen = SB_FREEZE_WRITE;
 975         smp_wmb();
 976
 977         sync_filesystem(sb);
 978
 979         sb->s_frozen = SB_FREEZE_TRANS;
 980         smp_wmb();
 981
 982         sync_blockdev(sb->s_bdev);
 983         if (sb->s_op->freeze_fs) {
 984                 ret = sb->s_op->freeze_fs(sb);
 985                 if (ret) {
 986                         printk(KERN_ERR
 987                                 "VFS:Filesystem freeze failed\n");
 988                         sb->s_frozen = SB_UNFROZEN;
 989                         deactivate_locked_super(sb);
 990                         return ret;
 991                 }
 992         }
 993         up_write(&sb->s_umount);
 994         return 0;
 995 }
 996 EXPORT_SYMBOL(freeze_super);
 997
 998 /**
 999  * thaw_super -- unlock filesystem
1000  * @sb: the super to thaw
1001  *
1002  * Unlocks the filesystem and marks it writeable again after freeze_super().
1003  */
1004 int thaw_super(struct super_block *sb)
1005 {
1006         int error;
1007
1008         down_write(&sb->s_umount);
1009         if (sb->s_frozen == SB_UNFROZEN) {
1010                 up_write(&sb->s_umount);
1011                 return -EINVAL;
1012         }
1013
1014         if (sb->s_flags & MS_RDONLY)
1015                 goto out;
1016
1017         if (sb->s_op->unfreeze_fs) {
1018                 error = sb->s_op->unfreeze_fs(sb);
1019                 if (error) {
1020                         printk(KERN_ERR
1021                                 "VFS:Filesystem thaw failed\n");
1022                         sb->s_frozen = SB_FREEZE_TRANS;
1023                         up_write(&sb->s_umount);
1024                         return error;
1025                 }
1026         }
1027
1028 out:
1029         sb->s_frozen = SB_UNFROZEN;
1030         smp_wmb();
1031         wake_up(&sb->s_wait_unfrozen);
1032         deactivate_locked_super(sb);
1033
1034         return 0;
1035 }
1036 EXPORT_SYMBOL(thaw_super);
1037
1038 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1039 {
1040         int err;
1041         const char *subtype = strchr(fstype, '.');
1042         if (subtype) {
1043                 subtype++;
1044                 err = -EINVAL;
1045                 if (!subtype[0])
1046                         goto err;
1047         } else
1048                 subtype = "";
1049
1050         mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1051         err = -ENOMEM;
1052         if (!mnt->mnt_sb->s_subtype)
1053                 goto err;
1054         return mnt;
1055
1056  err:
1057         mntput(mnt);
1058         return ERR_PTR(err);
1059 }
1060
1061 struct vfsmount *
1062 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1063 {
1064         struct file_system_type *type = get_fs_type(fstype);
1065         struct vfsmount *mnt;
1066         if (!type)
1067                 return ERR_PTR(-ENODEV);
1068         mnt = vfs_kern_mount(type, flags, name, data);
1069         if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1070             !mnt->mnt_sb->s_subtype)
1071                 mnt = fs_set_subtype(mnt, fstype);
1072         put_filesystem(type);
1073         return mnt;
1074 }
1075 EXPORT_SYMBOL_GPL(do_kern_mount);
1076
1077 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1078 {
1079         return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1080 }
1081
1082 EXPORT_SYMBOL_GPL(kern_mount_data);