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