fs/block_dev.c

   1 /*
   2  *  linux/fs/block_dev.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   6  */
   7
   8 #include <linux/init.h>
   9 #include <linux/mm.h>
  10 #include <linux/fcntl.h>
  11 #include <linux/slab.h>
  12 #include <linux/kmod.h>
  13 #include <linux/major.h>
  14 #include <linux/smp_lock.h>
  15 #include <linux/device_cgroup.h>
  16 #include <linux/highmem.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/module.h>
  19 #include <linux/blkpg.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/writeback.h>
  22 #include <linux/mpage.h>
  23 #include <linux/mount.h>
  24 #include <linux/uio.h>
  25 #include <linux/namei.h>
  26 #include <linux/log2.h>
  27 #include <asm/uaccess.h>
  28 #include "internal.h"
  29
  30 struct bdev_inode {
  31         struct block_device bdev;
  32         struct inode vfs_inode;
  33 };
  34
  35 static const struct address_space_operations def_blk_aops;
  36
  37 static inline struct bdev_inode *BDEV_I(struct inode *inode)
  38 {
  39         return container_of(inode, struct bdev_inode, vfs_inode);
  40 }
  41
  42 inline struct block_device *I_BDEV(struct inode *inode)
  43 {
  44         return &BDEV_I(inode)->bdev;
  45 }
  46
  47 EXPORT_SYMBOL(I_BDEV);
  48
  49 static sector_t max_block(struct block_device *bdev)
  50 {
  51         sector_t retval = ~((sector_t)0);
  52         loff_t sz = i_size_read(bdev->bd_inode);
  53
  54         if (sz) {
  55                 unsigned int size = block_size(bdev);
  56                 unsigned int sizebits = blksize_bits(size);
  57                 retval = (sz >> sizebits);
  58         }
  59         return retval;
  60 }
  61
  62 /* Kill _all_ buffers and pagecache , dirty or not.. */
  63 static void kill_bdev(struct block_device *bdev)
  64 {
  65         if (bdev->bd_inode->i_mapping->nrpages == 0)
  66                 return;
  67         invalidate_bh_lrus();
  68         truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  69 }
  70
  71 int set_blocksize(struct block_device *bdev, int size)
  72 {
  73         /* Size must be a power of two, and between 512 and PAGE_SIZE */
  74         if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
  75                 return -EINVAL;
  76
  77         /* Size cannot be smaller than the size supported by the device */
  78         if (size < bdev_hardsect_size(bdev))
  79                 return -EINVAL;
  80
  81         /* Don't change the size if it is same as current */
  82         if (bdev->bd_block_size != size) {
  83                 sync_blockdev(bdev);
  84                 bdev->bd_block_size = size;
  85                 bdev->bd_inode->i_blkbits = blksize_bits(size);
  86                 kill_bdev(bdev);
  87         }
  88         return 0;
  89 }
  90
  91 EXPORT_SYMBOL(set_blocksize);
  92
  93 int sb_set_blocksize(struct super_block *sb, int size)
  94 {
  95         if (set_blocksize(sb->s_bdev, size))
  96                 return 0;
  97         /* If we get here, we know size is power of two
  98          * and it's value is between 512 and PAGE_SIZE */
  99         sb->s_blocksize = size;
 100         sb->s_blocksize_bits = blksize_bits(size);
 101         return sb->s_blocksize;
 102 }
 103
 104 EXPORT_SYMBOL(sb_set_blocksize);
 105
 106 int sb_min_blocksize(struct super_block *sb, int size)
 107 {
 108         int minsize = bdev_hardsect_size(sb->s_bdev);
 109         if (size < minsize)
 110                 size = minsize;
 111         return sb_set_blocksize(sb, size);
 112 }
 113
 114 EXPORT_SYMBOL(sb_min_blocksize);
 115
 116 static int
 117 blkdev_get_block(struct inode *inode, sector_t iblock,
 118                 struct buffer_head *bh, int create)
 119 {
 120         if (iblock >= max_block(I_BDEV(inode))) {
 121                 if (create)
 122                         return -EIO;
 123
 124                 /*
 125                  * for reads, we're just trying to fill a partial page.
 126                  * return a hole, they will have to call get_block again
 127                  * before they can fill it, and they will get -EIO at that
 128                  * time
 129                  */
 130                 return 0;
 131         }
 132         bh->b_bdev = I_BDEV(inode);
 133         bh->b_blocknr = iblock;
 134         set_buffer_mapped(bh);
 135         return 0;
 136 }
 137
 138 static int
 139 blkdev_get_blocks(struct inode *inode, sector_t iblock,
 140                 struct buffer_head *bh, int create)
 141 {
 142         sector_t end_block = max_block(I_BDEV(inode));
 143         unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
 144
 145         if ((iblock + max_blocks) > end_block) {
 146                 max_blocks = end_block - iblock;
 147                 if ((long)max_blocks <= 0) {
 148                         if (create)
 149                                 return -EIO;    /* write fully beyond EOF */
 150                         /*
 151                          * It is a read which is fully beyond EOF.  We return
 152                          * a !buffer_mapped buffer
 153                          */
 154                         max_blocks = 0;
 155                 }
 156         }
 157
 158         bh->b_bdev = I_BDEV(inode);
 159         bh->b_blocknr = iblock;
 160         bh->b_size = max_blocks << inode->i_blkbits;
 161         if (max_blocks)
 162                 set_buffer_mapped(bh);
 163         return 0;
 164 }
 165
 166 static ssize_t
 167 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 168                         loff_t offset, unsigned long nr_segs)
 169 {
 170         struct file *file = iocb->ki_filp;
 171         struct inode *inode = file->f_mapping->host;
 172
 173         return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
 174                                 iov, offset, nr_segs, blkdev_get_blocks, NULL);
 175 }
 176
 177 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 178 {
 179         return block_write_full_page(page, blkdev_get_block, wbc);
 180 }
 181
 182 static int blkdev_readpage(struct file * file, struct page * page)
 183 {
 184         return block_read_full_page(page, blkdev_get_block);
 185 }
 186
 187 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 188                         loff_t pos, unsigned len, unsigned flags,
 189                         struct page **pagep, void **fsdata)
 190 {
 191         *pagep = NULL;
 192         return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
 193                                 blkdev_get_block);
 194 }
 195
 196 static int blkdev_write_end(struct file *file, struct address_space *mapping,
 197                         loff_t pos, unsigned len, unsigned copied,
 198                         struct page *page, void *fsdata)
 199 {
 200         int ret;
 201         ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 202
 203         unlock_page(page);
 204         page_cache_release(page);
 205
 206         return ret;
 207 }
 208
 209 /*
 210  * private llseek:
 211  * for a block special file file->f_path.dentry->d_inode->i_size is zero
 212  * so we compute the size by hand (just as in block_read/write above)
 213  */
 214 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 215 {
 216         struct inode *bd_inode = file->f_mapping->host;
 217         loff_t size;
 218         loff_t retval;
 219
 220         mutex_lock(&bd_inode->i_mutex);
 221         size = i_size_read(bd_inode);
 222
 223         switch (origin) {
 224                 case 2:
 225                         offset += size;
 226                         break;
 227                 case 1:
 228                         offset += file->f_pos;
 229         }
 230         retval = -EINVAL;
 231         if (offset >= 0 && offset <= size) {
 232                 if (offset != file->f_pos) {
 233                         file->f_pos = offset;
 234                 }
 235                 retval = offset;
 236         }
 237         mutex_unlock(&bd_inode->i_mutex);
 238         return retval;
 239 }
 240
 241 /*
 242  *      Filp is never NULL; the only case when ->fsync() is called with
 243  *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 244  */
 245
 246 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 247 {
 248         return sync_blockdev(I_BDEV(filp->f_mapping->host));
 249 }
 250
 251 /*
 252  * pseudo-fs
 253  */
 254
 255 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 256 static struct kmem_cache * bdev_cachep __read_mostly;
 257
 258 static struct inode *bdev_alloc_inode(struct super_block *sb)
 259 {
 260         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 261         if (!ei)
 262                 return NULL;
 263         return &ei->vfs_inode;
 264 }
 265
 266 static void bdev_destroy_inode(struct inode *inode)
 267 {
 268         struct bdev_inode *bdi = BDEV_I(inode);
 269
 270         bdi->bdev.bd_inode_backing_dev_info = NULL;
 271         kmem_cache_free(bdev_cachep, bdi);
 272 }
 273
 274 static void init_once(void *foo)
 275 {
 276         struct bdev_inode *ei = (struct bdev_inode *) foo;
 277         struct block_device *bdev = &ei->bdev;
 278
 279         memset(bdev, 0, sizeof(*bdev));
 280         mutex_init(&bdev->bd_mutex);
 281         sema_init(&bdev->bd_mount_sem, 1);
 282         INIT_LIST_HEAD(&bdev->bd_inodes);
 283         INIT_LIST_HEAD(&bdev->bd_list);
 284 #ifdef CONFIG_SYSFS
 285         INIT_LIST_HEAD(&bdev->bd_holder_list);
 286 #endif
 287         inode_init_once(&ei->vfs_inode);
 288         /* Initialize mutex for freeze. */
 289         mutex_init(&bdev->bd_fsfreeze_mutex);
 290 }
 291
 292 static inline void __bd_forget(struct inode *inode)
 293 {
 294         list_del_init(&inode->i_devices);
 295         inode->i_bdev = NULL;
 296         inode->i_mapping = &inode->i_data;
 297 }
 298
 299 static void bdev_clear_inode(struct inode *inode)
 300 {
 301         struct block_device *bdev = &BDEV_I(inode)->bdev;
 302         struct list_head *p;
 303         spin_lock(&bdev_lock);
 304         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 305                 __bd_forget(list_entry(p, struct inode, i_devices));
 306         }
 307         list_del_init(&bdev->bd_list);
 308         spin_unlock(&bdev_lock);
 309 }
 310
 311 static const struct super_operations bdev_sops = {
 312         .statfs = simple_statfs,
 313         .alloc_inode = bdev_alloc_inode,
 314         .destroy_inode = bdev_destroy_inode,
 315         .drop_inode = generic_delete_inode,
 316         .clear_inode = bdev_clear_inode,
 317 };
 318
 319 static int bd_get_sb(struct file_system_type *fs_type,
 320         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 321 {
 322         return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
 323 }
 324
 325 static struct file_system_type bd_type = {
 326         .name           = "bdev",
 327         .get_sb         = bd_get_sb,
 328         .kill_sb        = kill_anon_super,
 329 };
 330
 331 struct super_block *blockdev_superblock __read_mostly;
 332
 333 void __init bdev_cache_init(void)
 334 {
 335         int err;
 336         struct vfsmount *bd_mnt;
 337
 338         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 339                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 340                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 341                         init_once);
 342         err = register_filesystem(&bd_type);
 343         if (err)
 344                 panic("Cannot register bdev pseudo-fs");
 345         bd_mnt = kern_mount(&bd_type);
 346         if (IS_ERR(bd_mnt))
 347                 panic("Cannot create bdev pseudo-fs");
 348         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 349 }
 350
 351 /*
 352  * Most likely _very_ bad one - but then it's hardly critical for small
 353  * /dev and can be fixed when somebody will need really large one.
 354  * Keep in mind that it will be fed through icache hash function too.
 355  */
 356 static inline unsigned long hash(dev_t dev)
 357 {
 358         return MAJOR(dev)+MINOR(dev);
 359 }
 360
 361 static int bdev_test(struct inode *inode, void *data)
 362 {
 363         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 364 }
 365
 366 static int bdev_set(struct inode *inode, void *data)
 367 {
 368         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 369         return 0;
 370 }
 371
 372 static LIST_HEAD(all_bdevs);
 373
 374 struct block_device *bdget(dev_t dev)
 375 {
 376         struct block_device *bdev;
 377         struct inode *inode;
 378
 379         inode = iget5_locked(blockdev_superblock, hash(dev),
 380                         bdev_test, bdev_set, &dev);
 381
 382         if (!inode)
 383                 return NULL;
 384
 385         bdev = &BDEV_I(inode)->bdev;
 386
 387         if (inode->i_state & I_NEW) {
 388                 bdev->bd_contains = NULL;
 389                 bdev->bd_inode = inode;
 390                 bdev->bd_block_size = (1 << inode->i_blkbits);
 391                 bdev->bd_part_count = 0;
 392                 bdev->bd_invalidated = 0;
 393                 inode->i_mode = S_IFBLK;
 394                 inode->i_rdev = dev;
 395                 inode->i_bdev = bdev;
 396                 inode->i_data.a_ops = &def_blk_aops;
 397                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 398                 inode->i_data.backing_dev_info = &default_backing_dev_info;
 399                 spin_lock(&bdev_lock);
 400                 list_add(&bdev->bd_list, &all_bdevs);
 401                 spin_unlock(&bdev_lock);
 402                 unlock_new_inode(inode);
 403         }
 404         return bdev;
 405 }
 406
 407 EXPORT_SYMBOL(bdget);
 408
 409 long nr_blockdev_pages(void)
 410 {
 411         struct block_device *bdev;
 412         long ret = 0;
 413         spin_lock(&bdev_lock);
 414         list_for_each_entry(bdev, &all_bdevs, bd_list) {
 415                 ret += bdev->bd_inode->i_mapping->nrpages;
 416         }
 417         spin_unlock(&bdev_lock);
 418         return ret;
 419 }
 420
 421 void bdput(struct block_device *bdev)
 422 {
 423         iput(bdev->bd_inode);
 424 }
 425
 426 EXPORT_SYMBOL(bdput);
 427
 428 static struct block_device *bd_acquire(struct inode *inode)
 429 {
 430         struct block_device *bdev;
 431
 432         spin_lock(&bdev_lock);
 433         bdev = inode->i_bdev;
 434         if (bdev) {
 435                 atomic_inc(&bdev->bd_inode->i_count);
 436                 spin_unlock(&bdev_lock);
 437                 return bdev;
 438         }
 439         spin_unlock(&bdev_lock);
 440
 441         bdev = bdget(inode->i_rdev);
 442         if (bdev) {
 443                 spin_lock(&bdev_lock);
 444                 if (!inode->i_bdev) {
 445                         /*
 446                          * We take an additional bd_inode->i_count for inode,
 447                          * and it's released in clear_inode() of inode.
 448                          * So, we can access it via ->i_mapping always
 449                          * without igrab().
 450                          */
 451                         atomic_inc(&bdev->bd_inode->i_count);
 452                         inode->i_bdev = bdev;
 453                         inode->i_mapping = bdev->bd_inode->i_mapping;
 454                         list_add(&inode->i_devices, &bdev->bd_inodes);
 455                 }
 456                 spin_unlock(&bdev_lock);
 457         }
 458         return bdev;
 459 }
 460
 461 /* Call when you free inode */
 462
 463 void bd_forget(struct inode *inode)
 464 {
 465         struct block_device *bdev = NULL;
 466
 467         spin_lock(&bdev_lock);
 468         if (inode->i_bdev) {
 469                 if (!sb_is_blkdev_sb(inode->i_sb))
 470                         bdev = inode->i_bdev;
 471                 __bd_forget(inode);
 472         }
 473         spin_unlock(&bdev_lock);
 474
 475         if (bdev)
 476                 iput(bdev->bd_inode);
 477 }
 478
 479 int bd_claim(struct block_device *bdev, void *holder)
 480 {
 481         int res;
 482         spin_lock(&bdev_lock);
 483
 484         /* first decide result */
 485         if (bdev->bd_holder == holder)
 486                 res = 0;         /* already a holder */
 487         else if (bdev->bd_holder != NULL)
 488                 res = -EBUSY;    /* held by someone else */
 489         else if (bdev->bd_contains == bdev)
 490                 res = 0;         /* is a whole device which isn't held */
 491
 492         else if (bdev->bd_contains->bd_holder == bd_claim)
 493                 res = 0;         /* is a partition of a device that is being partitioned */
 494         else if (bdev->bd_contains->bd_holder != NULL)
 495                 res = -EBUSY;    /* is a partition of a held device */
 496         else
 497                 res = 0;         /* is a partition of an un-held device */
 498
 499         /* now impose change */
 500         if (res==0) {
 501                 /* note that for a whole device bd_holders
 502                  * will be incremented twice, and bd_holder will
 503                  * be set to bd_claim before being set to holder
 504                  */
 505                 bdev->bd_contains->bd_holders ++;
 506                 bdev->bd_contains->bd_holder = bd_claim;
 507                 bdev->bd_holders++;
 508                 bdev->bd_holder = holder;
 509         }
 510         spin_unlock(&bdev_lock);
 511         return res;
 512 }
 513
 514 EXPORT_SYMBOL(bd_claim);
 515
 516 void bd_release(struct block_device *bdev)
 517 {
 518         spin_lock(&bdev_lock);
 519         if (!--bdev->bd_contains->bd_holders)
 520                 bdev->bd_contains->bd_holder = NULL;
 521         if (!--bdev->bd_holders)
 522                 bdev->bd_holder = NULL;
 523         spin_unlock(&bdev_lock);
 524 }
 525
 526 EXPORT_SYMBOL(bd_release);
 527
 528 #ifdef CONFIG_SYSFS
 529 /*
 530  * Functions for bd_claim_by_kobject / bd_release_from_kobject
 531  *
 532  *     If a kobject is passed to bd_claim_by_kobject()
 533  *     and the kobject has a parent directory,
 534  *     following symlinks are created:
 535  *        o from the kobject to the claimed bdev
 536  *        o from "holders" directory of the bdev to the parent of the kobject
 537  *     bd_release_from_kobject() removes these symlinks.
 538  *
 539  *     Example:
 540  *        If /dev/dm-0 maps to /dev/sda, kobject corresponding to
 541  *        /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
 542  *           /sys/block/dm-0/slaves/sda --> /sys/block/sda
 543  *           /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 544  */
 545
 546 static int add_symlink(struct kobject *from, struct kobject *to)
 547 {
 548         if (!from || !to)
 549                 return 0;
 550         return sysfs_create_link(from, to, kobject_name(to));
 551 }
 552
 553 static void del_symlink(struct kobject *from, struct kobject *to)
 554 {
 555         if (!from || !to)
 556                 return;
 557         sysfs_remove_link(from, kobject_name(to));
 558 }
 559
 560 /*
 561  * 'struct bd_holder' contains pointers to kobjects symlinked by
 562  * bd_claim_by_kobject.
 563  * It's connected to bd_holder_list which is protected by bdev->bd_sem.
 564  */
 565 struct bd_holder {
 566         struct list_head list;  /* chain of holders of the bdev */
 567         int count;              /* references from the holder */
 568         struct kobject *sdir;   /* holder object, e.g. "/block/dm-0/slaves" */
 569         struct kobject *hdev;   /* e.g. "/block/dm-0" */
 570         struct kobject *hdir;   /* e.g. "/block/sda/holders" */
 571         struct kobject *sdev;   /* e.g. "/block/sda" */
 572 };
 573
 574 /*
 575  * Get references of related kobjects at once.
 576  * Returns 1 on success. 0 on failure.
 577  *
 578  * Should call bd_holder_release_dirs() after successful use.
 579  */
 580 static int bd_holder_grab_dirs(struct block_device *bdev,
 581                         struct bd_holder *bo)
 582 {
 583         if (!bdev || !bo)
 584                 return 0;
 585
 586         bo->sdir = kobject_get(bo->sdir);
 587         if (!bo->sdir)
 588                 return 0;
 589
 590         bo->hdev = kobject_get(bo->sdir->parent);
 591         if (!bo->hdev)
 592                 goto fail_put_sdir;
 593
 594         bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
 595         if (!bo->sdev)
 596                 goto fail_put_hdev;
 597
 598         bo->hdir = kobject_get(bdev->bd_part->holder_dir);
 599         if (!bo->hdir)
 600                 goto fail_put_sdev;
 601
 602         return 1;
 603
 604 fail_put_sdev:
 605         kobject_put(bo->sdev);
 606 fail_put_hdev:
 607         kobject_put(bo->hdev);
 608 fail_put_sdir:
 609         kobject_put(bo->sdir);
 610
 611         return 0;
 612 }
 613
 614 /* Put references of related kobjects at once. */
 615 static void bd_holder_release_dirs(struct bd_holder *bo)
 616 {
 617         kobject_put(bo->hdir);
 618         kobject_put(bo->sdev);
 619         kobject_put(bo->hdev);
 620         kobject_put(bo->sdir);
 621 }
 622
 623 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
 624 {
 625         struct bd_holder *bo;
 626
 627         bo = kzalloc(sizeof(*bo), GFP_KERNEL);
 628         if (!bo)
 629                 return NULL;
 630
 631         bo->count = 1;
 632         bo->sdir = kobj;
 633
 634         return bo;
 635 }
 636
 637 static void free_bd_holder(struct bd_holder *bo)
 638 {
 639         kfree(bo);
 640 }
 641
 642 /**
 643  * find_bd_holder - find matching struct bd_holder from the block device
 644  *
 645  * @bdev:       struct block device to be searched
 646  * @bo:         target struct bd_holder
 647  *
 648  * Returns matching entry with @bo in @bdev->bd_holder_list.
 649  * If found, increment the reference count and return the pointer.
 650  * If not found, returns NULL.
 651  */
 652 static struct bd_holder *find_bd_holder(struct block_device *bdev,
 653                                         struct bd_holder *bo)
 654 {
 655         struct bd_holder *tmp;
 656
 657         list_for_each_entry(tmp, &bdev->bd_holder_list, list)
 658                 if (tmp->sdir == bo->sdir) {
 659                         tmp->count++;
 660                         return tmp;
 661                 }
 662
 663         return NULL;
 664 }
 665
 666 /**
 667  * add_bd_holder - create sysfs symlinks for bd_claim() relationship
 668  *
 669  * @bdev:       block device to be bd_claimed
 670  * @bo:         preallocated and initialized by alloc_bd_holder()
 671  *
 672  * Add @bo to @bdev->bd_holder_list, create symlinks.
 673  *
 674  * Returns 0 if symlinks are created.
 675  * Returns -ve if something fails.
 676  */
 677 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
 678 {
 679         int err;
 680
 681         if (!bo)
 682                 return -EINVAL;
 683
 684         if (!bd_holder_grab_dirs(bdev, bo))
 685                 return -EBUSY;
 686
 687         err = add_symlink(bo->sdir, bo->sdev);
 688         if (err)
 689                 return err;
 690
 691         err = add_symlink(bo->hdir, bo->hdev);
 692         if (err) {
 693                 del_symlink(bo->sdir, bo->sdev);
 694                 return err;
 695         }
 696
 697         list_add_tail(&bo->list, &bdev->bd_holder_list);
 698         return 0;
 699 }
 700
 701 /**
 702  * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
 703  *
 704  * @bdev:       block device to be bd_claimed
 705  * @kobj:       holder's kobject
 706  *
 707  * If there is matching entry with @kobj in @bdev->bd_holder_list
 708  * and no other bd_claim() from the same kobject,
 709  * remove the struct bd_holder from the list, delete symlinks for it.
 710  *
 711  * Returns a pointer to the struct bd_holder when it's removed from the list
 712  * and ready to be freed.
 713  * Returns NULL if matching claim isn't found or there is other bd_claim()
 714  * by the same kobject.
 715  */
 716 static struct bd_holder *del_bd_holder(struct block_device *bdev,
 717                                         struct kobject *kobj)
 718 {
 719         struct bd_holder *bo;
 720
 721         list_for_each_entry(bo, &bdev->bd_holder_list, list) {
 722                 if (bo->sdir == kobj) {
 723                         bo->count--;
 724                         BUG_ON(bo->count < 0);
 725                         if (!bo->count) {
 726                                 list_del(&bo->list);
 727                                 del_symlink(bo->sdir, bo->sdev);
 728                                 del_symlink(bo->hdir, bo->hdev);
 729                                 bd_holder_release_dirs(bo);
 730                                 return bo;
 731                         }
 732                         break;
 733                 }
 734         }
 735
 736         return NULL;
 737 }
 738
 739 /**
 740  * bd_claim_by_kobject - bd_claim() with additional kobject signature
 741  *
 742  * @bdev:       block device to be claimed
 743  * @holder:     holder's signature
 744  * @kobj:       holder's kobject
 745  *
 746  * Do bd_claim() and if it succeeds, create sysfs symlinks between
 747  * the bdev and the holder's kobject.
 748  * Use bd_release_from_kobject() when relesing the claimed bdev.
 749  *
 750  * Returns 0 on success. (same as bd_claim())
 751  * Returns errno on failure.
 752  */
 753 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
 754                                 struct kobject *kobj)
 755 {
 756         int err;
 757         struct bd_holder *bo, *found;
 758
 759         if (!kobj)
 760                 return -EINVAL;
 761
 762         bo = alloc_bd_holder(kobj);
 763         if (!bo)
 764                 return -ENOMEM;
 765
 766         mutex_lock(&bdev->bd_mutex);
 767
 768         err = bd_claim(bdev, holder);
 769         if (err)
 770                 goto fail;
 771
 772         found = find_bd_holder(bdev, bo);
 773         if (found)
 774                 goto fail;
 775
 776         err = add_bd_holder(bdev, bo);
 777         if (err)
 778                 bd_release(bdev);
 779         else
 780                 bo = NULL;
 781 fail:
 782         mutex_unlock(&bdev->bd_mutex);
 783         free_bd_holder(bo);
 784         return err;
 785 }
 786
 787 /**
 788  * bd_release_from_kobject - bd_release() with additional kobject signature
 789  *
 790  * @bdev:       block device to be released
 791  * @kobj:       holder's kobject
 792  *
 793  * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
 794  */
 795 static void bd_release_from_kobject(struct block_device *bdev,
 796                                         struct kobject *kobj)
 797 {
 798         if (!kobj)
 799                 return;
 800
 801         mutex_lock(&bdev->bd_mutex);
 802         bd_release(bdev);
 803         free_bd_holder(del_bd_holder(bdev, kobj));
 804         mutex_unlock(&bdev->bd_mutex);
 805 }
 806
 807 /**
 808  * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
 809  *
 810  * @bdev:       block device to be claimed
 811  * @holder:     holder's signature
 812  * @disk:       holder's gendisk
 813  *
 814  * Call bd_claim_by_kobject() with getting @disk->slave_dir.
 815  */
 816 int bd_claim_by_disk(struct block_device *bdev, void *holder,
 817                         struct gendisk *disk)
 818 {
 819         return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
 820 }
 821 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
 822
 823 /**
 824  * bd_release_from_disk - wrapper function for bd_release_from_kobject()
 825  *
 826  * @bdev:       block device to be claimed
 827  * @disk:       holder's gendisk
 828  *
 829  * Call bd_release_from_kobject() and put @disk->slave_dir.
 830  */
 831 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
 832 {
 833         bd_release_from_kobject(bdev, disk->slave_dir);
 834         kobject_put(disk->slave_dir);
 835 }
 836 EXPORT_SYMBOL_GPL(bd_release_from_disk);
 837 #endif
 838
 839 /*
 840  * Tries to open block device by device number.  Use it ONLY if you
 841  * really do not have anything better - i.e. when you are behind a
 842  * truly sucky interface and all you are given is a device number.  _Never_
 843  * to be used for internal purposes.  If you ever need it - reconsider
 844  * your API.
 845  */
 846 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
 847 {
 848         struct block_device *bdev = bdget(dev);
 849         int err = -ENOMEM;
 850         if (bdev)
 851                 err = blkdev_get(bdev, mode);
 852         return err ? ERR_PTR(err) : bdev;
 853 }
 854
 855 EXPORT_SYMBOL(open_by_devnum);
 856
 857 /**
 858  * flush_disk - invalidates all buffer-cache entries on a disk
 859  *
 860  * @bdev:      struct block device to be flushed
 861  *
 862  * Invalidates all buffer-cache entries on a disk. It should be called
 863  * when a disk has been changed -- either by a media change or online
 864  * resize.
 865  */
 866 static void flush_disk(struct block_device *bdev)
 867 {
 868         if (__invalidate_device(bdev)) {
 869                 char name[BDEVNAME_SIZE] = "";
 870
 871                 if (bdev->bd_disk)
 872                         disk_name(bdev->bd_disk, 0, name);
 873                 printk(KERN_WARNING "VFS: busy inodes on changed media or "
 874                        "resized disk %s\n", name);
 875         }
 876
 877         if (!bdev->bd_disk)
 878                 return;
 879         if (disk_partitionable(bdev->bd_disk))
 880                 bdev->bd_invalidated = 1;
 881 }
 882
 883 /**
 884  * check_disk_size_change - checks for disk size change and adjusts bdev size.
 885  * @disk: struct gendisk to check
 886  * @bdev: struct bdev to adjust.
 887  *
 888  * This routine checks to see if the bdev size does not match the disk size
 889  * and adjusts it if it differs.
 890  */
 891 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 892 {
 893         loff_t disk_size, bdev_size;
 894
 895         disk_size = (loff_t)get_capacity(disk) << 9;
 896         bdev_size = i_size_read(bdev->bd_inode);
 897         if (disk_size != bdev_size) {
 898                 char name[BDEVNAME_SIZE];
 899
 900                 disk_name(disk, 0, name);
 901                 printk(KERN_INFO
 902                        "%s: detected capacity change from %lld to %lld\n",
 903                        name, bdev_size, disk_size);
 904                 i_size_write(bdev->bd_inode, disk_size);
 905                 flush_disk(bdev);
 906         }
 907 }
 908 EXPORT_SYMBOL(check_disk_size_change);
 909
 910 /**
 911  * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 912  * @disk: struct gendisk to be revalidated
 913  *
 914  * This routine is a wrapper for lower-level driver's revalidate_disk
 915  * call-backs.  It is used to do common pre and post operations needed
 916  * for all revalidate_disk operations.
 917  */
 918 int revalidate_disk(struct gendisk *disk)
 919 {
 920         struct block_device *bdev;
 921         int ret = 0;
 922
 923         if (disk->fops->revalidate_disk)
 924                 ret = disk->fops->revalidate_disk(disk);
 925
 926         bdev = bdget_disk(disk, 0);
 927         if (!bdev)
 928                 return ret;
 929
 930         mutex_lock(&bdev->bd_mutex);
 931         check_disk_size_change(disk, bdev);
 932         mutex_unlock(&bdev->bd_mutex);
 933         bdput(bdev);
 934         return ret;
 935 }
 936 EXPORT_SYMBOL(revalidate_disk);
 937
 938 /*
 939  * This routine checks whether a removable media has been changed,
 940  * and invalidates all buffer-cache-entries in that case. This
 941  * is a relatively slow routine, so we have to try to minimize using
 942  * it. Thus it is called only upon a 'mount' or 'open'. This
 943  * is the best way of combining speed and utility, I think.
 944  * People changing diskettes in the middle of an operation deserve
 945  * to lose :-)
 946  */
 947 int check_disk_change(struct block_device *bdev)
 948 {
 949         struct gendisk *disk = bdev->bd_disk;
 950         struct block_device_operations * bdops = disk->fops;
 951
 952         if (!bdops->media_changed)
 953                 return 0;
 954         if (!bdops->media_changed(bdev->bd_disk))
 955                 return 0;
 956
 957         flush_disk(bdev);
 958         if (bdops->revalidate_disk)
 959                 bdops->revalidate_disk(bdev->bd_disk);
 960         return 1;
 961 }
 962
 963 EXPORT_SYMBOL(check_disk_change);
 964
 965 void bd_set_size(struct block_device *bdev, loff_t size)
 966 {
 967         unsigned bsize = bdev_hardsect_size(bdev);
 968
 969         bdev->bd_inode->i_size = size;
 970         while (bsize < PAGE_CACHE_SIZE) {
 971                 if (size & bsize)
 972                         break;
 973                 bsize <<= 1;
 974         }
 975         bdev->bd_block_size = bsize;
 976         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 977 }
 978 EXPORT_SYMBOL(bd_set_size);
 979
 980 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
 981
 982 /*
 983  * bd_mutex locking:
 984  *
 985  *  mutex_lock(part->bd_mutex)
 986  *    mutex_lock_nested(whole->bd_mutex, 1)
 987  */
 988
 989 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
 990 {
 991         struct gendisk *disk;
 992         int ret;
 993         int partno;
 994         int perm = 0;
 995
 996         if (mode & FMODE_READ)
 997                 perm |= MAY_READ;
 998         if (mode & FMODE_WRITE)
 999                 perm |= MAY_WRITE;
1000         /*
1001          * hooks: /n/, see "layering violations".
1002          */
1003         ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1004         if (ret != 0) {
1005                 bdput(bdev);
1006                 return ret;
1007         }
1008
1009         lock_kernel();
1010  restart:
1011
1012         ret = -ENXIO;
1013         disk = get_gendisk(bdev->bd_dev, &partno);
1014         if (!disk)
1015                 goto out_unlock_kernel;
1016
1017         mutex_lock_nested(&bdev->bd_mutex, for_part);
1018         if (!bdev->bd_openers) {
1019                 bdev->bd_disk = disk;
1020                 bdev->bd_contains = bdev;
1021                 if (!partno) {
1022                         struct backing_dev_info *bdi;
1023
1024                         ret = -ENXIO;
1025                         bdev->bd_part = disk_get_part(disk, partno);
1026                         if (!bdev->bd_part)
1027                                 goto out_clear;
1028
1029                         if (disk->fops->open) {
1030                                 ret = disk->fops->open(bdev, mode);
1031                                 if (ret == -ERESTARTSYS) {
1032                                         /* Lost a race with 'disk' being
1033                                          * deleted, try again.
1034                                          * See md.c
1035                                          */
1036                                         disk_put_part(bdev->bd_part);
1037                                         bdev->bd_part = NULL;
1038                                         module_put(disk->fops->owner);
1039                                         put_disk(disk);
1040                                         bdev->bd_disk = NULL;
1041                                         mutex_unlock(&bdev->bd_mutex);
1042                                         goto restart;
1043                                 }
1044                                 if (ret)
1045                                         goto out_clear;
1046                         }
1047                         if (!bdev->bd_openers) {
1048                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1049                                 bdi = blk_get_backing_dev_info(bdev);
1050                                 if (bdi == NULL)
1051                                         bdi = &default_backing_dev_info;
1052                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
1053                         }
1054                         if (bdev->bd_invalidated)
1055                                 rescan_partitions(disk, bdev);
1056                 } else {
1057                         struct block_device *whole;
1058                         whole = bdget_disk(disk, 0);
1059                         ret = -ENOMEM;
1060                         if (!whole)
1061                                 goto out_clear;
1062                         BUG_ON(for_part);
1063                         ret = __blkdev_get(whole, mode, 1);
1064                         if (ret)
1065                                 goto out_clear;
1066                         bdev->bd_contains = whole;
1067                         bdev->bd_inode->i_data.backing_dev_info =
1068                            whole->bd_inode->i_data.backing_dev_info;
1069                         bdev->bd_part = disk_get_part(disk, partno);
1070                         if (!(disk->flags & GENHD_FL_UP) ||
1071                             !bdev->bd_part || !bdev->bd_part->nr_sects) {
1072                                 ret = -ENXIO;
1073                                 goto out_clear;
1074                         }
1075                         bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1076                 }
1077         } else {
1078                 put_disk(disk);
1079                 module_put(disk->fops->owner);
1080                 disk = NULL;
1081                 if (bdev->bd_contains == bdev) {
1082                         if (bdev->bd_disk->fops->open) {
1083                                 ret = bdev->bd_disk->fops->open(bdev, mode);
1084                                 if (ret)
1085                                         goto out_unlock_bdev;
1086                         }
1087                         if (bdev->bd_invalidated)
1088                                 rescan_partitions(bdev->bd_disk, bdev);
1089                 }
1090         }
1091         bdev->bd_openers++;
1092         if (for_part)
1093                 bdev->bd_part_count++;
1094         mutex_unlock(&bdev->bd_mutex);
1095         unlock_kernel();
1096         return 0;
1097
1098  out_clear:
1099         disk_put_part(bdev->bd_part);
1100         bdev->bd_disk = NULL;
1101         bdev->bd_part = NULL;
1102         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1103         if (bdev != bdev->bd_contains)
1104                 __blkdev_put(bdev->bd_contains, mode, 1);
1105         bdev->bd_contains = NULL;
1106  out_unlock_bdev:
1107         mutex_unlock(&bdev->bd_mutex);
1108  out_unlock_kernel:
1109         unlock_kernel();
1110
1111         if (disk)
1112                 module_put(disk->fops->owner);
1113         put_disk(disk);
1114         bdput(bdev);
1115
1116         return ret;
1117 }
1118
1119 int blkdev_get(struct block_device *bdev, fmode_t mode)
1120 {
1121         return __blkdev_get(bdev, mode, 0);
1122 }
1123 EXPORT_SYMBOL(blkdev_get);
1124
1125 static int blkdev_open(struct inode * inode, struct file * filp)
1126 {
1127         struct block_device *bdev;
1128         int res;
1129
1130         /*
1131          * Preserve backwards compatibility and allow large file access
1132          * even if userspace doesn't ask for it explicitly. Some mkfs
1133          * binary needs it. We might want to drop this workaround
1134          * during an unstable branch.
1135          */
1136         filp->f_flags |= O_LARGEFILE;
1137
1138         if (filp->f_flags & O_NDELAY)
1139                 filp->f_mode |= FMODE_NDELAY;
1140         if (filp->f_flags & O_EXCL)
1141                 filp->f_mode |= FMODE_EXCL;
1142         if ((filp->f_flags & O_ACCMODE) == 3)
1143                 filp->f_mode |= FMODE_WRITE_IOCTL;
1144
1145         bdev = bd_acquire(inode);
1146         if (bdev == NULL)
1147                 return -ENOMEM;
1148
1149         filp->f_mapping = bdev->bd_inode->i_mapping;
1150
1151         res = blkdev_get(bdev, filp->f_mode);
1152         if (res)
1153                 return res;
1154
1155         if (filp->f_mode & FMODE_EXCL) {
1156                 res = bd_claim(bdev, filp);
1157                 if (res)
1158                         goto out_blkdev_put;
1159         }
1160
1161         return 0;
1162
1163  out_blkdev_put:
1164         blkdev_put(bdev, filp->f_mode);
1165         return res;
1166 }
1167
1168 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1169 {
1170         int ret = 0;
1171         struct gendisk *disk = bdev->bd_disk;
1172         struct block_device *victim = NULL;
1173
1174         mutex_lock_nested(&bdev->bd_mutex, for_part);
1175         lock_kernel();
1176         if (for_part)
1177                 bdev->bd_part_count--;
1178
1179         if (!--bdev->bd_openers) {
1180                 sync_blockdev(bdev);
1181                 kill_bdev(bdev);
1182         }
1183         if (bdev->bd_contains == bdev) {
1184                 if (disk->fops->release)
1185                         ret = disk->fops->release(disk, mode);
1186         }
1187         if (!bdev->bd_openers) {
1188                 struct module *owner = disk->fops->owner;
1189
1190                 put_disk(disk);
1191                 module_put(owner);
1192                 disk_put_part(bdev->bd_part);
1193                 bdev->bd_part = NULL;
1194                 bdev->bd_disk = NULL;
1195                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1196                 if (bdev != bdev->bd_contains)
1197                         victim = bdev->bd_contains;
1198                 bdev->bd_contains = NULL;
1199         }
1200         unlock_kernel();
1201         mutex_unlock(&bdev->bd_mutex);
1202         bdput(bdev);
1203         if (victim)
1204                 __blkdev_put(victim, mode, 1);
1205         return ret;
1206 }
1207
1208 int blkdev_put(struct block_device *bdev, fmode_t mode)
1209 {
1210         return __blkdev_put(bdev, mode, 0);
1211 }
1212 EXPORT_SYMBOL(blkdev_put);
1213
1214 static int blkdev_close(struct inode * inode, struct file * filp)
1215 {
1216         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1217         if (bdev->bd_holder == filp)
1218                 bd_release(bdev);
1219         return blkdev_put(bdev, filp->f_mode);
1220 }
1221
1222 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1223 {
1224         struct block_device *bdev = I_BDEV(file->f_mapping->host);
1225         fmode_t mode = file->f_mode;
1226
1227         /*
1228          * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1229          * to updated it before every ioctl.
1230          */
1231         if (file->f_flags & O_NDELAY)
1232                 mode |= FMODE_NDELAY;
1233         else
1234                 mode &= ~FMODE_NDELAY;
1235
1236         return blkdev_ioctl(bdev, mode, cmd, arg);
1237 }
1238
1239 /*
1240  * Try to release a page associated with block device when the system
1241  * is under memory pressure.
1242  */
1243 static int blkdev_releasepage(struct page *page, gfp_t wait)
1244 {
1245         struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1246
1247         if (super && super->s_op->bdev_try_to_free_page)
1248                 return super->s_op->bdev_try_to_free_page(super, page, wait);
1249
1250         return try_to_free_buffers(page);
1251 }
1252
1253 static const struct address_space_operations def_blk_aops = {
1254         .readpage       = blkdev_readpage,
1255         .writepage      = blkdev_writepage,
1256         .sync_page      = block_sync_page,
1257         .write_begin    = blkdev_write_begin,
1258         .write_end      = blkdev_write_end,
1259         .writepages     = generic_writepages,
1260         .releasepage    = blkdev_releasepage,
1261         .direct_IO      = blkdev_direct_IO,
1262 };
1263
1264 const struct file_operations def_blk_fops = {
1265         .open           = blkdev_open,
1266         .release        = blkdev_close,
1267         .llseek         = block_llseek,
1268         .read           = do_sync_read,
1269         .write          = do_sync_write,
1270         .aio_read       = generic_file_aio_read,
1271         .aio_write      = generic_file_aio_write_nolock,
1272         .mmap           = generic_file_mmap,
1273         .fsync          = block_fsync,
1274         .unlocked_ioctl = block_ioctl,
1275 #ifdef CONFIG_COMPAT
1276         .compat_ioctl   = compat_blkdev_ioctl,
1277 #endif
1278         .splice_read    = generic_file_splice_read,
1279         .splice_write   = generic_file_splice_write,
1280 };
1281
1282 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1283 {
1284         int res;
1285         mm_segment_t old_fs = get_fs();
1286         set_fs(KERNEL_DS);
1287         res = blkdev_ioctl(bdev, 0, cmd, arg);
1288         set_fs(old_fs);
1289         return res;
1290 }
1291
1292 EXPORT_SYMBOL(ioctl_by_bdev);
1293
1294 /**
1295  * lookup_bdev  - lookup a struct block_device by name
1296  * @pathname:   special file representing the block device
1297  *
1298  * Get a reference to the blockdevice at @pathname in the current
1299  * namespace if possible and return it.  Return ERR_PTR(error)
1300  * otherwise.
1301  */
1302 struct block_device *lookup_bdev(const char *pathname)
1303 {
1304         struct block_device *bdev;
1305         struct inode *inode;
1306         struct path path;
1307         int error;
1308
1309         if (!pathname || !*pathname)
1310                 return ERR_PTR(-EINVAL);
1311
1312         error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1313         if (error)
1314                 return ERR_PTR(error);
1315
1316         inode = path.dentry->d_inode;
1317         error = -ENOTBLK;
1318         if (!S_ISBLK(inode->i_mode))
1319                 goto fail;
1320         error = -EACCES;
1321         if (path.mnt->mnt_flags & MNT_NODEV)
1322                 goto fail;
1323         error = -ENOMEM;
1324         bdev = bd_acquire(inode);
1325         if (!bdev)
1326                 goto fail;
1327 out:
1328         path_put(&path);
1329         return bdev;
1330 fail:
1331         bdev = ERR_PTR(error);
1332         goto out;
1333 }
1334 EXPORT_SYMBOL(lookup_bdev);
1335
1336 /**
1337  * open_bdev_exclusive  -  open a block device by name and set it up for use
1338  *
1339  * @path:       special file representing the block device
1340  * @mode:       FMODE_... combination to pass be used
1341  * @holder:     owner for exclusion
1342  *
1343  * Open the blockdevice described by the special file at @path, claim it
1344  * for the @holder.
1345  */
1346 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1347 {
1348         struct block_device *bdev;
1349         int error = 0;
1350
1351         bdev = lookup_bdev(path);
1352         if (IS_ERR(bdev))
1353                 return bdev;
1354
1355         error = blkdev_get(bdev, mode);
1356         if (error)
1357                 return ERR_PTR(error);
1358         error = -EACCES;
1359         if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1360                 goto blkdev_put;
1361         error = bd_claim(bdev, holder);
1362         if (error)
1363                 goto blkdev_put;
1364
1365         return bdev;
1366
1367 blkdev_put:
1368         blkdev_put(bdev, mode);
1369         return ERR_PTR(error);
1370 }
1371
1372 EXPORT_SYMBOL(open_bdev_exclusive);
1373
1374 /**
1375  * close_bdev_exclusive  -  close a blockdevice opened by open_bdev_exclusive()
1376  *
1377  * @bdev:       blockdevice to close
1378  * @mode:       mode, must match that used to open.
1379  *
1380  * This is the counterpart to open_bdev_exclusive().
1381  */
1382 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1383 {
1384         bd_release(bdev);
1385         blkdev_put(bdev, mode);
1386 }
1387
1388 EXPORT_SYMBOL(close_bdev_exclusive);
1389
1390 int __invalidate_device(struct block_device *bdev)
1391 {
1392         struct super_block *sb = get_super(bdev);
1393         int res = 0;
1394
1395         if (sb) {
1396                 /*
1397                  * no need to lock the super, get_super holds the
1398                  * read mutex so the filesystem cannot go away
1399                  * under us (->put_super runs with the write lock
1400                  * hold).
1401                  */
1402                 shrink_dcache_sb(sb);
1403                 res = invalidate_inodes(sb);
1404                 drop_super(sb);
1405         }
1406         invalidate_bdev(bdev);
1407         return res;
1408 }
1409 EXPORT_SYMBOL(__invalidate_device);