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/config.h>
   9 #include <linux/init.h>
  10 #include <linux/mm.h>
  11 #include <linux/fcntl.h>
  12 #include <linux/slab.h>
  13 #include <linux/kmod.h>
  14 #include <linux/major.h>
  15 #include <linux/devfs_fs_kernel.h>
  16 #include <linux/smp_lock.h>
  17 #include <linux/highmem.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/module.h>
  20 #include <linux/blkpg.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/mpage.h>
  23 #include <linux/mount.h>
  24 #include <linux/uio.h>
  25 #include <linux/namei.h>
  26 #include <asm/uaccess.h>
  27
  28 struct bdev_inode {
  29         struct block_device bdev;
  30         struct inode vfs_inode;
  31 };
  32
  33 static inline struct bdev_inode *BDEV_I(struct inode *inode)
  34 {
  35         return container_of(inode, struct bdev_inode, vfs_inode);
  36 }
  37
  38 inline struct block_device *I_BDEV(struct inode *inode)
  39 {
  40         return &BDEV_I(inode)->bdev;
  41 }
  42
  43 EXPORT_SYMBOL(I_BDEV);
  44
  45 static sector_t max_block(struct block_device *bdev)
  46 {
  47         sector_t retval = ~((sector_t)0);
  48         loff_t sz = i_size_read(bdev->bd_inode);
  49
  50         if (sz) {
  51                 unsigned int size = block_size(bdev);
  52                 unsigned int sizebits = blksize_bits(size);
  53                 retval = (sz >> sizebits);
  54         }
  55         return retval;
  56 }
  57
  58 /* Kill _all_ buffers, dirty or not.. */
  59 static void kill_bdev(struct block_device *bdev)
  60 {
  61         invalidate_bdev(bdev, 1);
  62         truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  63 }
  64
  65 int set_blocksize(struct block_device *bdev, int size)
  66 {
  67         /* Size must be a power of two, and between 512 and PAGE_SIZE */
  68         if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
  69                 return -EINVAL;
  70
  71         /* Size cannot be smaller than the size supported by the device */
  72         if (size < bdev_hardsect_size(bdev))
  73                 return -EINVAL;
  74
  75         /* Don't change the size if it is same as current */
  76         if (bdev->bd_block_size != size) {
  77                 sync_blockdev(bdev);
  78                 bdev->bd_block_size = size;
  79                 bdev->bd_inode->i_blkbits = blksize_bits(size);
  80                 kill_bdev(bdev);
  81         }
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(set_blocksize);
  86
  87 int sb_set_blocksize(struct super_block *sb, int size)
  88 {
  89         int bits = 9; /* 2^9 = 512 */
  90
  91         if (set_blocksize(sb->s_bdev, size))
  92                 return 0;
  93         /* If we get here, we know size is power of two
  94          * and it's value is between 512 and PAGE_SIZE */
  95         sb->s_blocksize = size;
  96         for (size >>= 10; size; size >>= 1)
  97                 ++bits;
  98         sb->s_blocksize_bits = bits;
  99         return sb->s_blocksize;
 100 }
 101
 102 EXPORT_SYMBOL(sb_set_blocksize);
 103
 104 int sb_min_blocksize(struct super_block *sb, int size)
 105 {
 106         int minsize = bdev_hardsect_size(sb->s_bdev);
 107         if (size < minsize)
 108                 size = minsize;
 109         return sb_set_blocksize(sb, size);
 110 }
 111
 112 EXPORT_SYMBOL(sb_min_blocksize);
 113
 114 static int
 115 blkdev_get_block(struct inode *inode, sector_t iblock,
 116                 struct buffer_head *bh, int create)
 117 {
 118         if (iblock >= max_block(I_BDEV(inode))) {
 119                 if (create)
 120                         return -EIO;
 121
 122                 /*
 123                  * for reads, we're just trying to fill a partial page.
 124                  * return a hole, they will have to call get_block again
 125                  * before they can fill it, and they will get -EIO at that
 126                  * time
 127                  */
 128                 return 0;
 129         }
 130         bh->b_bdev = I_BDEV(inode);
 131         bh->b_blocknr = iblock;
 132         set_buffer_mapped(bh);
 133         return 0;
 134 }
 135
 136 static int
 137 blkdev_get_blocks(struct inode *inode, sector_t iblock,
 138                 unsigned long max_blocks, struct buffer_head *bh, int create)
 139 {
 140         if ((iblock + max_blocks) > max_block(I_BDEV(inode)))
 141                 return -EIO;
 142
 143         bh->b_bdev = I_BDEV(inode);
 144         bh->b_blocknr = iblock;
 145         bh->b_size = max_blocks << inode->i_blkbits;
 146         set_buffer_mapped(bh);
 147         return 0;
 148 }
 149
 150 static ssize_t
 151 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 152                         loff_t offset, unsigned long nr_segs)
 153 {
 154         struct file *file = iocb->ki_filp;
 155         struct inode *inode = file->f_mapping->host;
 156
 157         return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
 158                                 iov, offset, nr_segs, blkdev_get_blocks, NULL);
 159 }
 160
 161 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 162 {
 163         return block_write_full_page(page, blkdev_get_block, wbc);
 164 }
 165
 166 static int blkdev_readpage(struct file * file, struct page * page)
 167 {
 168         return block_read_full_page(page, blkdev_get_block);
 169 }
 170
 171 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 172 {
 173         return block_prepare_write(page, from, to, blkdev_get_block);
 174 }
 175
 176 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
 177 {
 178         return block_commit_write(page, from, to);
 179 }
 180
 181 /*
 182  * private llseek:
 183  * for a block special file file->f_dentry->d_inode->i_size is zero
 184  * so we compute the size by hand (just as in block_read/write above)
 185  */
 186 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 187 {
 188         struct inode *bd_inode = file->f_mapping->host;
 189         loff_t size;
 190         loff_t retval;
 191
 192         down(&bd_inode->i_sem);
 193         size = i_size_read(bd_inode);
 194
 195         switch (origin) {
 196                 case 2:
 197                         offset += size;
 198                         break;
 199                 case 1:
 200                         offset += file->f_pos;
 201         }
 202         retval = -EINVAL;
 203         if (offset >= 0 && offset <= size) {
 204                 if (offset != file->f_pos) {
 205                         file->f_pos = offset;
 206                 }
 207                 retval = offset;
 208         }
 209         up(&bd_inode->i_sem);
 210         return retval;
 211 }
 212
 213 /*
 214  *      Filp is never NULL; the only case when ->fsync() is called with
 215  *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 216  */
 217
 218 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 219 {
 220         return sync_blockdev(I_BDEV(filp->f_mapping->host));
 221 }
 222
 223 /*
 224  * pseudo-fs
 225  */
 226
 227 static spinlock_t bdev_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 228 static kmem_cache_t * bdev_cachep;
 229
 230 static struct inode *bdev_alloc_inode(struct super_block *sb)
 231 {
 232         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
 233         if (!ei)
 234                 return NULL;
 235         return &ei->vfs_inode;
 236 }
 237
 238 static void bdev_destroy_inode(struct inode *inode)
 239 {
 240         struct bdev_inode *bdi = BDEV_I(inode);
 241
 242         bdi->bdev.bd_inode_backing_dev_info = NULL;
 243         kmem_cache_free(bdev_cachep, bdi);
 244 }
 245
 246 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 247 {
 248         struct bdev_inode *ei = (struct bdev_inode *) foo;
 249         struct block_device *bdev = &ei->bdev;
 250
 251         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 252             SLAB_CTOR_CONSTRUCTOR)
 253         {
 254                 memset(bdev, 0, sizeof(*bdev));
 255                 sema_init(&bdev->bd_sem, 1);
 256                 sema_init(&bdev->bd_mount_sem, 1);
 257                 INIT_LIST_HEAD(&bdev->bd_inodes);
 258                 INIT_LIST_HEAD(&bdev->bd_list);
 259                 inode_init_once(&ei->vfs_inode);
 260         }
 261 }
 262
 263 static inline void __bd_forget(struct inode *inode)
 264 {
 265         list_del_init(&inode->i_devices);
 266         inode->i_bdev = NULL;
 267         inode->i_mapping = &inode->i_data;
 268 }
 269
 270 static void bdev_clear_inode(struct inode *inode)
 271 {
 272         struct block_device *bdev = &BDEV_I(inode)->bdev;
 273         struct list_head *p;
 274         spin_lock(&bdev_lock);
 275         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 276                 __bd_forget(list_entry(p, struct inode, i_devices));
 277         }
 278         list_del_init(&bdev->bd_list);
 279         spin_unlock(&bdev_lock);
 280 }
 281
 282 static struct super_operations bdev_sops = {
 283         .statfs = simple_statfs,
 284         .alloc_inode = bdev_alloc_inode,
 285         .destroy_inode = bdev_destroy_inode,
 286         .drop_inode = generic_delete_inode,
 287         .clear_inode = bdev_clear_inode,
 288 };
 289
 290 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
 291         int flags, const char *dev_name, void *data)
 292 {
 293         return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
 294 }
 295
 296 static struct file_system_type bd_type = {
 297         .name           = "bdev",
 298         .get_sb         = bd_get_sb,
 299         .kill_sb        = kill_anon_super,
 300 };
 301
 302 static struct vfsmount *bd_mnt;
 303 struct super_block *blockdev_superblock;
 304
 305 void __init bdev_cache_init(void)
 306 {
 307         int err;
 308         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 309                         0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
 310                         init_once, NULL);
 311         err = register_filesystem(&bd_type);
 312         if (err)
 313                 panic("Cannot register bdev pseudo-fs");
 314         bd_mnt = kern_mount(&bd_type);
 315         err = PTR_ERR(bd_mnt);
 316         if (IS_ERR(bd_mnt))
 317                 panic("Cannot create bdev pseudo-fs");
 318         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 319 }
 320
 321 /*
 322  * Most likely _very_ bad one - but then it's hardly critical for small
 323  * /dev and can be fixed when somebody will need really large one.
 324  * Keep in mind that it will be fed through icache hash function too.
 325  */
 326 static inline unsigned long hash(dev_t dev)
 327 {
 328         return MAJOR(dev)+MINOR(dev);
 329 }
 330
 331 static int bdev_test(struct inode *inode, void *data)
 332 {
 333         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 334 }
 335
 336 static int bdev_set(struct inode *inode, void *data)
 337 {
 338         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 339         return 0;
 340 }
 341
 342 static LIST_HEAD(all_bdevs);
 343
 344 struct block_device *bdget(dev_t dev)
 345 {
 346         struct block_device *bdev;
 347         struct inode *inode;
 348
 349         inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
 350                         bdev_test, bdev_set, &dev);
 351
 352         if (!inode)
 353                 return NULL;
 354
 355         bdev = &BDEV_I(inode)->bdev;
 356
 357         if (inode->i_state & I_NEW) {
 358                 bdev->bd_contains = NULL;
 359                 bdev->bd_inode = inode;
 360                 bdev->bd_block_size = (1 << inode->i_blkbits);
 361                 bdev->bd_part_count = 0;
 362                 bdev->bd_invalidated = 0;
 363                 inode->i_mode = S_IFBLK;
 364                 inode->i_rdev = dev;
 365                 inode->i_bdev = bdev;
 366                 inode->i_data.a_ops = &def_blk_aops;
 367                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 368                 inode->i_data.backing_dev_info = &default_backing_dev_info;
 369                 spin_lock(&bdev_lock);
 370                 list_add(&bdev->bd_list, &all_bdevs);
 371                 spin_unlock(&bdev_lock);
 372                 unlock_new_inode(inode);
 373         }
 374         return bdev;
 375 }
 376
 377 EXPORT_SYMBOL(bdget);
 378
 379 long nr_blockdev_pages(void)
 380 {
 381         struct list_head *p;
 382         long ret = 0;
 383         spin_lock(&bdev_lock);
 384         list_for_each(p, &all_bdevs) {
 385                 struct block_device *bdev;
 386                 bdev = list_entry(p, struct block_device, bd_list);
 387                 ret += bdev->bd_inode->i_mapping->nrpages;
 388         }
 389         spin_unlock(&bdev_lock);
 390         return ret;
 391 }
 392
 393 void bdput(struct block_device *bdev)
 394 {
 395         iput(bdev->bd_inode);
 396 }
 397
 398 EXPORT_SYMBOL(bdput);
 399
 400 static struct block_device *bd_acquire(struct inode *inode)
 401 {
 402         struct block_device *bdev;
 403         spin_lock(&bdev_lock);
 404         bdev = inode->i_bdev;
 405         if (bdev && igrab(bdev->bd_inode)) {
 406                 spin_unlock(&bdev_lock);
 407                 return bdev;
 408         }
 409         spin_unlock(&bdev_lock);
 410         bdev = bdget(inode->i_rdev);
 411         if (bdev) {
 412                 spin_lock(&bdev_lock);
 413                 if (inode->i_bdev)
 414                         __bd_forget(inode);
 415                 inode->i_bdev = bdev;
 416                 inode->i_mapping = bdev->bd_inode->i_mapping;
 417                 list_add(&inode->i_devices, &bdev->bd_inodes);
 418                 spin_unlock(&bdev_lock);
 419         }
 420         return bdev;
 421 }
 422
 423 /* Call when you free inode */
 424
 425 void bd_forget(struct inode *inode)
 426 {
 427         spin_lock(&bdev_lock);
 428         if (inode->i_bdev)
 429                 __bd_forget(inode);
 430         spin_unlock(&bdev_lock);
 431 }
 432
 433 int bd_claim(struct block_device *bdev, void *holder)
 434 {
 435         int res;
 436         spin_lock(&bdev_lock);
 437
 438         /* first decide result */
 439         if (bdev->bd_holder == holder)
 440                 res = 0;         /* already a holder */
 441         else if (bdev->bd_holder != NULL)
 442                 res = -EBUSY;    /* held by someone else */
 443         else if (bdev->bd_contains == bdev)
 444                 res = 0;         /* is a whole device which isn't held */
 445
 446         else if (bdev->bd_contains->bd_holder == bd_claim)
 447                 res = 0;         /* is a partition of a device that is being partitioned */
 448         else if (bdev->bd_contains->bd_holder != NULL)
 449                 res = -EBUSY;    /* is a partition of a held device */
 450         else
 451                 res = 0;         /* is a partition of an un-held device */
 452
 453         /* now impose change */
 454         if (res==0) {
 455                 /* note that for a whole device bd_holders
 456                  * will be incremented twice, and bd_holder will
 457                  * be set to bd_claim before being set to holder
 458                  */
 459                 bdev->bd_contains->bd_holders ++;
 460                 bdev->bd_contains->bd_holder = bd_claim;
 461                 bdev->bd_holders++;
 462                 bdev->bd_holder = holder;
 463         }
 464         spin_unlock(&bdev_lock);
 465         return res;
 466 }
 467
 468 EXPORT_SYMBOL(bd_claim);
 469
 470 void bd_release(struct block_device *bdev)
 471 {
 472         spin_lock(&bdev_lock);
 473         if (!--bdev->bd_contains->bd_holders)
 474                 bdev->bd_contains->bd_holder = NULL;
 475         if (!--bdev->bd_holders)
 476                 bdev->bd_holder = NULL;
 477         spin_unlock(&bdev_lock);
 478 }
 479
 480 EXPORT_SYMBOL(bd_release);
 481
 482 /*
 483  * Tries to open block device by device number.  Use it ONLY if you
 484  * really do not have anything better - i.e. when you are behind a
 485  * truly sucky interface and all you are given is a device number.  _Never_
 486  * to be used for internal purposes.  If you ever need it - reconsider
 487  * your API.
 488  */
 489 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
 490 {
 491         struct block_device *bdev = bdget(dev);
 492         int err = -ENOMEM;
 493         int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
 494         if (bdev)
 495                 err = blkdev_get(bdev, mode, flags);
 496         return err ? ERR_PTR(err) : bdev;
 497 }
 498
 499 EXPORT_SYMBOL(open_by_devnum);
 500
 501 /*
 502  * This routine checks whether a removable media has been changed,
 503  * and invalidates all buffer-cache-entries in that case. This
 504  * is a relatively slow routine, so we have to try to minimize using
 505  * it. Thus it is called only upon a 'mount' or 'open'. This
 506  * is the best way of combining speed and utility, I think.
 507  * People changing diskettes in the middle of an operation deserve
 508  * to lose :-)
 509  */
 510 int check_disk_change(struct block_device *bdev)
 511 {
 512         struct gendisk *disk = bdev->bd_disk;
 513         struct block_device_operations * bdops = disk->fops;
 514
 515         if (!bdops->media_changed)
 516                 return 0;
 517         if (!bdops->media_changed(bdev->bd_disk))
 518                 return 0;
 519
 520         if (__invalidate_device(bdev, 0))
 521                 printk("VFS: busy inodes on changed media.\n");
 522
 523         if (bdops->revalidate_disk)
 524                 bdops->revalidate_disk(bdev->bd_disk);
 525         if (bdev->bd_disk->minors > 1)
 526                 bdev->bd_invalidated = 1;
 527         return 1;
 528 }
 529
 530 EXPORT_SYMBOL(check_disk_change);
 531
 532 void bd_set_size(struct block_device *bdev, loff_t size)
 533 {
 534         unsigned bsize = bdev_hardsect_size(bdev);
 535
 536         bdev->bd_inode->i_size = size;
 537         while (bsize < PAGE_CACHE_SIZE) {
 538                 if (size & bsize)
 539                         break;
 540                 bsize <<= 1;
 541         }
 542         bdev->bd_block_size = bsize;
 543         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 544 }
 545 EXPORT_SYMBOL(bd_set_size);
 546
 547 static int do_open(struct block_device *bdev, struct file *file)
 548 {
 549         struct module *owner = NULL;
 550         struct gendisk *disk;
 551         int ret = -ENXIO;
 552         int part;
 553
 554         file->f_mapping = bdev->bd_inode->i_mapping;
 555         lock_kernel();
 556         disk = get_gendisk(bdev->bd_dev, &part);
 557         if (!disk) {
 558                 unlock_kernel();
 559                 bdput(bdev);
 560                 return ret;
 561         }
 562         owner = disk->fops->owner;
 563
 564         down(&bdev->bd_sem);
 565         if (!bdev->bd_openers) {
 566                 bdev->bd_disk = disk;
 567                 bdev->bd_contains = bdev;
 568                 if (!part) {
 569                         struct backing_dev_info *bdi;
 570                         if (disk->fops->open) {
 571                                 ret = disk->fops->open(bdev->bd_inode, file);
 572                                 if (ret)
 573                                         goto out_first;
 574                         }
 575                         if (!bdev->bd_openers) {
 576                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
 577                                 bdi = blk_get_backing_dev_info(bdev);
 578                                 if (bdi == NULL)
 579                                         bdi = &default_backing_dev_info;
 580                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
 581                         }
 582                         if (bdev->bd_invalidated)
 583                                 rescan_partitions(disk, bdev);
 584                 } else {
 585                         struct hd_struct *p;
 586                         struct block_device *whole;
 587                         whole = bdget_disk(disk, 0);
 588                         ret = -ENOMEM;
 589                         if (!whole)
 590                                 goto out_first;
 591                         ret = blkdev_get(whole, file->f_mode, file->f_flags);
 592                         if (ret)
 593                                 goto out_first;
 594                         bdev->bd_contains = whole;
 595                         down(&whole->bd_sem);
 596                         whole->bd_part_count++;
 597                         p = disk->part[part - 1];
 598                         bdev->bd_inode->i_data.backing_dev_info =
 599                            whole->bd_inode->i_data.backing_dev_info;
 600                         if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
 601                                 whole->bd_part_count--;
 602                                 up(&whole->bd_sem);
 603                                 ret = -ENXIO;
 604                                 goto out_first;
 605                         }
 606                         kobject_get(&p->kobj);
 607                         bdev->bd_part = p;
 608                         bd_set_size(bdev, (loff_t) p->nr_sects << 9);
 609                         up(&whole->bd_sem);
 610                 }
 611         } else {
 612                 put_disk(disk);
 613                 module_put(owner);
 614                 if (bdev->bd_contains == bdev) {
 615                         if (bdev->bd_disk->fops->open) {
 616                                 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
 617                                 if (ret)
 618                                         goto out;
 619                         }
 620                         if (bdev->bd_invalidated)
 621                                 rescan_partitions(bdev->bd_disk, bdev);
 622                 } else {
 623                         down(&bdev->bd_contains->bd_sem);
 624                         bdev->bd_contains->bd_part_count++;
 625                         up(&bdev->bd_contains->bd_sem);
 626                 }
 627         }
 628         bdev->bd_openers++;
 629         up(&bdev->bd_sem);
 630         unlock_kernel();
 631         return 0;
 632
 633 out_first:
 634         bdev->bd_disk = NULL;
 635         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 636         if (bdev != bdev->bd_contains)
 637                 blkdev_put(bdev->bd_contains);
 638         bdev->bd_contains = NULL;
 639         put_disk(disk);
 640         module_put(owner);
 641 out:
 642         up(&bdev->bd_sem);
 643         unlock_kernel();
 644         if (ret)
 645                 bdput(bdev);
 646         return ret;
 647 }
 648
 649 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
 650 {
 651         /*
 652          * This crockload is due to bad choice of ->open() type.
 653          * It will go away.
 654          * For now, block device ->open() routine must _not_
 655          * examine anything in 'inode' argument except ->i_rdev.
 656          */
 657         struct file fake_file = {};
 658         struct dentry fake_dentry = {};
 659         fake_file.f_mode = mode;
 660         fake_file.f_flags = flags;
 661         fake_file.f_dentry = &fake_dentry;
 662         fake_dentry.d_inode = bdev->bd_inode;
 663
 664         return do_open(bdev, &fake_file);
 665 }
 666
 667 EXPORT_SYMBOL(blkdev_get);
 668
 669 int blkdev_open(struct inode * inode, struct file * filp)
 670 {
 671         struct block_device *bdev;
 672         int res;
 673
 674         /*
 675          * Preserve backwards compatibility and allow large file access
 676          * even if userspace doesn't ask for it explicitly. Some mkfs
 677          * binary needs it. We might want to drop this workaround
 678          * during an unstable branch.
 679          */
 680         filp->f_flags |= O_LARGEFILE;
 681
 682         bdev = bd_acquire(inode);
 683
 684         res = do_open(bdev, filp);
 685         if (res)
 686                 return res;
 687
 688         if (!(filp->f_flags & O_EXCL) )
 689                 return 0;
 690
 691         if (!(res = bd_claim(bdev, filp)))
 692                 return 0;
 693
 694         blkdev_put(bdev);
 695         return res;
 696 }
 697
 698 EXPORT_SYMBOL(blkdev_open);
 699
 700 int blkdev_put(struct block_device *bdev)
 701 {
 702         int ret = 0;
 703         struct inode *bd_inode = bdev->bd_inode;
 704         struct gendisk *disk = bdev->bd_disk;
 705
 706         down(&bdev->bd_sem);
 707         lock_kernel();
 708         if (!--bdev->bd_openers) {
 709                 sync_blockdev(bdev);
 710                 kill_bdev(bdev);
 711         }
 712         if (bdev->bd_contains == bdev) {
 713                 if (disk->fops->release)
 714                         ret = disk->fops->release(bd_inode, NULL);
 715         } else {
 716                 down(&bdev->bd_contains->bd_sem);
 717                 bdev->bd_contains->bd_part_count--;
 718                 up(&bdev->bd_contains->bd_sem);
 719         }
 720         if (!bdev->bd_openers) {
 721                 struct module *owner = disk->fops->owner;
 722
 723                 put_disk(disk);
 724                 module_put(owner);
 725
 726                 if (bdev->bd_contains != bdev) {
 727                         kobject_put(&bdev->bd_part->kobj);
 728                         bdev->bd_part = NULL;
 729                 }
 730                 bdev->bd_disk = NULL;
 731                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 732                 if (bdev != bdev->bd_contains) {
 733                         blkdev_put(bdev->bd_contains);
 734                 }
 735                 bdev->bd_contains = NULL;
 736         }
 737         unlock_kernel();
 738         up(&bdev->bd_sem);
 739         bdput(bdev);
 740         return ret;
 741 }
 742
 743 EXPORT_SYMBOL(blkdev_put);
 744
 745 static int blkdev_close(struct inode * inode, struct file * filp)
 746 {
 747         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
 748         if (bdev->bd_holder == filp)
 749                 bd_release(bdev);
 750         return blkdev_put(bdev);
 751 }
 752
 753 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
 754                                    size_t count, loff_t *ppos)
 755 {
 756         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 757
 758         return generic_file_write_nolock(file, &local_iov, 1, ppos);
 759 }
 760
 761 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
 762                                    size_t count, loff_t pos)
 763 {
 764         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 765
 766         return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
 767 }
 768
 769 static int block_ioctl(struct inode *inode, struct file *file, unsigned cmd,
 770                         unsigned long arg)
 771 {
 772         return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
 773 }
 774
 775 struct address_space_operations def_blk_aops = {
 776         .readpage       = blkdev_readpage,
 777         .writepage      = blkdev_writepage,
 778         .sync_page      = block_sync_page,
 779         .prepare_write  = blkdev_prepare_write,
 780         .commit_write   = blkdev_commit_write,
 781         .writepages     = generic_writepages,
 782         .direct_IO      = blkdev_direct_IO,
 783 };
 784
 785 struct file_operations def_blk_fops = {
 786         .open           = blkdev_open,
 787         .release        = blkdev_close,
 788         .llseek         = block_llseek,
 789         .read           = generic_file_read,
 790         .write          = blkdev_file_write,
 791         .aio_read       = generic_file_aio_read,
 792         .aio_write      = blkdev_file_aio_write,
 793         .mmap           = generic_file_mmap,
 794         .fsync          = block_fsync,
 795         .ioctl          = block_ioctl,
 796         .readv          = generic_file_readv,
 797         .writev         = generic_file_write_nolock,
 798         .sendfile       = generic_file_sendfile,
 799 };
 800
 801 EXPORT_SYMBOL(def_blk_fops);
 802
 803 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
 804 {
 805         int res;
 806         mm_segment_t old_fs = get_fs();
 807         set_fs(KERNEL_DS);
 808         res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
 809         set_fs(old_fs);
 810         return res;
 811 }
 812
 813 EXPORT_SYMBOL(ioctl_by_bdev);
 814
 815 /**
 816  * lookup_bdev  - lookup a struct block_device by name
 817  *
 818  * @path:       special file representing the block device
 819  *
 820  * Get a reference to the blockdevice at @path in the current
 821  * namespace if possible and return it.  Return ERR_PTR(error)
 822  * otherwise.
 823  */
 824 struct block_device *lookup_bdev(const char *path)
 825 {
 826         struct block_device *bdev;
 827         struct inode *inode;
 828         struct nameidata nd;
 829         int error;
 830
 831         if (!path || !*path)
 832                 return ERR_PTR(-EINVAL);
 833
 834         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
 835         if (error)
 836                 return ERR_PTR(error);
 837
 838         inode = nd.dentry->d_inode;
 839         error = -ENOTBLK;
 840         if (!S_ISBLK(inode->i_mode))
 841                 goto fail;
 842         error = -EACCES;
 843         if (nd.mnt->mnt_flags & MNT_NODEV)
 844                 goto fail;
 845         error = -ENOMEM;
 846         bdev = bd_acquire(inode);
 847         if (!bdev)
 848                 goto fail;
 849 out:
 850         path_release(&nd);
 851         return bdev;
 852 fail:
 853         bdev = ERR_PTR(error);
 854         goto out;
 855 }
 856
 857 /**
 858  * open_bdev_excl  -  open a block device by name and set it up for use
 859  *
 860  * @path:       special file representing the block device
 861  * @flags:      %MS_RDONLY for opening read-only
 862  * @holder:     owner for exclusion
 863  *
 864  * Open the blockdevice described by the special file at @path, claim it
 865  * for the @holder.
 866  */
 867 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
 868 {
 869         struct block_device *bdev;
 870         mode_t mode = FMODE_READ;
 871         int error = 0;
 872
 873         bdev = lookup_bdev(path);
 874         if (IS_ERR(bdev))
 875                 return bdev;
 876
 877         if (!(flags & MS_RDONLY))
 878                 mode |= FMODE_WRITE;
 879         error = blkdev_get(bdev, mode, 0);
 880         if (error)
 881                 return ERR_PTR(error);
 882         error = -EACCES;
 883         if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
 884                 goto blkdev_put;
 885         error = bd_claim(bdev, holder);
 886         if (error)
 887                 goto blkdev_put;
 888
 889         return bdev;
 890
 891 blkdev_put:
 892         blkdev_put(bdev);
 893         return ERR_PTR(error);
 894 }
 895
 896 EXPORT_SYMBOL(open_bdev_excl);
 897
 898 /**
 899  * close_bdev_excl  -  release a blockdevice openen by open_bdev_excl()
 900  *
 901  * @bdev:       blockdevice to close
 902  *
 903  * This is the counterpart to open_bdev_excl().
 904  */
 905 void close_bdev_excl(struct block_device *bdev)
 906 {
 907         bd_release(bdev);
 908         blkdev_put(bdev);
 909 }
 910
 911 EXPORT_SYMBOL(close_bdev_excl);