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         int oldsize;
  68
  69         /* Size must be a power of two, and between 512 and PAGE_SIZE */
  70         if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
  71                 return -EINVAL;
  72
  73         /* Size cannot be smaller than the size supported by the device */
  74         if (size < bdev_hardsect_size(bdev))
  75                 return -EINVAL;
  76
  77         oldsize = bdev->bd_block_size;
  78         if (oldsize == size)
  79                 return 0;
  80
  81         /* Ok, we're actually changing the blocksize.. */
  82         sync_blockdev(bdev);
  83         bdev->bd_block_size = size;
  84         bdev->bd_inode->i_blkbits = blksize_bits(size);
  85         kill_bdev(bdev);
  86         return 0;
  87 }
  88
  89 EXPORT_SYMBOL(set_blocksize);
  90
  91 int sb_set_blocksize(struct super_block *sb, int size)
  92 {
  93         int bits;
  94         if (set_blocksize(sb->s_bdev, size) < 0)
  95                 return 0;
  96         sb->s_blocksize = size;
  97         for (bits = 9, size >>= 9; size >>= 1; bits++)
  98                 ;
  99         sb->s_blocksize_bits = bits;
 100         return sb->s_blocksize;
 101 }
 102
 103 EXPORT_SYMBOL(sb_set_blocksize);
 104
 105 int sb_min_blocksize(struct super_block *sb, int size)
 106 {
 107         int minsize = bdev_hardsect_size(sb->s_bdev);
 108         if (size < minsize)
 109                 size = minsize;
 110         return sb_set_blocksize(sb, size);
 111 }
 112
 113 EXPORT_SYMBOL(sb_min_blocksize);
 114
 115 static int
 116 blkdev_get_block(struct inode *inode, sector_t iblock,
 117                 struct buffer_head *bh, int create)
 118 {
 119         if (iblock >= max_block(I_BDEV(inode))) {
 120                 if (create)
 121                         return -EIO;
 122
 123                 /*
 124                  * for reads, we're just trying to fill a partial page.
 125                  * return a hole, they will have to call get_block again
 126                  * before they can fill it, and they will get -EIO at that
 127                  * time
 128                  */
 129                 return 0;
 130         }
 131         bh->b_bdev = I_BDEV(inode);
 132         bh->b_blocknr = iblock;
 133         set_buffer_mapped(bh);
 134         return 0;
 135 }
 136
 137 static int
 138 blkdev_get_blocks(struct inode *inode, sector_t iblock,
 139                 unsigned long max_blocks, struct buffer_head *bh, int create)
 140 {
 141         if ((iblock + max_blocks) > max_block(I_BDEV(inode)))
 142                 return -EIO;
 143
 144         bh->b_bdev = I_BDEV(inode);
 145         bh->b_blocknr = iblock;
 146         bh->b_size = max_blocks << inode->i_blkbits;
 147         set_buffer_mapped(bh);
 148         return 0;
 149 }
 150
 151 static int
 152 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 153                         loff_t offset, unsigned long nr_segs)
 154 {
 155         struct file *file = iocb->ki_filp;
 156         struct inode *inode = file->f_mapping->host;
 157
 158         return blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 159                                 nr_segs, blkdev_get_blocks, NULL);
 160 }
 161
 162 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 163 {
 164         return block_write_full_page(page, blkdev_get_block, wbc);
 165 }
 166
 167 static int blkdev_readpage(struct file * file, struct page * page)
 168 {
 169         return block_read_full_page(page, blkdev_get_block);
 170 }
 171
 172 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 173 {
 174         return block_prepare_write(page, from, to, blkdev_get_block);
 175 }
 176
 177 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
 178 {
 179         return block_commit_write(page, from, to);
 180 }
 181
 182 /*
 183  * private llseek:
 184  * for a block special file file->f_dentry->d_inode->i_size is zero
 185  * so we compute the size by hand (just as in block_read/write above)
 186  */
 187 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 188 {
 189         struct inode *bd_inode = file->f_mapping->host;
 190         loff_t size;
 191         loff_t retval;
 192
 193         down(&bd_inode->i_sem);
 194         size = i_size_read(bd_inode);
 195
 196         switch (origin) {
 197                 case 2:
 198                         offset += size;
 199                         break;
 200                 case 1:
 201                         offset += file->f_pos;
 202         }
 203         retval = -EINVAL;
 204         if (offset >= 0 && offset <= size) {
 205                 if (offset != file->f_pos) {
 206                         file->f_pos = offset;
 207                 }
 208                 retval = offset;
 209         }
 210         up(&bd_inode->i_sem);
 211         return retval;
 212 }
 213
 214 /*
 215  *      Filp is never NULL; the only case when ->fsync() is called with
 216  *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 217  */
 218
 219 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 220 {
 221         return sync_blockdev(I_BDEV(filp->f_mapping->host));
 222 }
 223
 224 /*
 225  * pseudo-fs
 226  */
 227
 228 static spinlock_t bdev_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 229 static kmem_cache_t * bdev_cachep;
 230
 231 static struct inode *bdev_alloc_inode(struct super_block *sb)
 232 {
 233         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
 234         if (!ei)
 235                 return NULL;
 236         return &ei->vfs_inode;
 237 }
 238
 239 static void bdev_destroy_inode(struct inode *inode)
 240 {
 241         kmem_cache_free(bdev_cachep, BDEV_I(inode));
 242 }
 243
 244 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 245 {
 246         struct bdev_inode *ei = (struct bdev_inode *) foo;
 247         struct block_device *bdev = &ei->bdev;
 248
 249         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 250             SLAB_CTOR_CONSTRUCTOR)
 251         {
 252                 memset(bdev, 0, sizeof(*bdev));
 253                 sema_init(&bdev->bd_sem, 1);
 254                 INIT_LIST_HEAD(&bdev->bd_inodes);
 255                 INIT_LIST_HEAD(&bdev->bd_list);
 256                 inode_init_once(&ei->vfs_inode);
 257         }
 258 }
 259
 260 static inline void __bd_forget(struct inode *inode)
 261 {
 262         list_del_init(&inode->i_devices);
 263         inode->i_bdev = NULL;
 264         inode->i_mapping = &inode->i_data;
 265 }
 266
 267 static void bdev_clear_inode(struct inode *inode)
 268 {
 269         struct block_device *bdev = &BDEV_I(inode)->bdev;
 270         struct list_head *p;
 271         spin_lock(&bdev_lock);
 272         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 273                 __bd_forget(list_entry(p, struct inode, i_devices));
 274         }
 275         list_del_init(&bdev->bd_list);
 276         spin_unlock(&bdev_lock);
 277 }
 278
 279 static struct super_operations bdev_sops = {
 280         .statfs = simple_statfs,
 281         .alloc_inode = bdev_alloc_inode,
 282         .destroy_inode = bdev_destroy_inode,
 283         .drop_inode = generic_delete_inode,
 284         .clear_inode = bdev_clear_inode,
 285 };
 286
 287 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
 288         int flags, const char *dev_name, void *data)
 289 {
 290         return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
 291 }
 292
 293 static struct file_system_type bd_type = {
 294         .name           = "bdev",
 295         .get_sb         = bd_get_sb,
 296         .kill_sb        = kill_anon_super,
 297 };
 298
 299 static struct vfsmount *bd_mnt;
 300 struct super_block *blockdev_superblock;
 301
 302 void __init bdev_cache_init(void)
 303 {
 304         int err;
 305         bdev_cachep = kmem_cache_create("bdev_cache",
 306                                         sizeof(struct bdev_inode),
 307                                         0,
 308                                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
 309                                         init_once,
 310                                         NULL);
 311         if (!bdev_cachep)
 312                 panic("Cannot create bdev_cache SLAB cache");
 313         err = register_filesystem(&bd_type);
 314         if (err)
 315                 panic("Cannot register bdev pseudo-fs");
 316         bd_mnt = kern_mount(&bd_type);
 317         err = PTR_ERR(bd_mnt);
 318         if (IS_ERR(bd_mnt))
 319                 panic("Cannot create bdev pseudo-fs");
 320         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 321 }
 322
 323 /*
 324  * Most likely _very_ bad one - but then it's hardly critical for small
 325  * /dev and can be fixed when somebody will need really large one.
 326  * Keep in mind that it will be fed through icache hash function too.
 327  */
 328 static inline unsigned long hash(dev_t dev)
 329 {
 330         return MAJOR(dev)+MINOR(dev);
 331 }
 332
 333 static int bdev_test(struct inode *inode, void *data)
 334 {
 335         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 336 }
 337
 338 static int bdev_set(struct inode *inode, void *data)
 339 {
 340         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 341         return 0;
 342 }
 343
 344 static LIST_HEAD(all_bdevs);
 345
 346 struct block_device *bdget(dev_t dev)
 347 {
 348         struct block_device *bdev;
 349         struct inode *inode;
 350
 351         inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
 352                         bdev_test, bdev_set, &dev);
 353
 354         if (!inode)
 355                 return NULL;
 356
 357         bdev = &BDEV_I(inode)->bdev;
 358
 359         if (inode->i_state & I_NEW) {
 360                 bdev->bd_contains = NULL;
 361                 bdev->bd_inode = inode;
 362                 bdev->bd_block_size = (1 << inode->i_blkbits);
 363                 bdev->bd_part_count = 0;
 364                 bdev->bd_invalidated = 0;
 365                 inode->i_mode = S_IFBLK;
 366                 inode->i_rdev = dev;
 367                 inode->i_bdev = bdev;
 368                 inode->i_data.a_ops = &def_blk_aops;
 369                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 370                 inode->i_data.backing_dev_info = &default_backing_dev_info;
 371                 spin_lock(&bdev_lock);
 372                 list_add(&bdev->bd_list, &all_bdevs);
 373                 spin_unlock(&bdev_lock);
 374                 unlock_new_inode(inode);
 375         }
 376         return bdev;
 377 }
 378
 379 EXPORT_SYMBOL(bdget);
 380
 381 long nr_blockdev_pages(void)
 382 {
 383         struct list_head *p;
 384         long ret = 0;
 385         spin_lock(&bdev_lock);
 386         list_for_each(p, &all_bdevs) {
 387                 struct block_device *bdev;
 388                 bdev = list_entry(p, struct block_device, bd_list);
 389                 ret += bdev->bd_inode->i_mapping->nrpages;
 390         }
 391         spin_unlock(&bdev_lock);
 392         return ret;
 393 }
 394
 395 void bdput(struct block_device *bdev)
 396 {
 397         iput(bdev->bd_inode);
 398 }
 399
 400 EXPORT_SYMBOL(bdput);
 401
 402 static struct block_device *bd_acquire(struct inode *inode)
 403 {
 404         struct block_device *bdev;
 405         spin_lock(&bdev_lock);
 406         bdev = inode->i_bdev;
 407         if (bdev && igrab(bdev->bd_inode)) {
 408                 spin_unlock(&bdev_lock);
 409                 return bdev;
 410         }
 411         spin_unlock(&bdev_lock);
 412         bdev = bdget(inode->i_rdev);
 413         if (bdev) {
 414                 spin_lock(&bdev_lock);
 415                 if (inode->i_bdev)
 416                         __bd_forget(inode);
 417                 inode->i_bdev = bdev;
 418                 inode->i_mapping = bdev->bd_inode->i_mapping;
 419                 list_add(&inode->i_devices, &bdev->bd_inodes);
 420                 spin_unlock(&bdev_lock);
 421         }
 422         return bdev;
 423 }
 424
 425 /* Call when you free inode */
 426
 427 void bd_forget(struct inode *inode)
 428 {
 429         spin_lock(&bdev_lock);
 430         if (inode->i_bdev)
 431                 __bd_forget(inode);
 432         spin_unlock(&bdev_lock);
 433 }
 434
 435 int bd_claim(struct block_device *bdev, void *holder)
 436 {
 437         int res;
 438         spin_lock(&bdev_lock);
 439
 440         /* first decide result */
 441         if (bdev->bd_holder == holder)
 442                 res = 0;         /* already a holder */
 443         else if (bdev->bd_holder != NULL)
 444                 res = -EBUSY;    /* held by someone else */
 445         else if (bdev->bd_contains == bdev)
 446                 res = 0;         /* is a whole device which isn't held */
 447
 448         else if (bdev->bd_contains->bd_holder == bd_claim)
 449                 res = 0;         /* is a partition of a device that is being partitioned */
 450         else if (bdev->bd_contains->bd_holder != NULL)
 451                 res = -EBUSY;    /* is a partition of a held device */
 452         else
 453                 res = 0;         /* is a partition of an un-held device */
 454
 455         /* now impose change */
 456         if (res==0) {
 457                 /* note that for a whole device bd_holders
 458                  * will be incremented twice, and bd_holder will
 459                  * be set to bd_claim before being set to holder
 460                  */
 461                 bdev->bd_contains->bd_holders ++;
 462                 bdev->bd_contains->bd_holder = bd_claim;
 463                 bdev->bd_holders++;
 464                 bdev->bd_holder = holder;
 465         }
 466         spin_unlock(&bdev_lock);
 467         return res;
 468 }
 469
 470 EXPORT_SYMBOL(bd_claim);
 471
 472 void bd_release(struct block_device *bdev)
 473 {
 474         spin_lock(&bdev_lock);
 475         if (!--bdev->bd_contains->bd_holders)
 476                 bdev->bd_contains->bd_holder = NULL;
 477         if (!--bdev->bd_holders)
 478                 bdev->bd_holder = NULL;
 479         spin_unlock(&bdev_lock);
 480 }
 481
 482 EXPORT_SYMBOL(bd_release);
 483
 484 /*
 485  * Tries to open block device by device number.  Use it ONLY if you
 486  * really do not have anything better - i.e. when you are behind a
 487  * truly sucky interface and all you are given is a device number.  _Never_
 488  * to be used for internal purposes.  If you ever need it - reconsider
 489  * your API.
 490  */
 491 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
 492 {
 493         struct block_device *bdev = bdget(dev);
 494         int err = -ENOMEM;
 495         int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
 496         if (bdev)
 497                 err = blkdev_get(bdev, mode, flags);
 498         return err ? ERR_PTR(err) : bdev;
 499 }
 500
 501 EXPORT_SYMBOL(open_by_devnum);
 502
 503 /*
 504  * This routine checks whether a removable media has been changed,
 505  * and invalidates all buffer-cache-entries in that case. This
 506  * is a relatively slow routine, so we have to try to minimize using
 507  * it. Thus it is called only upon a 'mount' or 'open'. This
 508  * is the best way of combining speed and utility, I think.
 509  * People changing diskettes in the middle of an operation deserve
 510  * to lose :-)
 511  */
 512 int check_disk_change(struct block_device *bdev)
 513 {
 514         struct gendisk *disk = bdev->bd_disk;
 515         struct block_device_operations * bdops = disk->fops;
 516
 517         if (!bdops->media_changed)
 518                 return 0;
 519         if (!bdops->media_changed(bdev->bd_disk))
 520                 return 0;
 521
 522         if (__invalidate_device(bdev, 0))
 523                 printk("VFS: busy inodes on changed media.\n");
 524
 525         if (bdops->revalidate_disk)
 526                 bdops->revalidate_disk(bdev->bd_disk);
 527         if (bdev->bd_disk->minors > 1)
 528                 bdev->bd_invalidated = 1;
 529         return 1;
 530 }
 531
 532 EXPORT_SYMBOL(check_disk_change);
 533
 534 void bd_set_size(struct block_device *bdev, loff_t size)
 535 {
 536         unsigned bsize = bdev_hardsect_size(bdev);
 537
 538         bdev->bd_inode->i_size = size;
 539         while (bsize < PAGE_CACHE_SIZE) {
 540                 if (size & bsize)
 541                         break;
 542                 bsize <<= 1;
 543         }
 544         bdev->bd_block_size = bsize;
 545         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 546 }
 547 EXPORT_SYMBOL(bd_set_size);
 548
 549 static int do_open(struct block_device *bdev, struct file *file)
 550 {
 551         struct module *owner = NULL;
 552         struct gendisk *disk;
 553         int ret = -ENXIO;
 554         int part;
 555
 556         file->f_mapping = bdev->bd_inode->i_mapping;
 557         lock_kernel();
 558         disk = get_gendisk(bdev->bd_dev, &part);
 559         if (!disk) {
 560                 unlock_kernel();
 561                 bdput(bdev);
 562                 return ret;
 563         }
 564         owner = disk->fops->owner;
 565
 566         down(&bdev->bd_sem);
 567         if (!bdev->bd_openers) {
 568                 bdev->bd_disk = disk;
 569                 bdev->bd_contains = bdev;
 570                 if (!part) {
 571                         struct backing_dev_info *bdi;
 572                         if (disk->fops->open) {
 573                                 ret = disk->fops->open(bdev->bd_inode, file);
 574                                 if (ret)
 575                                         goto out_first;
 576                         }
 577                         if (!bdev->bd_openers) {
 578                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
 579                                 bdi = blk_get_backing_dev_info(bdev);
 580                                 if (bdi == NULL)
 581                                         bdi = &default_backing_dev_info;
 582                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
 583                         }
 584                         if (bdev->bd_invalidated)
 585                                 rescan_partitions(disk, bdev);
 586                 } else {
 587                         struct hd_struct *p;
 588                         struct block_device *whole;
 589                         whole = bdget_disk(disk, 0);
 590                         ret = -ENOMEM;
 591                         if (!whole)
 592                                 goto out_first;
 593                         ret = blkdev_get(whole, file->f_mode, file->f_flags);
 594                         if (ret)
 595                                 goto out_first;
 596                         bdev->bd_contains = whole;
 597                         down(&whole->bd_sem);
 598                         whole->bd_part_count++;
 599                         p = disk->part[part - 1];
 600                         bdev->bd_inode->i_data.backing_dev_info =
 601                            whole->bd_inode->i_data.backing_dev_info;
 602                         if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
 603                                 whole->bd_part_count--;
 604                                 up(&whole->bd_sem);
 605                                 ret = -ENXIO;
 606                                 goto out_first;
 607                         }
 608                         kobject_get(&p->kobj);
 609                         bdev->bd_part = p;
 610                         bd_set_size(bdev, (loff_t) p->nr_sects << 9);
 611                         up(&whole->bd_sem);
 612                 }
 613         } else {
 614                 put_disk(disk);
 615                 module_put(owner);
 616                 if (bdev->bd_contains == bdev) {
 617                         if (bdev->bd_disk->fops->open) {
 618                                 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
 619                                 if (ret)
 620                                         goto out;
 621                         }
 622                         if (bdev->bd_invalidated)
 623                                 rescan_partitions(bdev->bd_disk, bdev);
 624                 } else {
 625                         down(&bdev->bd_contains->bd_sem);
 626                         bdev->bd_contains->bd_part_count++;
 627                         up(&bdev->bd_contains->bd_sem);
 628                 }
 629         }
 630         bdev->bd_openers++;
 631         up(&bdev->bd_sem);
 632         unlock_kernel();
 633         return 0;
 634
 635 out_first:
 636         bdev->bd_disk = NULL;
 637         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 638         if (bdev != bdev->bd_contains)
 639                 blkdev_put(bdev->bd_contains);
 640         bdev->bd_contains = NULL;
 641         put_disk(disk);
 642         module_put(owner);
 643 out:
 644         up(&bdev->bd_sem);
 645         unlock_kernel();
 646         if (ret)
 647                 bdput(bdev);
 648         return ret;
 649 }
 650
 651 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
 652 {
 653         /*
 654          * This crockload is due to bad choice of ->open() type.
 655          * It will go away.
 656          * For now, block device ->open() routine must _not_
 657          * examine anything in 'inode' argument except ->i_rdev.
 658          */
 659         struct file fake_file = {};
 660         struct dentry fake_dentry = {};
 661         fake_file.f_mode = mode;
 662         fake_file.f_flags = flags;
 663         fake_file.f_dentry = &fake_dentry;
 664         fake_dentry.d_inode = bdev->bd_inode;
 665
 666         return do_open(bdev, &fake_file);
 667 }
 668
 669 EXPORT_SYMBOL(blkdev_get);
 670
 671 int blkdev_open(struct inode * inode, struct file * filp)
 672 {
 673         struct block_device *bdev;
 674         int res;
 675
 676         /*
 677          * Preserve backwards compatibility and allow large file access
 678          * even if userspace doesn't ask for it explicitly. Some mkfs
 679          * binary needs it. We might want to drop this workaround
 680          * during an unstable branch.
 681          */
 682         filp->f_flags |= O_LARGEFILE;
 683
 684         bdev = bd_acquire(inode);
 685
 686         res = do_open(bdev, filp);
 687         if (res)
 688                 return res;
 689
 690         if (!(filp->f_flags & O_EXCL) )
 691                 return 0;
 692
 693         if (!(res = bd_claim(bdev, filp)))
 694                 return 0;
 695
 696         blkdev_put(bdev);
 697         return res;
 698 }
 699
 700 EXPORT_SYMBOL(blkdev_open);
 701
 702 int blkdev_put(struct block_device *bdev)
 703 {
 704         int ret = 0;
 705         struct inode *bd_inode = bdev->bd_inode;
 706         struct gendisk *disk = bdev->bd_disk;
 707
 708         down(&bdev->bd_sem);
 709         lock_kernel();
 710         if (!--bdev->bd_openers) {
 711                 sync_blockdev(bdev);
 712                 kill_bdev(bdev);
 713         }
 714         if (bdev->bd_contains == bdev) {
 715                 if (disk->fops->release)
 716                         ret = disk->fops->release(bd_inode, NULL);
 717         } else {
 718                 down(&bdev->bd_contains->bd_sem);
 719                 bdev->bd_contains->bd_part_count--;
 720                 up(&bdev->bd_contains->bd_sem);
 721         }
 722         if (!bdev->bd_openers) {
 723                 struct module *owner = disk->fops->owner;
 724
 725                 put_disk(disk);
 726                 module_put(owner);
 727
 728                 if (bdev->bd_contains != bdev) {
 729                         kobject_put(&bdev->bd_part->kobj);
 730                         bdev->bd_part = NULL;
 731                 }
 732                 bdev->bd_disk = NULL;
 733                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 734                 if (bdev != bdev->bd_contains) {
 735                         blkdev_put(bdev->bd_contains);
 736                 }
 737                 bdev->bd_contains = NULL;
 738         }
 739         unlock_kernel();
 740         up(&bdev->bd_sem);
 741         bdput(bdev);
 742         return ret;
 743 }
 744
 745 EXPORT_SYMBOL(blkdev_put);
 746
 747 static int blkdev_close(struct inode * inode, struct file * filp)
 748 {
 749         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
 750         if (bdev->bd_holder == filp)
 751                 bd_release(bdev);
 752         return blkdev_put(bdev);
 753 }
 754
 755 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
 756                                    size_t count, loff_t *ppos)
 757 {
 758         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 759
 760         return generic_file_write_nolock(file, &local_iov, 1, ppos);
 761 }
 762
 763 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
 764                                    size_t count, loff_t pos)
 765 {
 766         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 767
 768         return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
 769 }
 770
 771 static int block_ioctl(struct inode *inode, struct file *file, unsigned cmd,
 772                         unsigned long arg)
 773 {
 774         return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
 775 }
 776
 777 struct address_space_operations def_blk_aops = {
 778         .readpage       = blkdev_readpage,
 779         .writepage      = blkdev_writepage,
 780         .sync_page      = block_sync_page,
 781         .prepare_write  = blkdev_prepare_write,
 782         .commit_write   = blkdev_commit_write,
 783         .writepages     = generic_writepages,
 784         .direct_IO      = blkdev_direct_IO,
 785 };
 786
 787 struct file_operations def_blk_fops = {
 788         .open           = blkdev_open,
 789         .release        = blkdev_close,
 790         .llseek         = block_llseek,
 791         .read           = generic_file_read,
 792         .write          = blkdev_file_write,
 793         .aio_read       = generic_file_aio_read,
 794         .aio_write      = blkdev_file_aio_write,
 795         .mmap           = generic_file_mmap,
 796         .fsync          = block_fsync,
 797         .ioctl          = block_ioctl,
 798         .readv          = generic_file_readv,
 799         .writev         = generic_file_writev,
 800         .sendfile       = generic_file_sendfile,
 801 };
 802
 803 EXPORT_SYMBOL(def_blk_fops);
 804
 805 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
 806 {
 807         int res;
 808         mm_segment_t old_fs = get_fs();
 809         set_fs(KERNEL_DS);
 810         res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
 811         set_fs(old_fs);
 812         return res;
 813 }
 814
 815 EXPORT_SYMBOL(ioctl_by_bdev);
 816
 817 /**
 818  * lookup_bdev  - lookup a struct block_device by name
 819  *
 820  * @path:       special file representing the block device
 821  *
 822  * Get a reference to the blockdevice at @path in the current
 823  * namespace if possible and return it.  Return ERR_PTR(error)
 824  * otherwise.
 825  */
 826 struct block_device *lookup_bdev(const char *path)
 827 {
 828         struct block_device *bdev;
 829         struct inode *inode;
 830         struct nameidata nd;
 831         int error;
 832
 833         if (!path || !*path)
 834                 return ERR_PTR(-EINVAL);
 835
 836         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
 837         if (error)
 838                 return ERR_PTR(error);
 839
 840         inode = nd.dentry->d_inode;
 841         error = -ENOTBLK;
 842         if (!S_ISBLK(inode->i_mode))
 843                 goto fail;
 844         error = -EACCES;
 845         if (nd.mnt->mnt_flags & MNT_NODEV)
 846                 goto fail;
 847         error = -ENOMEM;
 848         bdev = bd_acquire(inode);
 849         if (!bdev)
 850                 goto fail;
 851 out:
 852         path_release(&nd);
 853         return bdev;
 854 fail:
 855         bdev = ERR_PTR(error);
 856         goto out;
 857 }
 858
 859 /**
 860  * open_bdev_excl  -  open a block device by name and set it up for use
 861  *
 862  * @path:       special file representing the block device
 863  * @flags:      %MS_RDONLY for opening read-only
 864  * @holder:     owner for exclusion
 865  *
 866  * Open the blockdevice described by the special file at @path, claim it
 867  * for the @holder.
 868  */
 869 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
 870 {
 871         struct block_device *bdev;
 872         mode_t mode = FMODE_READ;
 873         int error = 0;
 874
 875         bdev = lookup_bdev(path);
 876         if (IS_ERR(bdev))
 877                 return bdev;
 878
 879         if (!(flags & MS_RDONLY))
 880                 mode |= FMODE_WRITE;
 881         error = blkdev_get(bdev, mode, 0);
 882         if (error)
 883                 return ERR_PTR(error);
 884         error = -EACCES;
 885         if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
 886                 goto blkdev_put;
 887         error = bd_claim(bdev, holder);
 888         if (error)
 889                 goto blkdev_put;
 890
 891         return bdev;
 892
 893 blkdev_put:
 894         blkdev_put(bdev);
 895         return ERR_PTR(error);
 896 }
 897
 898 EXPORT_SYMBOL(open_bdev_excl);
 899
 900 /**
 901  * close_bdev_excl  -  release a blockdevice openen by open_bdev_excl()
 902  *
 903  * @bdev:       blockdevice to close
 904  *
 905  * This is the counterpart to open_bdev_excl().
 906  */
 907 void close_bdev_excl(struct block_device *bdev)
 908 {
 909         bd_release(bdev);
 910         blkdev_put(bdev);
 911 }
 912
 913 EXPORT_SYMBOL(close_bdev_excl);