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
  29 static sector_t max_block(struct block_device *bdev)
  30 {
  31         sector_t retval = ~((sector_t)0);
  32         loff_t sz = i_size_read(bdev->bd_inode);
  33
  34         if (sz) {
  35                 unsigned int size = block_size(bdev);
  36                 unsigned int sizebits = blksize_bits(size);
  37                 retval = (sz >> sizebits);
  38         }
  39         return retval;
  40 }
  41
  42 /* Kill _all_ buffers, dirty or not.. */
  43 static void kill_bdev(struct block_device *bdev)
  44 {
  45         invalidate_bdev(bdev, 1);
  46         truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  47 }
  48
  49 int set_blocksize(struct block_device *bdev, int size)
  50 {
  51         int oldsize;
  52
  53         /* Size must be a power of two, and between 512 and PAGE_SIZE */
  54         if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
  55                 return -EINVAL;
  56
  57         /* Size cannot be smaller than the size supported by the device */
  58         if (size < bdev_hardsect_size(bdev))
  59                 return -EINVAL;
  60
  61         oldsize = bdev->bd_block_size;
  62         if (oldsize == size)
  63                 return 0;
  64
  65         /* Ok, we're actually changing the blocksize.. */
  66         sync_blockdev(bdev);
  67         bdev->bd_block_size = size;
  68         bdev->bd_inode->i_blkbits = blksize_bits(size);
  69         kill_bdev(bdev);
  70         return 0;
  71 }
  72
  73 int sb_set_blocksize(struct super_block *sb, int size)
  74 {
  75         int bits;
  76         if (set_blocksize(sb->s_bdev, size) < 0)
  77                 return 0;
  78         sb->s_blocksize = size;
  79         for (bits = 9, size >>= 9; size >>= 1; bits++)
  80                 ;
  81         sb->s_blocksize_bits = bits;
  82         return sb->s_blocksize;
  83 }
  84
  85 int sb_min_blocksize(struct super_block *sb, int size)
  86 {
  87         int minsize = bdev_hardsect_size(sb->s_bdev);
  88         if (size < minsize)
  89                 size = minsize;
  90         return sb_set_blocksize(sb, size);
  91 }
  92
  93 static int
  94 blkdev_get_block(struct inode *inode, sector_t iblock,
  95                 struct buffer_head *bh, int create)
  96 {
  97         if (iblock >= max_block(inode->i_bdev))
  98                 return -EIO;
  99
 100         bh->b_bdev = inode->i_bdev;
 101         bh->b_blocknr = iblock;
 102         set_buffer_mapped(bh);
 103         return 0;
 104 }
 105
 106 static int
 107 blkdev_get_blocks(struct inode *inode, sector_t iblock,
 108                 unsigned long max_blocks, struct buffer_head *bh, int create)
 109 {
 110         if ((iblock + max_blocks) > max_block(inode->i_bdev))
 111                 return -EIO;
 112
 113         bh->b_bdev = inode->i_bdev;
 114         bh->b_blocknr = iblock;
 115         bh->b_size = max_blocks << inode->i_blkbits;
 116         set_buffer_mapped(bh);
 117         return 0;
 118 }
 119
 120 static int
 121 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 122                         loff_t offset, unsigned long nr_segs)
 123 {
 124         struct file *file = iocb->ki_filp;
 125         struct inode *inode = file->f_dentry->d_inode->i_mapping->host;
 126
 127         return blockdev_direct_IO(rw, iocb, inode, inode->i_bdev, iov, offset,
 128                                 nr_segs, blkdev_get_blocks);
 129 }
 130
 131 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 132 {
 133         return block_write_full_page(page, blkdev_get_block, wbc);
 134 }
 135
 136 static int blkdev_readpage(struct file * file, struct page * page)
 137 {
 138         return block_read_full_page(page, blkdev_get_block);
 139 }
 140
 141 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 142 {
 143         return block_prepare_write(page, from, to, blkdev_get_block);
 144 }
 145
 146 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
 147 {
 148         return block_commit_write(page, from, to);
 149 }
 150
 151 /*
 152  * private llseek:
 153  * for a block special file file->f_dentry->d_inode->i_size is zero
 154  * so we compute the size by hand (just as in block_read/write above)
 155  */
 156 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 157 {
 158         struct inode *bd_inode;
 159         loff_t size;
 160         loff_t retval;
 161
 162         bd_inode = file->f_dentry->d_inode->i_bdev->bd_inode;
 163         down(&bd_inode->i_sem);
 164         size = i_size_read(bd_inode);
 165
 166         switch (origin) {
 167                 case 2:
 168                         offset += size;
 169                         break;
 170                 case 1:
 171                         offset += file->f_pos;
 172         }
 173         retval = -EINVAL;
 174         if (offset >= 0 && offset <= size) {
 175                 if (offset != file->f_pos) {
 176                         file->f_pos = offset;
 177                 }
 178                 retval = offset;
 179         }
 180         up(&bd_inode->i_sem);
 181         return retval;
 182 }
 183
 184 /*
 185  *      Filp may be NULL when we are called by an msync of a vma
 186  *      since the vma has no handle.
 187  */
 188
 189 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 190 {
 191         struct inode * inode = dentry->d_inode;
 192
 193         return sync_blockdev(inode->i_bdev);
 194 }
 195
 196 /*
 197  * pseudo-fs
 198  */
 199
 200 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
 201         int flags, const char *dev_name, void *data)
 202 {
 203         return get_sb_pseudo(fs_type, "bdev:", NULL, 0x62646576);
 204 }
 205
 206 static struct file_system_type bd_type = {
 207         .name           = "bdev",
 208         .get_sb         = bd_get_sb,
 209         .kill_sb        = kill_anon_super,
 210 };
 211
 212 static struct vfsmount *bd_mnt;
 213 struct super_block *blockdev_superblock;
 214
 215 /*
 216  * bdev cache handling - shamelessly stolen from inode.c
 217  * We use smaller hashtable, though.
 218  */
 219
 220 #define HASH_BITS       6
 221 #define HASH_SIZE       (1UL << HASH_BITS)
 222 #define HASH_MASK       (HASH_SIZE-1)
 223 static struct list_head bdev_hashtable[HASH_SIZE];
 224 static spinlock_t bdev_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 225 static kmem_cache_t * bdev_cachep;
 226
 227 #define alloc_bdev() \
 228          ((struct block_device *) kmem_cache_alloc(bdev_cachep, SLAB_KERNEL))
 229 #define destroy_bdev(bdev) kmem_cache_free(bdev_cachep, (bdev))
 230
 231 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 232 {
 233         struct block_device * bdev = (struct block_device *) foo;
 234
 235         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 236             SLAB_CTOR_CONSTRUCTOR)
 237         {
 238                 memset(bdev, 0, sizeof(*bdev));
 239                 sema_init(&bdev->bd_sem, 1);
 240                 INIT_LIST_HEAD(&bdev->bd_inodes);
 241         }
 242 }
 243
 244 void __init bdev_cache_init(void)
 245 {
 246         int i, err;
 247         struct list_head *head = bdev_hashtable;
 248
 249         i = HASH_SIZE;
 250         do {
 251                 INIT_LIST_HEAD(head);
 252                 head++;
 253                 i--;
 254         } while (i);
 255
 256         bdev_cachep = kmem_cache_create("bdev_cache",
 257                                          sizeof(struct block_device),
 258                                          0, SLAB_HWCACHE_ALIGN, init_once,
 259                                          NULL);
 260         if (!bdev_cachep)
 261                 panic("Cannot create bdev_cache SLAB cache");
 262         err = register_filesystem(&bd_type);
 263         if (err)
 264                 panic("Cannot register bdev pseudo-fs");
 265         bd_mnt = kern_mount(&bd_type);
 266         err = PTR_ERR(bd_mnt);
 267         if (IS_ERR(bd_mnt))
 268                 panic("Cannot create bdev pseudo-fs");
 269         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 270 }
 271
 272 /*
 273  * Most likely _very_ bad one - but then it's hardly critical for small
 274  * /dev and can be fixed when somebody will need really large one.
 275  */
 276 static inline unsigned long hash(dev_t dev)
 277 {
 278         unsigned long tmp = dev;
 279         tmp = tmp + (tmp >> HASH_BITS) + (tmp >> HASH_BITS*2);
 280         return tmp & HASH_MASK;
 281 }
 282
 283 static struct block_device *bdfind(dev_t dev, struct list_head *head)
 284 {
 285         struct list_head *p;
 286         struct block_device *bdev;
 287         list_for_each(p, head) {
 288                 bdev = list_entry(p, struct block_device, bd_hash);
 289                 if (bdev->bd_dev != dev)
 290                         continue;
 291                 atomic_inc(&bdev->bd_count);
 292                 return bdev;
 293         }
 294         return NULL;
 295 }
 296
 297 struct block_device *bdget(dev_t dev)
 298 {
 299         struct list_head * head = bdev_hashtable + hash(dev);
 300         struct block_device *bdev, *new_bdev;
 301         spin_lock(&bdev_lock);
 302         bdev = bdfind(dev, head);
 303         spin_unlock(&bdev_lock);
 304         if (bdev)
 305                 return bdev;
 306         new_bdev = alloc_bdev();
 307         if (new_bdev) {
 308                 struct inode *inode = new_inode(bd_mnt->mnt_sb);
 309                 if (inode) {
 310                         kdev_t kdev = to_kdev_t(dev);
 311
 312                         atomic_set(&new_bdev->bd_count,1);
 313                         new_bdev->bd_dev = dev;
 314                         new_bdev->bd_contains = NULL;
 315                         new_bdev->bd_inode = inode;
 316                         new_bdev->bd_block_size = (1 << inode->i_blkbits);
 317                         new_bdev->bd_part_count = 0;
 318                         new_bdev->bd_invalidated = 0;
 319                         inode->i_mode = S_IFBLK;
 320                         inode->i_rdev = kdev;
 321                         inode->i_bdev = new_bdev;
 322                         inode->i_data.a_ops = &def_blk_aops;
 323                         inode->i_data.gfp_mask = GFP_USER;
 324                         inode->i_data.backing_dev_info = &default_backing_dev_info;
 325                         spin_lock(&bdev_lock);
 326                         bdev = bdfind(dev, head);
 327                         if (!bdev) {
 328                                 list_add(&new_bdev->bd_hash, head);
 329                                 spin_unlock(&bdev_lock);
 330                                 return new_bdev;
 331                         }
 332                         spin_unlock(&bdev_lock);
 333                         iput(new_bdev->bd_inode);
 334                 }
 335                 destroy_bdev(new_bdev);
 336         }
 337         return bdev;
 338 }
 339
 340 long nr_blockdev_pages(void)
 341 {
 342         long ret = 0;
 343         int i;
 344
 345         spin_lock(&bdev_lock);
 346         for (i = 0; i < ARRAY_SIZE(bdev_hashtable); i++) {
 347                 struct list_head *head = &bdev_hashtable[i];
 348                 struct list_head *lh;
 349
 350                 if (head == NULL)
 351                         continue;
 352                 list_for_each(lh, head) {
 353                         struct block_device *bdev;
 354
 355                         bdev = list_entry(lh, struct block_device, bd_hash);
 356                         ret += bdev->bd_inode->i_mapping->nrpages;
 357                 }
 358         }
 359         spin_unlock(&bdev_lock);
 360         return ret;
 361 }
 362
 363 static inline void __bd_forget(struct inode *inode)
 364 {
 365         list_del_init(&inode->i_devices);
 366         inode->i_bdev = NULL;
 367         inode->i_mapping = &inode->i_data;
 368 }
 369
 370 void bdput(struct block_device *bdev)
 371 {
 372         if (atomic_dec_and_lock(&bdev->bd_count, &bdev_lock)) {
 373                 struct list_head *p;
 374                 if (bdev->bd_openers)
 375                         BUG();
 376                 list_del(&bdev->bd_hash);
 377                 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 378                         __bd_forget(list_entry(p, struct inode, i_devices));
 379                 }
 380                 spin_unlock(&bdev_lock);
 381                 iput(bdev->bd_inode);
 382                 destroy_bdev(bdev);
 383         }
 384 }
 385
 386 int bd_acquire(struct inode *inode)
 387 {
 388         struct block_device *bdev;
 389         spin_lock(&bdev_lock);
 390         if (inode->i_bdev) {
 391                 atomic_inc(&inode->i_bdev->bd_count);
 392                 spin_unlock(&bdev_lock);
 393                 return 0;
 394         }
 395         spin_unlock(&bdev_lock);
 396         bdev = bdget(kdev_t_to_nr(inode->i_rdev));
 397         if (!bdev)
 398                 return -ENOMEM;
 399         spin_lock(&bdev_lock);
 400         if (!inode->i_bdev) {
 401                 inode->i_bdev = bdev;
 402                 inode->i_mapping = bdev->bd_inode->i_mapping;
 403                 list_add(&inode->i_devices, &bdev->bd_inodes);
 404         } else if (inode->i_bdev != bdev)
 405                 BUG();
 406         spin_unlock(&bdev_lock);
 407         return 0;
 408 }
 409
 410 /* Call when you free inode */
 411
 412 void bd_forget(struct inode *inode)
 413 {
 414         spin_lock(&bdev_lock);
 415         if (inode->i_bdev)
 416                 __bd_forget(inode);
 417         spin_unlock(&bdev_lock);
 418 }
 419
 420 int bd_claim(struct block_device *bdev, void *holder)
 421 {
 422         int res = -EBUSY;
 423         spin_lock(&bdev_lock);
 424         if (!bdev->bd_holder || bdev->bd_holder == holder) {
 425                 bdev->bd_holder = holder;
 426                 bdev->bd_holders++;
 427                 res = 0;
 428         }
 429         spin_unlock(&bdev_lock);
 430         return res;
 431 }
 432
 433 void bd_release(struct block_device *bdev)
 434 {
 435         spin_lock(&bdev_lock);
 436         if (!--bdev->bd_holders)
 437                 bdev->bd_holder = NULL;
 438         spin_unlock(&bdev_lock);
 439 }
 440
 441 /*
 442  * Tries to open block device by device number.  Use it ONLY if you
 443  * really do not have anything better - i.e. when you are behind a
 444  * truly sucky interface and all you are given is a device number.  _Never_
 445  * to be used for internal purposes.  If you ever need it - reconsider
 446  * your API.
 447  */
 448 struct block_device *open_by_devnum(dev_t dev, unsigned mode, int kind)
 449 {
 450         struct block_device *bdev = bdget(dev);
 451         int err = -ENOMEM;
 452         int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
 453         if (bdev)
 454                 err = blkdev_get(bdev, mode, flags, kind);
 455         return err ? ERR_PTR(err) : bdev;
 456 }
 457
 458 /*
 459  * This routine checks whether a removable media has been changed,
 460  * and invalidates all buffer-cache-entries in that case. This
 461  * is a relatively slow routine, so we have to try to minimize using
 462  * it. Thus it is called only upon a 'mount' or 'open'. This
 463  * is the best way of combining speed and utility, I think.
 464  * People changing diskettes in the middle of an operation deserve
 465  * to lose :-)
 466  */
 467 int check_disk_change(struct block_device *bdev)
 468 {
 469         struct gendisk *disk = bdev->bd_disk;
 470         struct block_device_operations * bdops = disk->fops;
 471
 472         if (!bdops->media_changed)
 473                 return 0;
 474         if (!bdops->media_changed(bdev->bd_disk))
 475                 return 0;
 476
 477         if (__invalidate_device(bdev, 0))
 478                 printk("VFS: busy inodes on changed media.\n");
 479
 480         if (bdops->revalidate_disk)
 481                 bdops->revalidate_disk(bdev->bd_disk);
 482         if (bdev->bd_disk->minors > 1)
 483                 bdev->bd_invalidated = 1;
 484         return 1;
 485 }
 486
 487 static void bd_set_size(struct block_device *bdev, loff_t size)
 488 {
 489         unsigned bsize = bdev_hardsect_size(bdev);
 490         i_size_write(bdev->bd_inode, size);
 491         while (bsize < PAGE_CACHE_SIZE) {
 492                 if (size & bsize)
 493                         break;
 494                 bsize <<= 1;
 495         }
 496         bdev->bd_block_size = bsize;
 497         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 498 }
 499
 500 static int do_open(struct block_device *bdev, struct inode *inode, struct file *file)
 501 {
 502         struct module *owner = NULL;
 503         struct gendisk *disk;
 504         int ret = -ENXIO;
 505         int part;
 506
 507         lock_kernel();
 508         disk = get_gendisk(bdev->bd_dev, &part);
 509         if (!disk) {
 510                 unlock_kernel();
 511                 bdput(bdev);
 512                 return ret;
 513         }
 514         owner = disk->fops->owner;
 515
 516         down(&bdev->bd_sem);
 517         if (!bdev->bd_openers) {
 518                 bdev->bd_disk = disk;
 519                 bdev->bd_contains = bdev;
 520                 if (!part) {
 521                         struct backing_dev_info *bdi;
 522                         if (disk->fops->open) {
 523                                 ret = disk->fops->open(inode, file);
 524                                 if (ret)
 525                                         goto out_first;
 526                         }
 527                         bdev->bd_offset = 0;
 528                         if (!bdev->bd_openers) {
 529                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
 530                                 bdi = blk_get_backing_dev_info(bdev);
 531                                 if (bdi == NULL)
 532                                         bdi = &default_backing_dev_info;
 533                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
 534                         }
 535                         if (bdev->bd_invalidated)
 536                                 rescan_partitions(disk, bdev);
 537                 } else {
 538                         struct hd_struct *p;
 539                         struct block_device *whole;
 540                         whole = bdget_disk(disk, 0);
 541                         ret = -ENOMEM;
 542                         if (!whole)
 543                                 goto out_first;
 544                         ret = blkdev_get(whole, file->f_mode, file->f_flags, BDEV_RAW);
 545                         if (ret)
 546                                 goto out_first;
 547                         bdev->bd_contains = whole;
 548                         down(&whole->bd_sem);
 549                         whole->bd_part_count++;
 550                         p = disk->part[part - 1];
 551                         bdev->bd_inode->i_data.backing_dev_info =
 552                            whole->bd_inode->i_data.backing_dev_info;
 553                         if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
 554                                 whole->bd_part_count--;
 555                                 up(&whole->bd_sem);
 556                                 ret = -ENXIO;
 557                                 goto out_first;
 558                         }
 559                         bdev->bd_offset = p->start_sect;
 560                         bd_set_size(bdev, (loff_t) p->nr_sects << 9);
 561                         up(&whole->bd_sem);
 562                 }
 563         } else {
 564                 put_disk(disk);
 565                 module_put(owner);
 566                 if (bdev->bd_contains == bdev) {
 567                         if (bdev->bd_disk->fops->open) {
 568                                 ret = bdev->bd_disk->fops->open(inode, file);
 569                                 if (ret)
 570                                         goto out;
 571                         }
 572                         if (bdev->bd_invalidated)
 573                                 rescan_partitions(bdev->bd_disk, bdev);
 574                 } else {
 575                         down(&bdev->bd_contains->bd_sem);
 576                         bdev->bd_contains->bd_part_count++;
 577                         up(&bdev->bd_contains->bd_sem);
 578                 }
 579         }
 580         bdev->bd_openers++;
 581         up(&bdev->bd_sem);
 582         unlock_kernel();
 583         return 0;
 584
 585 out_first:
 586         bdev->bd_disk = NULL;
 587         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 588         if (bdev != bdev->bd_contains)
 589                 blkdev_put(bdev->bd_contains, BDEV_RAW);
 590         bdev->bd_contains = NULL;
 591         put_disk(disk);
 592         module_put(owner);
 593 out:
 594         up(&bdev->bd_sem);
 595         unlock_kernel();
 596         if (ret)
 597                 bdput(bdev);
 598         return ret;
 599 }
 600
 601 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags, int kind)
 602 {
 603         /*
 604          * This crockload is due to bad choice of ->open() type.
 605          * It will go away.
 606          * For now, block device ->open() routine must _not_
 607          * examine anything in 'inode' argument except ->i_rdev.
 608          */
 609         struct file fake_file = {};
 610         struct dentry fake_dentry = {};
 611         fake_file.f_mode = mode;
 612         fake_file.f_flags = flags;
 613         fake_file.f_dentry = &fake_dentry;
 614         fake_dentry.d_inode = bdev->bd_inode;
 615
 616         return do_open(bdev, bdev->bd_inode, &fake_file);
 617 }
 618
 619 int blkdev_open(struct inode * inode, struct file * filp)
 620 {
 621         struct block_device *bdev;
 622
 623         /*
 624          * Preserve backwards compatibility and allow large file access
 625          * even if userspace doesn't ask for it explicitly. Some mkfs
 626          * binary needs it. We might want to drop this workaround
 627          * during an unstable branch.
 628          */
 629         filp->f_flags |= O_LARGEFILE;
 630
 631         bd_acquire(inode);
 632         bdev = inode->i_bdev;
 633
 634         return do_open(bdev, inode, filp);
 635 }
 636
 637 int blkdev_put(struct block_device *bdev, int kind)
 638 {
 639         int ret = 0;
 640         struct inode *bd_inode = bdev->bd_inode;
 641         struct gendisk *disk = bdev->bd_disk;
 642
 643         down(&bdev->bd_sem);
 644         lock_kernel();
 645         if (!--bdev->bd_openers) {
 646                 switch (kind) {
 647                 case BDEV_FILE:
 648                 case BDEV_FS:
 649                         sync_blockdev(bd_inode->i_bdev);
 650                         break;
 651                 }
 652                 kill_bdev(bdev);
 653         }
 654         if (bdev->bd_contains == bdev) {
 655                 if (disk->fops->release)
 656                         ret = disk->fops->release(bd_inode, NULL);
 657         } else {
 658                 down(&bdev->bd_contains->bd_sem);
 659                 bdev->bd_contains->bd_part_count--;
 660                 up(&bdev->bd_contains->bd_sem);
 661         }
 662         if (!bdev->bd_openers) {
 663                 struct module *owner = disk->fops->owner;
 664
 665                 put_disk(disk);
 666                 module_put(owner);
 667
 668                 bdev->bd_disk = NULL;
 669                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 670                 if (bdev != bdev->bd_contains) {
 671                         blkdev_put(bdev->bd_contains, BDEV_RAW);
 672                 }
 673                 bdev->bd_contains = NULL;
 674         }
 675         unlock_kernel();
 676         up(&bdev->bd_sem);
 677         bdput(bdev);
 678         return ret;
 679 }
 680
 681 int blkdev_close(struct inode * inode, struct file * filp)
 682 {
 683         return blkdev_put(inode->i_bdev, BDEV_FILE);
 684 }
 685
 686 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
 687                                    size_t count, loff_t *ppos)
 688 {
 689         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 690
 691         return generic_file_write_nolock(file, &local_iov, 1, ppos);
 692 }
 693
 694 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
 695                                    size_t count, loff_t pos)
 696 {
 697         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 698
 699         return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
 700 }
 701
 702
 703 struct address_space_operations def_blk_aops = {
 704         .readpage       = blkdev_readpage,
 705         .writepage      = blkdev_writepage,
 706         .sync_page      = block_sync_page,
 707         .prepare_write  = blkdev_prepare_write,
 708         .commit_write   = blkdev_commit_write,
 709         .writepages     = generic_writepages,
 710         .direct_IO      = blkdev_direct_IO,
 711 };
 712
 713 struct file_operations def_blk_fops = {
 714         .open           = blkdev_open,
 715         .release        = blkdev_close,
 716         .llseek         = block_llseek,
 717         .read           = generic_file_read,
 718         .write          = blkdev_file_write,
 719         .aio_read       = generic_file_aio_read,
 720         .aio_write      = blkdev_file_aio_write,
 721         .mmap           = generic_file_mmap,
 722         .fsync          = block_fsync,
 723         .ioctl          = blkdev_ioctl,
 724         .readv          = generic_file_readv,
 725         .writev         = generic_file_writev,
 726         .sendfile       = generic_file_sendfile,
 727 };
 728
 729 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
 730 {
 731         int res;
 732         mm_segment_t old_fs = get_fs();
 733         set_fs(KERNEL_DS);
 734         res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
 735         set_fs(old_fs);
 736         return res;
 737 }
 738
 739 /**
 740  * lookup_bdev  - lookup a struct block_device by name
 741  *
 742  * @path:       special file representing the block device
 743  *
 744  * Get a reference to the blockdevice at @path in the current
 745  * namespace if possible and return it.  Return ERR_PTR(error)
 746  * otherwise.
 747  */
 748 struct block_device *lookup_bdev(const char *path)
 749 {
 750         struct block_device *bdev;
 751         struct inode *inode;
 752         struct nameidata nd;
 753         int error;
 754
 755         if (!path || !*path)
 756                 return ERR_PTR(-EINVAL);
 757
 758         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
 759         if (error)
 760                 return ERR_PTR(error);
 761
 762         inode = nd.dentry->d_inode;
 763         error = -ENOTBLK;
 764         if (!S_ISBLK(inode->i_mode))
 765                 goto fail;
 766         error = -EACCES;
 767         if (nd.mnt->mnt_flags & MNT_NODEV)
 768                 goto fail;
 769         error = bd_acquire(inode);
 770         if (error)
 771                 goto fail;
 772         bdev = inode->i_bdev;
 773
 774 out:
 775         path_release(&nd);
 776         return bdev;
 777 fail:
 778         bdev = ERR_PTR(error);
 779         goto out;
 780 }
 781
 782 /**
 783  * open_bdev_excl  -  open a block device by name and set it up for use
 784  *
 785  * @path:       special file representing the block device
 786  * @flags:      %MS_RDONLY for opening read-only
 787  * @kind:       usage (same as the 4th paramter to blkdev_get)
 788  * @holder:     owner for exclusion
 789  *
 790  * Open the blockdevice described by the special file at @path, claim it
 791  * for the @holder and properly set it up for @kind usage.
 792  */
 793 struct block_device *open_bdev_excl(const char *path, int flags,
 794                                     int kind, void *holder)
 795 {
 796         struct block_device *bdev;
 797         mode_t mode = FMODE_READ;
 798         int error = 0;
 799
 800         bdev = lookup_bdev(path);
 801         if (IS_ERR(bdev))
 802                 return bdev;
 803
 804         if (!(flags & MS_RDONLY))
 805                 mode |= FMODE_WRITE;
 806         error = blkdev_get(bdev, mode, 0, kind);
 807         if (error)
 808                 return ERR_PTR(error);
 809         error = -EACCES;
 810         if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
 811                 goto blkdev_put;
 812         error = bd_claim(bdev, holder);
 813         if (error)
 814                 goto blkdev_put;
 815
 816         return bdev;
 817
 818 blkdev_put:
 819         blkdev_put(bdev, BDEV_FS);
 820         return ERR_PTR(error);
 821 }
 822
 823 /**
 824  * close_bdev_excl  -  release a blockdevice openen by open_bdev_excl()
 825  *
 826  * @bdev:       blockdevice to close
 827  * @kind:       usage (same as the 4th paramter to blkdev_get)
 828  *
 829  * This is the counterpart to open_bdev_excl().
 830  */
 831 void close_bdev_excl(struct block_device *bdev, int kind)
 832 {
 833         bd_release(bdev);
 834         blkdev_put(bdev, kind);
 835 }