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/init.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/acct.h>
  28 #include <linux/blkdev.h>
  29 #include <linux/quotaops.h>
  30 #include <linux/namei.h>
  31 #include <linux/mount.h>
  32 #include <linux/security.h>
  33 #include <linux/syscalls.h>
  34 #include <linux/vfs.h>
  35 #include <linux/writeback.h>            /* for the emergency remount stuff */
  36 #include <linux/idr.h>
  37 #include <linux/kobject.h>
  38 #include <linux/mutex.h>
  39 #include <linux/file.h>
  40 #include <asm/uaccess.h>
  41 #include "internal.h"
  42
  43
  44 LIST_HEAD(super_blocks);
  45 DEFINE_SPINLOCK(sb_lock);
  46
  47 /**
  48  *      alloc_super     -       create new superblock
  49  *      @type:  filesystem type superblock should belong to
  50  *
  51  *      Allocates and initializes a new &struct super_block.  alloc_super()
  52  *      returns a pointer new superblock or %NULL if allocation had failed.
  53  */
  54 static struct super_block *alloc_super(struct file_system_type *type)
  55 {
  56         struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
  57         static struct super_operations default_op;
  58
  59         if (s) {
  60                 if (security_sb_alloc(s)) {
  61                         kfree(s);
  62                         s = NULL;
  63                         goto out;
  64                 }
  65                 INIT_LIST_HEAD(&s->s_dirty);
  66                 INIT_LIST_HEAD(&s->s_io);
  67                 INIT_LIST_HEAD(&s->s_more_io);
  68                 INIT_LIST_HEAD(&s->s_files);
  69                 INIT_LIST_HEAD(&s->s_instances);
  70                 INIT_HLIST_HEAD(&s->s_anon);
  71                 INIT_LIST_HEAD(&s->s_inodes);
  72                 INIT_LIST_HEAD(&s->s_dentry_lru);
  73                 init_rwsem(&s->s_umount);
  74                 mutex_init(&s->s_lock);
  75                 lockdep_set_class(&s->s_umount, &type->s_umount_key);
  76                 /*
  77                  * The locking rules for s_lock are up to the
  78                  * filesystem. For example ext3fs has different
  79                  * lock ordering than usbfs:
  80                  */
  81                 lockdep_set_class(&s->s_lock, &type->s_lock_key);
  82                 /*
  83                  * sget() can have s_umount recursion.
  84                  *
  85                  * When it cannot find a suitable sb, it allocates a new
  86                  * one (this one), and tries again to find a suitable old
  87                  * one.
  88                  *
  89                  * In case that succeeds, it will acquire the s_umount
  90                  * lock of the old one. Since these are clearly distrinct
  91                  * locks, and this object isn't exposed yet, there's no
  92                  * risk of deadlocks.
  93                  *
  94                  * Annotate this by putting this lock in a different
  95                  * subclass.
  96                  */
  97                 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
  98                 s->s_count = S_BIAS;
  99                 atomic_set(&s->s_active, 1);
 100                 mutex_init(&s->s_vfs_rename_mutex);
 101                 mutex_init(&s->s_dquot.dqio_mutex);
 102                 mutex_init(&s->s_dquot.dqonoff_mutex);
 103                 init_rwsem(&s->s_dquot.dqptr_sem);
 104                 init_waitqueue_head(&s->s_wait_unfrozen);
 105                 s->s_maxbytes = MAX_NON_LFS;
 106                 s->dq_op = sb_dquot_ops;
 107                 s->s_qcop = sb_quotactl_ops;
 108                 s->s_op = &default_op;
 109                 s->s_time_gran = 1000000000;
 110         }
 111 out:
 112         return s;
 113 }
 114
 115 /**
 116  *      destroy_super   -       frees a superblock
 117  *      @s: superblock to free
 118  *
 119  *      Frees a superblock.
 120  */
 121 static inline void destroy_super(struct super_block *s)
 122 {
 123         security_sb_free(s);
 124         kfree(s->s_subtype);
 125         kfree(s->s_options);
 126         kfree(s);
 127 }
 128
 129 /* Superblock refcounting  */
 130
 131 /*
 132  * Drop a superblock's refcount.  Returns non-zero if the superblock was
 133  * destroyed.  The caller must hold sb_lock.
 134  */
 135 static int __put_super(struct super_block *sb)
 136 {
 137         int ret = 0;
 138
 139         if (!--sb->s_count) {
 140                 destroy_super(sb);
 141                 ret = 1;
 142         }
 143         return ret;
 144 }
 145
 146 /*
 147  * Drop a superblock's refcount.
 148  * Returns non-zero if the superblock is about to be destroyed and
 149  * at least is already removed from super_blocks list, so if we are
 150  * making a loop through super blocks then we need to restart.
 151  * The caller must hold sb_lock.
 152  */
 153 int __put_super_and_need_restart(struct super_block *sb)
 154 {
 155         /* check for race with generic_shutdown_super() */
 156         if (list_empty(&sb->s_list)) {
 157                 /* super block is removed, need to restart... */
 158                 __put_super(sb);
 159                 return 1;
 160         }
 161         /* can't be the last, since s_list is still in use */
 162         sb->s_count--;
 163         BUG_ON(sb->s_count == 0);
 164         return 0;
 165 }
 166
 167 /**
 168  *      put_super       -       drop a temporary reference to superblock
 169  *      @sb: superblock in question
 170  *
 171  *      Drops a temporary reference, frees superblock if there's no
 172  *      references left.
 173  */
 174 static void put_super(struct super_block *sb)
 175 {
 176         spin_lock(&sb_lock);
 177         __put_super(sb);
 178         spin_unlock(&sb_lock);
 179 }
 180
 181
 182 /**
 183  *      deactivate_super        -       drop an active reference to superblock
 184  *      @s: superblock to deactivate
 185  *
 186  *      Drops an active reference to superblock, acquiring a temprory one if
 187  *      there is no active references left.  In that case we lock superblock,
 188  *      tell fs driver to shut it down and drop the temporary reference we
 189  *      had just acquired.
 190  */
 191 void deactivate_super(struct super_block *s)
 192 {
 193         struct file_system_type *fs = s->s_type;
 194         if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
 195                 s->s_count -= S_BIAS-1;
 196                 spin_unlock(&sb_lock);
 197                 vfs_dq_off(s, 0);
 198                 down_write(&s->s_umount);
 199                 fs->kill_sb(s);
 200                 put_filesystem(fs);
 201                 put_super(s);
 202         }
 203 }
 204
 205 EXPORT_SYMBOL(deactivate_super);
 206
 207 /**
 208  *      deactivate_locked_super -       drop an active reference to superblock
 209  *      @s: superblock to deactivate
 210  *
 211  *      Equivalent of up_write(&s->s_umount); deactivate_super(s);, except that
 212  *      it does not unlock it until it's all over.  As the result, it's safe to
 213  *      use to dispose of new superblock on ->get_sb() failure exits - nobody
 214  *      will see the sucker until it's all over.  Equivalent using up_write +
 215  *      deactivate_super is safe for that purpose only if superblock is either
 216  *      safe to use or has NULL ->s_root when we unlock.
 217  */
 218 void deactivate_locked_super(struct super_block *s)
 219 {
 220         struct file_system_type *fs = s->s_type;
 221         if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
 222                 s->s_count -= S_BIAS-1;
 223                 spin_unlock(&sb_lock);
 224                 vfs_dq_off(s, 0);
 225                 fs->kill_sb(s);
 226                 put_filesystem(fs);
 227                 put_super(s);
 228         } else {
 229                 up_write(&s->s_umount);
 230         }
 231 }
 232
 233 EXPORT_SYMBOL(deactivate_locked_super);
 234
 235 /**
 236  *      grab_super - acquire an active reference
 237  *      @s: reference we are trying to make active
 238  *
 239  *      Tries to acquire an active reference.  grab_super() is used when we
 240  *      had just found a superblock in super_blocks or fs_type->fs_supers
 241  *      and want to turn it into a full-blown active reference.  grab_super()
 242  *      is called with sb_lock held and drops it.  Returns 1 in case of
 243  *      success, 0 if we had failed (superblock contents was already dead or
 244  *      dying when grab_super() had been called).
 245  */
 246 static int grab_super(struct super_block *s) __releases(sb_lock)
 247 {
 248         s->s_count++;
 249         spin_unlock(&sb_lock);
 250         down_write(&s->s_umount);
 251         if (s->s_root) {
 252                 spin_lock(&sb_lock);
 253                 if (s->s_count > S_BIAS) {
 254                         atomic_inc(&s->s_active);
 255                         s->s_count--;
 256                         spin_unlock(&sb_lock);
 257                         return 1;
 258                 }
 259                 spin_unlock(&sb_lock);
 260         }
 261         up_write(&s->s_umount);
 262         put_super(s);
 263         yield();
 264         return 0;
 265 }
 266
 267 /*
 268  * Superblock locking.  We really ought to get rid of these two.
 269  */
 270 void lock_super(struct super_block * sb)
 271 {
 272         get_fs_excl();
 273         mutex_lock(&sb->s_lock);
 274 }
 275
 276 void unlock_super(struct super_block * sb)
 277 {
 278         put_fs_excl();
 279         mutex_unlock(&sb->s_lock);
 280 }
 281
 282 EXPORT_SYMBOL(lock_super);
 283 EXPORT_SYMBOL(unlock_super);
 284
 285 /**
 286  *      generic_shutdown_super  -       common helper for ->kill_sb()
 287  *      @sb: superblock to kill
 288  *
 289  *      generic_shutdown_super() does all fs-independent work on superblock
 290  *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
 291  *      that need destruction out of superblock, call generic_shutdown_super()
 292  *      and release aforementioned objects.  Note: dentries and inodes _are_
 293  *      taken care of and do not need specific handling.
 294  *
 295  *      Upon calling this function, the filesystem may no longer alter or
 296  *      rearrange the set of dentries belonging to this super_block, nor may it
 297  *      change the attachments of dentries to inodes.
 298  */
 299 void generic_shutdown_super(struct super_block *sb)
 300 {
 301         const struct super_operations *sop = sb->s_op;
 302
 303
 304         if (sb->s_root) {
 305                 shrink_dcache_for_umount(sb);
 306                 sync_filesystem(sb);
 307                 get_fs_excl();
 308                 sb->s_flags &= ~MS_ACTIVE;
 309
 310                 /* bad name - it should be evict_inodes() */
 311                 invalidate_inodes(sb);
 312                 lock_kernel();
 313
 314                 if (sop->put_super)
 315                         sop->put_super(sb);
 316
 317                 /* Forget any remaining inodes */
 318                 if (invalidate_inodes(sb)) {
 319                         printk("VFS: Busy inodes after unmount of %s. "
 320                            "Self-destruct in 5 seconds.  Have a nice day...\n",
 321                            sb->s_id);
 322                 }
 323
 324                 unlock_kernel();
 325                 put_fs_excl();
 326         }
 327         spin_lock(&sb_lock);
 328         /* should be initialized for __put_super_and_need_restart() */
 329         list_del_init(&sb->s_list);
 330         list_del(&sb->s_instances);
 331         spin_unlock(&sb_lock);
 332         up_write(&sb->s_umount);
 333 }
 334
 335 EXPORT_SYMBOL(generic_shutdown_super);
 336
 337 /**
 338  *      sget    -       find or create a superblock
 339  *      @type:  filesystem type superblock should belong to
 340  *      @test:  comparison callback
 341  *      @set:   setup callback
 342  *      @data:  argument to each of them
 343  */
 344 struct super_block *sget(struct file_system_type *type,
 345                         int (*test)(struct super_block *,void *),
 346                         int (*set)(struct super_block *,void *),
 347                         void *data)
 348 {
 349         struct super_block *s = NULL;
 350         struct super_block *old;
 351         int err;
 352
 353 retry:
 354         spin_lock(&sb_lock);
 355         if (test) {
 356                 list_for_each_entry(old, &type->fs_supers, s_instances) {
 357                         if (!test(old, data))
 358                                 continue;
 359                         if (!grab_super(old))
 360                                 goto retry;
 361                         if (s) {
 362                                 up_write(&s->s_umount);
 363                                 destroy_super(s);
 364                         }
 365                         return old;
 366                 }
 367         }
 368         if (!s) {
 369                 spin_unlock(&sb_lock);
 370                 s = alloc_super(type);
 371                 if (!s)
 372                         return ERR_PTR(-ENOMEM);
 373                 goto retry;
 374         }
 375
 376         err = set(s, data);
 377         if (err) {
 378                 spin_unlock(&sb_lock);
 379                 up_write(&s->s_umount);
 380                 destroy_super(s);
 381                 return ERR_PTR(err);
 382         }
 383         s->s_type = type;
 384         strlcpy(s->s_id, type->name, sizeof(s->s_id));
 385         list_add_tail(&s->s_list, &super_blocks);
 386         list_add(&s->s_instances, &type->fs_supers);
 387         spin_unlock(&sb_lock);
 388         get_filesystem(type);
 389         return s;
 390 }
 391
 392 EXPORT_SYMBOL(sget);
 393
 394 void drop_super(struct super_block *sb)
 395 {
 396         up_read(&sb->s_umount);
 397         put_super(sb);
 398 }
 399
 400 EXPORT_SYMBOL(drop_super);
 401
 402 /**
 403  * sync_supers - helper for periodic superblock writeback
 404  *
 405  * Call the write_super method if present on all dirty superblocks in
 406  * the system.  This is for the periodic writeback used by most older
 407  * filesystems.  For data integrity superblock writeback use
 408  * sync_filesystems() instead.
 409  *
 410  * Note: check the dirty flag before waiting, so we don't
 411  * hold up the sync while mounting a device. (The newly
 412  * mounted device won't need syncing.)
 413  */
 414 void sync_supers(void)
 415 {
 416         struct super_block *sb;
 417
 418         spin_lock(&sb_lock);
 419 restart:
 420         list_for_each_entry(sb, &super_blocks, s_list) {
 421                 if (sb->s_op->write_super && sb->s_dirt) {
 422                         sb->s_count++;
 423                         spin_unlock(&sb_lock);
 424
 425                         down_read(&sb->s_umount);
 426                         lock_super(sb);
 427                         if (sb->s_root && sb->s_dirt)
 428                                 sb->s_op->write_super(sb);
 429                         unlock_super(sb);
 430                         up_read(&sb->s_umount);
 431
 432                         spin_lock(&sb_lock);
 433                         if (__put_super_and_need_restart(sb))
 434                                 goto restart;
 435                 }
 436         }
 437         spin_unlock(&sb_lock);
 438 }
 439
 440 /**
 441  *      get_super - get the superblock of a device
 442  *      @bdev: device to get the superblock for
 443  *
 444  *      Scans the superblock list and finds the superblock of the file system
 445  *      mounted on the device given. %NULL is returned if no match is found.
 446  */
 447
 448 struct super_block * get_super(struct block_device *bdev)
 449 {
 450         struct super_block *sb;
 451
 452         if (!bdev)
 453                 return NULL;
 454
 455         spin_lock(&sb_lock);
 456 rescan:
 457         list_for_each_entry(sb, &super_blocks, s_list) {
 458                 if (sb->s_bdev == bdev) {
 459                         sb->s_count++;
 460                         spin_unlock(&sb_lock);
 461                         down_read(&sb->s_umount);
 462                         if (sb->s_root)
 463                                 return sb;
 464                         up_read(&sb->s_umount);
 465                         /* restart only when sb is no longer on the list */
 466                         spin_lock(&sb_lock);
 467                         if (__put_super_and_need_restart(sb))
 468                                 goto rescan;
 469                 }
 470         }
 471         spin_unlock(&sb_lock);
 472         return NULL;
 473 }
 474
 475 EXPORT_SYMBOL(get_super);
 476
 477 struct super_block * user_get_super(dev_t dev)
 478 {
 479         struct super_block *sb;
 480
 481         spin_lock(&sb_lock);
 482 rescan:
 483         list_for_each_entry(sb, &super_blocks, s_list) {
 484                 if (sb->s_dev ==  dev) {
 485                         sb->s_count++;
 486                         spin_unlock(&sb_lock);
 487                         down_read(&sb->s_umount);
 488                         if (sb->s_root)
 489                                 return sb;
 490                         up_read(&sb->s_umount);
 491                         /* restart only when sb is no longer on the list */
 492                         spin_lock(&sb_lock);
 493                         if (__put_super_and_need_restart(sb))
 494                                 goto rescan;
 495                 }
 496         }
 497         spin_unlock(&sb_lock);
 498         return NULL;
 499 }
 500
 501 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
 502 {
 503         struct super_block *s;
 504         struct ustat tmp;
 505         struct kstatfs sbuf;
 506         int err = -EINVAL;
 507
 508         s = user_get_super(new_decode_dev(dev));
 509         if (s == NULL)
 510                 goto out;
 511         err = vfs_statfs(s->s_root, &sbuf);
 512         drop_super(s);
 513         if (err)
 514                 goto out;
 515
 516         memset(&tmp,0,sizeof(struct ustat));
 517         tmp.f_tfree = sbuf.f_bfree;
 518         tmp.f_tinode = sbuf.f_ffree;
 519
 520         err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
 521 out:
 522         return err;
 523 }
 524
 525 /**
 526  *      do_remount_sb - asks filesystem to change mount options.
 527  *      @sb:    superblock in question
 528  *      @flags: numeric part of options
 529  *      @data:  the rest of options
 530  *      @force: whether or not to force the change
 531  *
 532  *      Alters the mount options of a mounted file system.
 533  */
 534 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 535 {
 536         int retval;
 537         int remount_rw;
 538
 539 #ifdef CONFIG_BLOCK
 540         if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
 541                 return -EACCES;
 542 #endif
 543         if (flags & MS_RDONLY)
 544                 acct_auto_close(sb);
 545         shrink_dcache_sb(sb);
 546         sync_filesystem(sb);
 547
 548         /* If we are remounting RDONLY and current sb is read/write,
 549            make sure there are no rw files opened */
 550         if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
 551                 if (force)
 552                         mark_files_ro(sb);
 553                 else if (!fs_may_remount_ro(sb))
 554                         return -EBUSY;
 555                 retval = vfs_dq_off(sb, 1);
 556                 if (retval < 0 && retval != -ENOSYS)
 557                         return -EBUSY;
 558         }
 559         remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
 560
 561         if (sb->s_op->remount_fs) {
 562                 lock_super(sb);
 563                 retval = sb->s_op->remount_fs(sb, &flags, data);
 564                 unlock_super(sb);
 565                 if (retval)
 566                         return retval;
 567         }
 568         sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
 569         if (remount_rw)
 570                 vfs_dq_quota_on_remount(sb);
 571         return 0;
 572 }
 573
 574 static void do_emergency_remount(struct work_struct *work)
 575 {
 576         struct super_block *sb;
 577
 578         spin_lock(&sb_lock);
 579         list_for_each_entry(sb, &super_blocks, s_list) {
 580                 sb->s_count++;
 581                 spin_unlock(&sb_lock);
 582                 down_write(&sb->s_umount);
 583                 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
 584                         /*
 585                          * ->remount_fs needs lock_kernel().
 586                          *
 587                          * What lock protects sb->s_flags??
 588                          */
 589                         lock_kernel();
 590                         do_remount_sb(sb, MS_RDONLY, NULL, 1);
 591                         unlock_kernel();
 592                 }
 593                 up_write(&sb->s_umount);
 594                 put_super(sb);
 595                 spin_lock(&sb_lock);
 596         }
 597         spin_unlock(&sb_lock);
 598         kfree(work);
 599         printk("Emergency Remount complete\n");
 600 }
 601
 602 void emergency_remount(void)
 603 {
 604         struct work_struct *work;
 605
 606         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 607         if (work) {
 608                 INIT_WORK(work, do_emergency_remount);
 609                 schedule_work(work);
 610         }
 611 }
 612
 613 /*
 614  * Unnamed block devices are dummy devices used by virtual
 615  * filesystems which don't use real block-devices.  -- jrs
 616  */
 617
 618 static DEFINE_IDA(unnamed_dev_ida);
 619 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
 620
 621 int set_anon_super(struct super_block *s, void *data)
 622 {
 623         int dev;
 624         int error;
 625
 626  retry:
 627         if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
 628                 return -ENOMEM;
 629         spin_lock(&unnamed_dev_lock);
 630         error = ida_get_new(&unnamed_dev_ida, &dev);
 631         spin_unlock(&unnamed_dev_lock);
 632         if (error == -EAGAIN)
 633                 /* We raced and lost with another CPU. */
 634                 goto retry;
 635         else if (error)
 636                 return -EAGAIN;
 637
 638         if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
 639                 spin_lock(&unnamed_dev_lock);
 640                 ida_remove(&unnamed_dev_ida, dev);
 641                 spin_unlock(&unnamed_dev_lock);
 642                 return -EMFILE;
 643         }
 644         s->s_dev = MKDEV(0, dev & MINORMASK);
 645         return 0;
 646 }
 647
 648 EXPORT_SYMBOL(set_anon_super);
 649
 650 void kill_anon_super(struct super_block *sb)
 651 {
 652         int slot = MINOR(sb->s_dev);
 653
 654         generic_shutdown_super(sb);
 655         spin_lock(&unnamed_dev_lock);
 656         ida_remove(&unnamed_dev_ida, slot);
 657         spin_unlock(&unnamed_dev_lock);
 658 }
 659
 660 EXPORT_SYMBOL(kill_anon_super);
 661
 662 void kill_litter_super(struct super_block *sb)
 663 {
 664         if (sb->s_root)
 665                 d_genocide(sb->s_root);
 666         kill_anon_super(sb);
 667 }
 668
 669 EXPORT_SYMBOL(kill_litter_super);
 670
 671 static int ns_test_super(struct super_block *sb, void *data)
 672 {
 673         return sb->s_fs_info == data;
 674 }
 675
 676 static int ns_set_super(struct super_block *sb, void *data)
 677 {
 678         sb->s_fs_info = data;
 679         return set_anon_super(sb, NULL);
 680 }
 681
 682 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
 683         int (*fill_super)(struct super_block *, void *, int),
 684         struct vfsmount *mnt)
 685 {
 686         struct super_block *sb;
 687
 688         sb = sget(fs_type, ns_test_super, ns_set_super, data);
 689         if (IS_ERR(sb))
 690                 return PTR_ERR(sb);
 691
 692         if (!sb->s_root) {
 693                 int err;
 694                 sb->s_flags = flags;
 695                 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
 696                 if (err) {
 697                         deactivate_locked_super(sb);
 698                         return err;
 699                 }
 700
 701                 sb->s_flags |= MS_ACTIVE;
 702         }
 703
 704         simple_set_mnt(mnt, sb);
 705         return 0;
 706 }
 707
 708 EXPORT_SYMBOL(get_sb_ns);
 709
 710 #ifdef CONFIG_BLOCK
 711 static int set_bdev_super(struct super_block *s, void *data)
 712 {
 713         s->s_bdev = data;
 714         s->s_dev = s->s_bdev->bd_dev;
 715         return 0;
 716 }
 717
 718 static int test_bdev_super(struct super_block *s, void *data)
 719 {
 720         return (void *)s->s_bdev == data;
 721 }
 722
 723 int get_sb_bdev(struct file_system_type *fs_type,
 724         int flags, const char *dev_name, void *data,
 725         int (*fill_super)(struct super_block *, void *, int),
 726         struct vfsmount *mnt)
 727 {
 728         struct block_device *bdev;
 729         struct super_block *s;
 730         fmode_t mode = FMODE_READ;
 731         int error = 0;
 732
 733         if (!(flags & MS_RDONLY))
 734                 mode |= FMODE_WRITE;
 735
 736         bdev = open_bdev_exclusive(dev_name, mode, fs_type);
 737         if (IS_ERR(bdev))
 738                 return PTR_ERR(bdev);
 739
 740         /*
 741          * once the super is inserted into the list by sget, s_umount
 742          * will protect the lockfs code from trying to start a snapshot
 743          * while we are mounting
 744          */
 745         down(&bdev->bd_mount_sem);
 746         s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
 747         up(&bdev->bd_mount_sem);
 748         if (IS_ERR(s))
 749                 goto error_s;
 750
 751         if (s->s_root) {
 752                 if ((flags ^ s->s_flags) & MS_RDONLY) {
 753                         deactivate_locked_super(s);
 754                         error = -EBUSY;
 755                         goto error_bdev;
 756                 }
 757
 758                 close_bdev_exclusive(bdev, mode);
 759         } else {
 760                 char b[BDEVNAME_SIZE];
 761
 762                 s->s_flags = flags;
 763                 s->s_mode = mode;
 764                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
 765                 sb_set_blocksize(s, block_size(bdev));
 766                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 767                 if (error) {
 768                         deactivate_locked_super(s);
 769                         goto error;
 770                 }
 771
 772                 s->s_flags |= MS_ACTIVE;
 773                 bdev->bd_super = s;
 774         }
 775
 776         simple_set_mnt(mnt, s);
 777         return 0;
 778
 779 error_s:
 780         error = PTR_ERR(s);
 781 error_bdev:
 782         close_bdev_exclusive(bdev, mode);
 783 error:
 784         return error;
 785 }
 786
 787 EXPORT_SYMBOL(get_sb_bdev);
 788
 789 void kill_block_super(struct super_block *sb)
 790 {
 791         struct block_device *bdev = sb->s_bdev;
 792         fmode_t mode = sb->s_mode;
 793
 794         bdev->bd_super = NULL;
 795         generic_shutdown_super(sb);
 796         sync_blockdev(bdev);
 797         close_bdev_exclusive(bdev, mode);
 798 }
 799
 800 EXPORT_SYMBOL(kill_block_super);
 801 #endif
 802
 803 int get_sb_nodev(struct file_system_type *fs_type,
 804         int flags, void *data,
 805         int (*fill_super)(struct super_block *, void *, int),
 806         struct vfsmount *mnt)
 807 {
 808         int error;
 809         struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
 810
 811         if (IS_ERR(s))
 812                 return PTR_ERR(s);
 813
 814         s->s_flags = flags;
 815
 816         error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 817         if (error) {
 818                 deactivate_locked_super(s);
 819                 return error;
 820         }
 821         s->s_flags |= MS_ACTIVE;
 822         simple_set_mnt(mnt, s);
 823         return 0;
 824 }
 825
 826 EXPORT_SYMBOL(get_sb_nodev);
 827
 828 static int compare_single(struct super_block *s, void *p)
 829 {
 830         return 1;
 831 }
 832
 833 int get_sb_single(struct file_system_type *fs_type,
 834         int flags, void *data,
 835         int (*fill_super)(struct super_block *, void *, int),
 836         struct vfsmount *mnt)
 837 {
 838         struct super_block *s;
 839         int error;
 840
 841         s = sget(fs_type, compare_single, set_anon_super, NULL);
 842         if (IS_ERR(s))
 843                 return PTR_ERR(s);
 844         if (!s->s_root) {
 845                 s->s_flags = flags;
 846                 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
 847                 if (error) {
 848                         deactivate_locked_super(s);
 849                         return error;
 850                 }
 851                 s->s_flags |= MS_ACTIVE;
 852         }
 853         do_remount_sb(s, flags, data, 0);
 854         simple_set_mnt(mnt, s);
 855         return 0;
 856 }
 857
 858 EXPORT_SYMBOL(get_sb_single);
 859
 860 struct vfsmount *
 861 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 862 {
 863         struct vfsmount *mnt;
 864         char *secdata = NULL;
 865         int error;
 866
 867         if (!type)
 868                 return ERR_PTR(-ENODEV);
 869
 870         error = -ENOMEM;
 871         mnt = alloc_vfsmnt(name);
 872         if (!mnt)
 873                 goto out;
 874
 875         if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
 876                 secdata = alloc_secdata();
 877                 if (!secdata)
 878                         goto out_mnt;
 879
 880                 error = security_sb_copy_data(data, secdata);
 881                 if (error)
 882                         goto out_free_secdata;
 883         }
 884
 885         error = type->get_sb(type, flags, name, data, mnt);
 886         if (error < 0)
 887                 goto out_free_secdata;
 888         BUG_ON(!mnt->mnt_sb);
 889
 890         error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
 891         if (error)
 892                 goto out_sb;
 893
 894         mnt->mnt_mountpoint = mnt->mnt_root;
 895         mnt->mnt_parent = mnt;
 896         up_write(&mnt->mnt_sb->s_umount);
 897         free_secdata(secdata);
 898         return mnt;
 899 out_sb:
 900         dput(mnt->mnt_root);
 901         deactivate_locked_super(mnt->mnt_sb);
 902 out_free_secdata:
 903         free_secdata(secdata);
 904 out_mnt:
 905         free_vfsmnt(mnt);
 906 out:
 907         return ERR_PTR(error);
 908 }
 909
 910 EXPORT_SYMBOL_GPL(vfs_kern_mount);
 911
 912 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
 913 {
 914         int err;
 915         const char *subtype = strchr(fstype, '.');
 916         if (subtype) {
 917                 subtype++;
 918                 err = -EINVAL;
 919                 if (!subtype[0])
 920                         goto err;
 921         } else
 922                 subtype = "";
 923
 924         mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
 925         err = -ENOMEM;
 926         if (!mnt->mnt_sb->s_subtype)
 927                 goto err;
 928         return mnt;
 929
 930  err:
 931         mntput(mnt);
 932         return ERR_PTR(err);
 933 }
 934
 935 struct vfsmount *
 936 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
 937 {
 938         struct file_system_type *type = get_fs_type(fstype);
 939         struct vfsmount *mnt;
 940         if (!type)
 941                 return ERR_PTR(-ENODEV);
 942         mnt = vfs_kern_mount(type, flags, name, data);
 943         if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
 944             !mnt->mnt_sb->s_subtype)
 945                 mnt = fs_set_subtype(mnt, fstype);
 946         put_filesystem(type);
 947         return mnt;
 948 }
 949 EXPORT_SYMBOL_GPL(do_kern_mount);
 950
 951 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
 952 {
 953         return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
 954 }
 955
 956 EXPORT_SYMBOL_GPL(kern_mount_data);