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         sector_t end_block = max_block(I_BDEV(inode));
 141
 142         if ((iblock + max_blocks) > end_block) {
 143                 max_blocks = end_block - iblock;
 144                 if ((long)max_blocks <= 0) {
 145                         if (create)
 146                                 return -EIO;    /* write fully beyond EOF */
 147                         /*
 148                          * It is a read which is fully beyond EOF.  We return
 149                          * a !buffer_mapped buffer
 150                          */
 151                         max_blocks = 0;
 152                 }
 153         }
 154
 155         bh->b_bdev = I_BDEV(inode);
 156         bh->b_blocknr = iblock;
 157         bh->b_size = max_blocks << inode->i_blkbits;
 158         if (max_blocks)
 159                 set_buffer_mapped(bh);
 160         return 0;
 161 }
 162
 163 static ssize_t
 164 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 165                         loff_t offset, unsigned long nr_segs)
 166 {
 167         struct file *file = iocb->ki_filp;
 168         struct inode *inode = file->f_mapping->host;
 169
 170         return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
 171                                 iov, offset, nr_segs, blkdev_get_blocks, NULL);
 172 }
 173
 174 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 175 {
 176         return block_write_full_page(page, blkdev_get_block, wbc);
 177 }
 178
 179 static int blkdev_readpage(struct file * file, struct page * page)
 180 {
 181         return block_read_full_page(page, blkdev_get_block);
 182 }
 183
 184 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 185 {
 186         return block_prepare_write(page, from, to, blkdev_get_block);
 187 }
 188
 189 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
 190 {
 191         return block_commit_write(page, from, to);
 192 }
 193
 194 /*
 195  * private llseek:
 196  * for a block special file file->f_dentry->d_inode->i_size is zero
 197  * so we compute the size by hand (just as in block_read/write above)
 198  */
 199 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 200 {
 201         struct inode *bd_inode = file->f_mapping->host;
 202         loff_t size;
 203         loff_t retval;
 204
 205         mutex_lock(&bd_inode->i_mutex);
 206         size = i_size_read(bd_inode);
 207
 208         switch (origin) {
 209                 case 2:
 210                         offset += size;
 211                         break;
 212                 case 1:
 213                         offset += file->f_pos;
 214         }
 215         retval = -EINVAL;
 216         if (offset >= 0 && offset <= size) {
 217                 if (offset != file->f_pos) {
 218                         file->f_pos = offset;
 219                 }
 220                 retval = offset;
 221         }
 222         mutex_unlock(&bd_inode->i_mutex);
 223         return retval;
 224 }
 225
 226 /*
 227  *      Filp is never NULL; the only case when ->fsync() is called with
 228  *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 229  */
 230
 231 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 232 {
 233         return sync_blockdev(I_BDEV(filp->f_mapping->host));
 234 }
 235
 236 /*
 237  * pseudo-fs
 238  */
 239
 240 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 241 static kmem_cache_t * bdev_cachep;
 242
 243 static struct inode *bdev_alloc_inode(struct super_block *sb)
 244 {
 245         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
 246         if (!ei)
 247                 return NULL;
 248         return &ei->vfs_inode;
 249 }
 250
 251 static void bdev_destroy_inode(struct inode *inode)
 252 {
 253         struct bdev_inode *bdi = BDEV_I(inode);
 254
 255         bdi->bdev.bd_inode_backing_dev_info = NULL;
 256         kmem_cache_free(bdev_cachep, bdi);
 257 }
 258
 259 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 260 {
 261         struct bdev_inode *ei = (struct bdev_inode *) foo;
 262         struct block_device *bdev = &ei->bdev;
 263
 264         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 265             SLAB_CTOR_CONSTRUCTOR)
 266         {
 267                 memset(bdev, 0, sizeof(*bdev));
 268                 mutex_init(&bdev->bd_mutex);
 269                 mutex_init(&bdev->bd_mount_mutex);
 270                 INIT_LIST_HEAD(&bdev->bd_inodes);
 271                 INIT_LIST_HEAD(&bdev->bd_list);
 272                 inode_init_once(&ei->vfs_inode);
 273         }
 274 }
 275
 276 static inline void __bd_forget(struct inode *inode)
 277 {
 278         list_del_init(&inode->i_devices);
 279         inode->i_bdev = NULL;
 280         inode->i_mapping = &inode->i_data;
 281 }
 282
 283 static void bdev_clear_inode(struct inode *inode)
 284 {
 285         struct block_device *bdev = &BDEV_I(inode)->bdev;
 286         struct list_head *p;
 287         spin_lock(&bdev_lock);
 288         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 289                 __bd_forget(list_entry(p, struct inode, i_devices));
 290         }
 291         list_del_init(&bdev->bd_list);
 292         spin_unlock(&bdev_lock);
 293 }
 294
 295 static struct super_operations bdev_sops = {
 296         .statfs = simple_statfs,
 297         .alloc_inode = bdev_alloc_inode,
 298         .destroy_inode = bdev_destroy_inode,
 299         .drop_inode = generic_delete_inode,
 300         .clear_inode = bdev_clear_inode,
 301 };
 302
 303 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
 304         int flags, const char *dev_name, void *data)
 305 {
 306         return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
 307 }
 308
 309 static struct file_system_type bd_type = {
 310         .name           = "bdev",
 311         .get_sb         = bd_get_sb,
 312         .kill_sb        = kill_anon_super,
 313 };
 314
 315 static struct vfsmount *bd_mnt;
 316 struct super_block *blockdev_superblock;
 317
 318 void __init bdev_cache_init(void)
 319 {
 320         int err;
 321         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 322                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 323                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 324                         init_once, NULL);
 325         err = register_filesystem(&bd_type);
 326         if (err)
 327                 panic("Cannot register bdev pseudo-fs");
 328         bd_mnt = kern_mount(&bd_type);
 329         err = PTR_ERR(bd_mnt);
 330         if (IS_ERR(bd_mnt))
 331                 panic("Cannot create bdev pseudo-fs");
 332         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 333 }
 334
 335 /*
 336  * Most likely _very_ bad one - but then it's hardly critical for small
 337  * /dev and can be fixed when somebody will need really large one.
 338  * Keep in mind that it will be fed through icache hash function too.
 339  */
 340 static inline unsigned long hash(dev_t dev)
 341 {
 342         return MAJOR(dev)+MINOR(dev);
 343 }
 344
 345 static int bdev_test(struct inode *inode, void *data)
 346 {
 347         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 348 }
 349
 350 static int bdev_set(struct inode *inode, void *data)
 351 {
 352         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 353         return 0;
 354 }
 355
 356 static LIST_HEAD(all_bdevs);
 357
 358 struct block_device *bdget(dev_t dev)
 359 {
 360         struct block_device *bdev;
 361         struct inode *inode;
 362
 363         inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
 364                         bdev_test, bdev_set, &dev);
 365
 366         if (!inode)
 367                 return NULL;
 368
 369         bdev = &BDEV_I(inode)->bdev;
 370
 371         if (inode->i_state & I_NEW) {
 372                 bdev->bd_contains = NULL;
 373                 bdev->bd_inode = inode;
 374                 bdev->bd_block_size = (1 << inode->i_blkbits);
 375                 bdev->bd_part_count = 0;
 376                 bdev->bd_invalidated = 0;
 377                 inode->i_mode = S_IFBLK;
 378                 inode->i_rdev = dev;
 379                 inode->i_bdev = bdev;
 380                 inode->i_data.a_ops = &def_blk_aops;
 381                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 382                 inode->i_data.backing_dev_info = &default_backing_dev_info;
 383                 spin_lock(&bdev_lock);
 384                 list_add(&bdev->bd_list, &all_bdevs);
 385                 spin_unlock(&bdev_lock);
 386                 unlock_new_inode(inode);
 387         }
 388         return bdev;
 389 }
 390
 391 EXPORT_SYMBOL(bdget);
 392
 393 long nr_blockdev_pages(void)
 394 {
 395         struct list_head *p;
 396         long ret = 0;
 397         spin_lock(&bdev_lock);
 398         list_for_each(p, &all_bdevs) {
 399                 struct block_device *bdev;
 400                 bdev = list_entry(p, struct block_device, bd_list);
 401                 ret += bdev->bd_inode->i_mapping->nrpages;
 402         }
 403         spin_unlock(&bdev_lock);
 404         return ret;
 405 }
 406
 407 void bdput(struct block_device *bdev)
 408 {
 409         iput(bdev->bd_inode);
 410 }
 411
 412 EXPORT_SYMBOL(bdput);
 413
 414 static struct block_device *bd_acquire(struct inode *inode)
 415 {
 416         struct block_device *bdev;
 417         spin_lock(&bdev_lock);
 418         bdev = inode->i_bdev;
 419         if (bdev && igrab(bdev->bd_inode)) {
 420                 spin_unlock(&bdev_lock);
 421                 return bdev;
 422         }
 423         spin_unlock(&bdev_lock);
 424         bdev = bdget(inode->i_rdev);
 425         if (bdev) {
 426                 spin_lock(&bdev_lock);
 427                 if (inode->i_bdev)
 428                         __bd_forget(inode);
 429                 inode->i_bdev = bdev;
 430                 inode->i_mapping = bdev->bd_inode->i_mapping;
 431                 list_add(&inode->i_devices, &bdev->bd_inodes);
 432                 spin_unlock(&bdev_lock);
 433         }
 434         return bdev;
 435 }
 436
 437 /* Call when you free inode */
 438
 439 void bd_forget(struct inode *inode)
 440 {
 441         spin_lock(&bdev_lock);
 442         if (inode->i_bdev)
 443                 __bd_forget(inode);
 444         spin_unlock(&bdev_lock);
 445 }
 446
 447 int bd_claim(struct block_device *bdev, void *holder)
 448 {
 449         int res;
 450         spin_lock(&bdev_lock);
 451
 452         /* first decide result */
 453         if (bdev->bd_holder == holder)
 454                 res = 0;         /* already a holder */
 455         else if (bdev->bd_holder != NULL)
 456                 res = -EBUSY;    /* held by someone else */
 457         else if (bdev->bd_contains == bdev)
 458                 res = 0;         /* is a whole device which isn't held */
 459
 460         else if (bdev->bd_contains->bd_holder == bd_claim)
 461                 res = 0;         /* is a partition of a device that is being partitioned */
 462         else if (bdev->bd_contains->bd_holder != NULL)
 463                 res = -EBUSY;    /* is a partition of a held device */
 464         else
 465                 res = 0;         /* is a partition of an un-held device */
 466
 467         /* now impose change */
 468         if (res==0) {
 469                 /* note that for a whole device bd_holders
 470                  * will be incremented twice, and bd_holder will
 471                  * be set to bd_claim before being set to holder
 472                  */
 473                 bdev->bd_contains->bd_holders ++;
 474                 bdev->bd_contains->bd_holder = bd_claim;
 475                 bdev->bd_holders++;
 476                 bdev->bd_holder = holder;
 477         }
 478         spin_unlock(&bdev_lock);
 479         return res;
 480 }
 481
 482 EXPORT_SYMBOL(bd_claim);
 483
 484 void bd_release(struct block_device *bdev)
 485 {
 486         spin_lock(&bdev_lock);
 487         if (!--bdev->bd_contains->bd_holders)
 488                 bdev->bd_contains->bd_holder = NULL;
 489         if (!--bdev->bd_holders)
 490                 bdev->bd_holder = NULL;
 491         spin_unlock(&bdev_lock);
 492 }
 493
 494 EXPORT_SYMBOL(bd_release);
 495
 496 /*
 497  * Tries to open block device by device number.  Use it ONLY if you
 498  * really do not have anything better - i.e. when you are behind a
 499  * truly sucky interface and all you are given is a device number.  _Never_
 500  * to be used for internal purposes.  If you ever need it - reconsider
 501  * your API.
 502  */
 503 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
 504 {
 505         struct block_device *bdev = bdget(dev);
 506         int err = -ENOMEM;
 507         int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
 508         if (bdev)
 509                 err = blkdev_get(bdev, mode, flags);
 510         return err ? ERR_PTR(err) : bdev;
 511 }
 512
 513 EXPORT_SYMBOL(open_by_devnum);
 514
 515 /*
 516  * This routine checks whether a removable media has been changed,
 517  * and invalidates all buffer-cache-entries in that case. This
 518  * is a relatively slow routine, so we have to try to minimize using
 519  * it. Thus it is called only upon a 'mount' or 'open'. This
 520  * is the best way of combining speed and utility, I think.
 521  * People changing diskettes in the middle of an operation deserve
 522  * to lose :-)
 523  */
 524 int check_disk_change(struct block_device *bdev)
 525 {
 526         struct gendisk *disk = bdev->bd_disk;
 527         struct block_device_operations * bdops = disk->fops;
 528
 529         if (!bdops->media_changed)
 530                 return 0;
 531         if (!bdops->media_changed(bdev->bd_disk))
 532                 return 0;
 533
 534         if (__invalidate_device(bdev))
 535                 printk("VFS: busy inodes on changed media.\n");
 536
 537         if (bdops->revalidate_disk)
 538                 bdops->revalidate_disk(bdev->bd_disk);
 539         if (bdev->bd_disk->minors > 1)
 540                 bdev->bd_invalidated = 1;
 541         return 1;
 542 }
 543
 544 EXPORT_SYMBOL(check_disk_change);
 545
 546 void bd_set_size(struct block_device *bdev, loff_t size)
 547 {
 548         unsigned bsize = bdev_hardsect_size(bdev);
 549
 550         bdev->bd_inode->i_size = size;
 551         while (bsize < PAGE_CACHE_SIZE) {
 552                 if (size & bsize)
 553                         break;
 554                 bsize <<= 1;
 555         }
 556         bdev->bd_block_size = bsize;
 557         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 558 }
 559 EXPORT_SYMBOL(bd_set_size);
 560
 561 static int do_open(struct block_device *bdev, struct file *file)
 562 {
 563         struct module *owner = NULL;
 564         struct gendisk *disk;
 565         int ret = -ENXIO;
 566         int part;
 567
 568         file->f_mapping = bdev->bd_inode->i_mapping;
 569         lock_kernel();
 570         disk = get_gendisk(bdev->bd_dev, &part);
 571         if (!disk) {
 572                 unlock_kernel();
 573                 bdput(bdev);
 574                 return ret;
 575         }
 576         owner = disk->fops->owner;
 577
 578         mutex_lock(&bdev->bd_mutex);
 579         if (!bdev->bd_openers) {
 580                 bdev->bd_disk = disk;
 581                 bdev->bd_contains = bdev;
 582                 if (!part) {
 583                         struct backing_dev_info *bdi;
 584                         if (disk->fops->open) {
 585                                 ret = disk->fops->open(bdev->bd_inode, file);
 586                                 if (ret)
 587                                         goto out_first;
 588                         }
 589                         if (!bdev->bd_openers) {
 590                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
 591                                 bdi = blk_get_backing_dev_info(bdev);
 592                                 if (bdi == NULL)
 593                                         bdi = &default_backing_dev_info;
 594                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
 595                         }
 596                         if (bdev->bd_invalidated)
 597                                 rescan_partitions(disk, bdev);
 598                 } else {
 599                         struct hd_struct *p;
 600                         struct block_device *whole;
 601                         whole = bdget_disk(disk, 0);
 602                         ret = -ENOMEM;
 603                         if (!whole)
 604                                 goto out_first;
 605                         ret = blkdev_get(whole, file->f_mode, file->f_flags);
 606                         if (ret)
 607                                 goto out_first;
 608                         bdev->bd_contains = whole;
 609                         mutex_lock(&whole->bd_mutex);
 610                         whole->bd_part_count++;
 611                         p = disk->part[part - 1];
 612                         bdev->bd_inode->i_data.backing_dev_info =
 613                            whole->bd_inode->i_data.backing_dev_info;
 614                         if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
 615                                 whole->bd_part_count--;
 616                                 mutex_unlock(&whole->bd_mutex);
 617                                 ret = -ENXIO;
 618                                 goto out_first;
 619                         }
 620                         kobject_get(&p->kobj);
 621                         bdev->bd_part = p;
 622                         bd_set_size(bdev, (loff_t) p->nr_sects << 9);
 623                         mutex_unlock(&whole->bd_mutex);
 624                 }
 625         } else {
 626                 put_disk(disk);
 627                 module_put(owner);
 628                 if (bdev->bd_contains == bdev) {
 629                         if (bdev->bd_disk->fops->open) {
 630                                 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
 631                                 if (ret)
 632                                         goto out;
 633                         }
 634                         if (bdev->bd_invalidated)
 635                                 rescan_partitions(bdev->bd_disk, bdev);
 636                 } else {
 637                         mutex_lock(&bdev->bd_contains->bd_mutex);
 638                         bdev->bd_contains->bd_part_count++;
 639                         mutex_unlock(&bdev->bd_contains->bd_mutex);
 640                 }
 641         }
 642         bdev->bd_openers++;
 643         mutex_unlock(&bdev->bd_mutex);
 644         unlock_kernel();
 645         return 0;
 646
 647 out_first:
 648         bdev->bd_disk = NULL;
 649         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 650         if (bdev != bdev->bd_contains)
 651                 blkdev_put(bdev->bd_contains);
 652         bdev->bd_contains = NULL;
 653         put_disk(disk);
 654         module_put(owner);
 655 out:
 656         mutex_unlock(&bdev->bd_mutex);
 657         unlock_kernel();
 658         if (ret)
 659                 bdput(bdev);
 660         return ret;
 661 }
 662
 663 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
 664 {
 665         /*
 666          * This crockload is due to bad choice of ->open() type.
 667          * It will go away.
 668          * For now, block device ->open() routine must _not_
 669          * examine anything in 'inode' argument except ->i_rdev.
 670          */
 671         struct file fake_file = {};
 672         struct dentry fake_dentry = {};
 673         fake_file.f_mode = mode;
 674         fake_file.f_flags = flags;
 675         fake_file.f_dentry = &fake_dentry;
 676         fake_dentry.d_inode = bdev->bd_inode;
 677
 678         return do_open(bdev, &fake_file);
 679 }
 680
 681 EXPORT_SYMBOL(blkdev_get);
 682
 683 static int blkdev_open(struct inode * inode, struct file * filp)
 684 {
 685         struct block_device *bdev;
 686         int res;
 687
 688         /*
 689          * Preserve backwards compatibility and allow large file access
 690          * even if userspace doesn't ask for it explicitly. Some mkfs
 691          * binary needs it. We might want to drop this workaround
 692          * during an unstable branch.
 693          */
 694         filp->f_flags |= O_LARGEFILE;
 695
 696         bdev = bd_acquire(inode);
 697
 698         res = do_open(bdev, filp);
 699         if (res)
 700                 return res;
 701
 702         if (!(filp->f_flags & O_EXCL) )
 703                 return 0;
 704
 705         if (!(res = bd_claim(bdev, filp)))
 706                 return 0;
 707
 708         blkdev_put(bdev);
 709         return res;
 710 }
 711
 712 int blkdev_put(struct block_device *bdev)
 713 {
 714         int ret = 0;
 715         struct inode *bd_inode = bdev->bd_inode;
 716         struct gendisk *disk = bdev->bd_disk;
 717
 718         mutex_lock(&bdev->bd_mutex);
 719         lock_kernel();
 720         if (!--bdev->bd_openers) {
 721                 sync_blockdev(bdev);
 722                 kill_bdev(bdev);
 723         }
 724         if (bdev->bd_contains == bdev) {
 725                 if (disk->fops->release)
 726                         ret = disk->fops->release(bd_inode, NULL);
 727         } else {
 728                 mutex_lock(&bdev->bd_contains->bd_mutex);
 729                 bdev->bd_contains->bd_part_count--;
 730                 mutex_unlock(&bdev->bd_contains->bd_mutex);
 731         }
 732         if (!bdev->bd_openers) {
 733                 struct module *owner = disk->fops->owner;
 734
 735                 put_disk(disk);
 736                 module_put(owner);
 737
 738                 if (bdev->bd_contains != bdev) {
 739                         kobject_put(&bdev->bd_part->kobj);
 740                         bdev->bd_part = NULL;
 741                 }
 742                 bdev->bd_disk = NULL;
 743                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 744                 if (bdev != bdev->bd_contains) {
 745                         blkdev_put(bdev->bd_contains);
 746                 }
 747                 bdev->bd_contains = NULL;
 748         }
 749         unlock_kernel();
 750         mutex_unlock(&bdev->bd_mutex);
 751         bdput(bdev);
 752         return ret;
 753 }
 754
 755 EXPORT_SYMBOL(blkdev_put);
 756
 757 static int blkdev_close(struct inode * inode, struct file * filp)
 758 {
 759         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
 760         if (bdev->bd_holder == filp)
 761                 bd_release(bdev);
 762         return blkdev_put(bdev);
 763 }
 764
 765 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
 766                                    size_t count, loff_t *ppos)
 767 {
 768         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 769
 770         return generic_file_write_nolock(file, &local_iov, 1, ppos);
 771 }
 772
 773 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
 774                                    size_t count, loff_t pos)
 775 {
 776         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 777
 778         return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
 779 }
 780
 781 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 782 {
 783         return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
 784 }
 785
 786 struct address_space_operations def_blk_aops = {
 787         .readpage       = blkdev_readpage,
 788         .writepage      = blkdev_writepage,
 789         .sync_page      = block_sync_page,
 790         .prepare_write  = blkdev_prepare_write,
 791         .commit_write   = blkdev_commit_write,
 792         .writepages     = generic_writepages,
 793         .direct_IO      = blkdev_direct_IO,
 794 };
 795
 796 struct file_operations def_blk_fops = {
 797         .open           = blkdev_open,
 798         .release        = blkdev_close,
 799         .llseek         = block_llseek,
 800         .read           = generic_file_read,
 801         .write          = blkdev_file_write,
 802         .aio_read       = generic_file_aio_read,
 803         .aio_write      = blkdev_file_aio_write,
 804         .mmap           = generic_file_mmap,
 805         .fsync          = block_fsync,
 806         .unlocked_ioctl = block_ioctl,
 807 #ifdef CONFIG_COMPAT
 808         .compat_ioctl   = compat_blkdev_ioctl,
 809 #endif
 810         .readv          = generic_file_readv,
 811         .writev         = generic_file_write_nolock,
 812         .sendfile       = generic_file_sendfile,
 813 };
 814
 815 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
 816 {
 817         int res;
 818         mm_segment_t old_fs = get_fs();
 819         set_fs(KERNEL_DS);
 820         res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
 821         set_fs(old_fs);
 822         return res;
 823 }
 824
 825 EXPORT_SYMBOL(ioctl_by_bdev);
 826
 827 /**
 828  * lookup_bdev  - lookup a struct block_device by name
 829  *
 830  * @path:       special file representing the block device
 831  *
 832  * Get a reference to the blockdevice at @path in the current
 833  * namespace if possible and return it.  Return ERR_PTR(error)
 834  * otherwise.
 835  */
 836 struct block_device *lookup_bdev(const char *path)
 837 {
 838         struct block_device *bdev;
 839         struct inode *inode;
 840         struct nameidata nd;
 841         int error;
 842
 843         if (!path || !*path)
 844                 return ERR_PTR(-EINVAL);
 845
 846         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
 847         if (error)
 848                 return ERR_PTR(error);
 849
 850         inode = nd.dentry->d_inode;
 851         error = -ENOTBLK;
 852         if (!S_ISBLK(inode->i_mode))
 853                 goto fail;
 854         error = -EACCES;
 855         if (nd.mnt->mnt_flags & MNT_NODEV)
 856                 goto fail;
 857         error = -ENOMEM;
 858         bdev = bd_acquire(inode);
 859         if (!bdev)
 860                 goto fail;
 861 out:
 862         path_release(&nd);
 863         return bdev;
 864 fail:
 865         bdev = ERR_PTR(error);
 866         goto out;
 867 }
 868
 869 /**
 870  * open_bdev_excl  -  open a block device by name and set it up for use
 871  *
 872  * @path:       special file representing the block device
 873  * @flags:      %MS_RDONLY for opening read-only
 874  * @holder:     owner for exclusion
 875  *
 876  * Open the blockdevice described by the special file at @path, claim it
 877  * for the @holder.
 878  */
 879 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
 880 {
 881         struct block_device *bdev;
 882         mode_t mode = FMODE_READ;
 883         int error = 0;
 884
 885         bdev = lookup_bdev(path);
 886         if (IS_ERR(bdev))
 887                 return bdev;
 888
 889         if (!(flags & MS_RDONLY))
 890                 mode |= FMODE_WRITE;
 891         error = blkdev_get(bdev, mode, 0);
 892         if (error)
 893                 return ERR_PTR(error);
 894         error = -EACCES;
 895         if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
 896                 goto blkdev_put;
 897         error = bd_claim(bdev, holder);
 898         if (error)
 899                 goto blkdev_put;
 900
 901         return bdev;
 902
 903 blkdev_put:
 904         blkdev_put(bdev);
 905         return ERR_PTR(error);
 906 }
 907
 908 EXPORT_SYMBOL(open_bdev_excl);
 909
 910 /**
 911  * close_bdev_excl  -  release a blockdevice openen by open_bdev_excl()
 912  *
 913  * @bdev:       blockdevice to close
 914  *
 915  * This is the counterpart to open_bdev_excl().
 916  */
 917 void close_bdev_excl(struct block_device *bdev)
 918 {
 919         bd_release(bdev);
 920         blkdev_put(bdev);
 921 }
 922
 923 EXPORT_SYMBOL(close_bdev_excl);