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         if (set_blocksize(sb->s_bdev, size))
  90                 return 0;
  91         /* If we get here, we know size is power of two
  92          * and it's value is between 512 and PAGE_SIZE */
  93         sb->s_blocksize = size;
  94         sb->s_blocksize_bits = blksize_bits(size);
  95         return sb->s_blocksize;
  96 }
  97
  98 EXPORT_SYMBOL(sb_set_blocksize);
  99
 100 int sb_min_blocksize(struct super_block *sb, int size)
 101 {
 102         int minsize = bdev_hardsect_size(sb->s_bdev);
 103         if (size < minsize)
 104                 size = minsize;
 105         return sb_set_blocksize(sb, size);
 106 }
 107
 108 EXPORT_SYMBOL(sb_min_blocksize);
 109
 110 static int
 111 blkdev_get_block(struct inode *inode, sector_t iblock,
 112                 struct buffer_head *bh, int create)
 113 {
 114         if (iblock >= max_block(I_BDEV(inode))) {
 115                 if (create)
 116                         return -EIO;
 117
 118                 /*
 119                  * for reads, we're just trying to fill a partial page.
 120                  * return a hole, they will have to call get_block again
 121                  * before they can fill it, and they will get -EIO at that
 122                  * time
 123                  */
 124                 return 0;
 125         }
 126         bh->b_bdev = I_BDEV(inode);
 127         bh->b_blocknr = iblock;
 128         set_buffer_mapped(bh);
 129         return 0;
 130 }
 131
 132 static int
 133 blkdev_get_blocks(struct inode *inode, sector_t iblock,
 134                 struct buffer_head *bh, int create)
 135 {
 136         sector_t end_block = max_block(I_BDEV(inode));
 137         unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
 138
 139         if ((iblock + max_blocks) > end_block) {
 140                 max_blocks = end_block - iblock;
 141                 if ((long)max_blocks <= 0) {
 142                         if (create)
 143                                 return -EIO;    /* write fully beyond EOF */
 144                         /*
 145                          * It is a read which is fully beyond EOF.  We return
 146                          * a !buffer_mapped buffer
 147                          */
 148                         max_blocks = 0;
 149                 }
 150         }
 151
 152         bh->b_bdev = I_BDEV(inode);
 153         bh->b_blocknr = iblock;
 154         bh->b_size = max_blocks << inode->i_blkbits;
 155         if (max_blocks)
 156                 set_buffer_mapped(bh);
 157         return 0;
 158 }
 159
 160 static ssize_t
 161 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 162                         loff_t offset, unsigned long nr_segs)
 163 {
 164         struct file *file = iocb->ki_filp;
 165         struct inode *inode = file->f_mapping->host;
 166
 167         return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
 168                                 iov, offset, nr_segs, blkdev_get_blocks, NULL);
 169 }
 170
 171 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 172 {
 173         return block_write_full_page(page, blkdev_get_block, wbc);
 174 }
 175
 176 static int blkdev_readpage(struct file * file, struct page * page)
 177 {
 178         return block_read_full_page(page, blkdev_get_block);
 179 }
 180
 181 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
 182 {
 183         return block_prepare_write(page, from, to, blkdev_get_block);
 184 }
 185
 186 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
 187 {
 188         return block_commit_write(page, from, to);
 189 }
 190
 191 /*
 192  * private llseek:
 193  * for a block special file file->f_dentry->d_inode->i_size is zero
 194  * so we compute the size by hand (just as in block_read/write above)
 195  */
 196 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 197 {
 198         struct inode *bd_inode = file->f_mapping->host;
 199         loff_t size;
 200         loff_t retval;
 201
 202         mutex_lock(&bd_inode->i_mutex);
 203         size = i_size_read(bd_inode);
 204
 205         switch (origin) {
 206                 case 2:
 207                         offset += size;
 208                         break;
 209                 case 1:
 210                         offset += file->f_pos;
 211         }
 212         retval = -EINVAL;
 213         if (offset >= 0 && offset <= size) {
 214                 if (offset != file->f_pos) {
 215                         file->f_pos = offset;
 216                 }
 217                 retval = offset;
 218         }
 219         mutex_unlock(&bd_inode->i_mutex);
 220         return retval;
 221 }
 222
 223 /*
 224  *      Filp is never NULL; the only case when ->fsync() is called with
 225  *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 226  */
 227
 228 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 229 {
 230         return sync_blockdev(I_BDEV(filp->f_mapping->host));
 231 }
 232
 233 /*
 234  * pseudo-fs
 235  */
 236
 237 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 238 static kmem_cache_t * bdev_cachep __read_mostly;
 239
 240 static struct inode *bdev_alloc_inode(struct super_block *sb)
 241 {
 242         struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
 243         if (!ei)
 244                 return NULL;
 245         return &ei->vfs_inode;
 246 }
 247
 248 static void bdev_destroy_inode(struct inode *inode)
 249 {
 250         struct bdev_inode *bdi = BDEV_I(inode);
 251
 252         bdi->bdev.bd_inode_backing_dev_info = NULL;
 253         kmem_cache_free(bdev_cachep, bdi);
 254 }
 255
 256 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 257 {
 258         struct bdev_inode *ei = (struct bdev_inode *) foo;
 259         struct block_device *bdev = &ei->bdev;
 260
 261         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 262             SLAB_CTOR_CONSTRUCTOR)
 263         {
 264                 memset(bdev, 0, sizeof(*bdev));
 265                 mutex_init(&bdev->bd_mutex);
 266                 mutex_init(&bdev->bd_mount_mutex);
 267                 INIT_LIST_HEAD(&bdev->bd_inodes);
 268                 INIT_LIST_HEAD(&bdev->bd_list);
 269                 inode_init_once(&ei->vfs_inode);
 270         }
 271 }
 272
 273 static inline void __bd_forget(struct inode *inode)
 274 {
 275         list_del_init(&inode->i_devices);
 276         inode->i_bdev = NULL;
 277         inode->i_mapping = &inode->i_data;
 278 }
 279
 280 static void bdev_clear_inode(struct inode *inode)
 281 {
 282         struct block_device *bdev = &BDEV_I(inode)->bdev;
 283         struct list_head *p;
 284         spin_lock(&bdev_lock);
 285         while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 286                 __bd_forget(list_entry(p, struct inode, i_devices));
 287         }
 288         list_del_init(&bdev->bd_list);
 289         spin_unlock(&bdev_lock);
 290 }
 291
 292 static struct super_operations bdev_sops = {
 293         .statfs = simple_statfs,
 294         .alloc_inode = bdev_alloc_inode,
 295         .destroy_inode = bdev_destroy_inode,
 296         .drop_inode = generic_delete_inode,
 297         .clear_inode = bdev_clear_inode,
 298 };
 299
 300 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
 301         int flags, const char *dev_name, void *data)
 302 {
 303         return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
 304 }
 305
 306 static struct file_system_type bd_type = {
 307         .name           = "bdev",
 308         .get_sb         = bd_get_sb,
 309         .kill_sb        = kill_anon_super,
 310 };
 311
 312 static struct vfsmount *bd_mnt __read_mostly;
 313 struct super_block *blockdev_superblock;
 314
 315 void __init bdev_cache_init(void)
 316 {
 317         int err;
 318         bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 319                         0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 320                                 SLAB_MEM_SPREAD|SLAB_PANIC),
 321                         init_once, NULL);
 322         err = register_filesystem(&bd_type);
 323         if (err)
 324                 panic("Cannot register bdev pseudo-fs");
 325         bd_mnt = kern_mount(&bd_type);
 326         err = PTR_ERR(bd_mnt);
 327         if (IS_ERR(bd_mnt))
 328                 panic("Cannot create bdev pseudo-fs");
 329         blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 330 }
 331
 332 /*
 333  * Most likely _very_ bad one - but then it's hardly critical for small
 334  * /dev and can be fixed when somebody will need really large one.
 335  * Keep in mind that it will be fed through icache hash function too.
 336  */
 337 static inline unsigned long hash(dev_t dev)
 338 {
 339         return MAJOR(dev)+MINOR(dev);
 340 }
 341
 342 static int bdev_test(struct inode *inode, void *data)
 343 {
 344         return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 345 }
 346
 347 static int bdev_set(struct inode *inode, void *data)
 348 {
 349         BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 350         return 0;
 351 }
 352
 353 static LIST_HEAD(all_bdevs);
 354
 355 struct block_device *bdget(dev_t dev)
 356 {
 357         struct block_device *bdev;
 358         struct inode *inode;
 359
 360         inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
 361                         bdev_test, bdev_set, &dev);
 362
 363         if (!inode)
 364                 return NULL;
 365
 366         bdev = &BDEV_I(inode)->bdev;
 367
 368         if (inode->i_state & I_NEW) {
 369                 bdev->bd_contains = NULL;
 370                 bdev->bd_inode = inode;
 371                 bdev->bd_block_size = (1 << inode->i_blkbits);
 372                 bdev->bd_part_count = 0;
 373                 bdev->bd_invalidated = 0;
 374                 inode->i_mode = S_IFBLK;
 375                 inode->i_rdev = dev;
 376                 inode->i_bdev = bdev;
 377                 inode->i_data.a_ops = &def_blk_aops;
 378                 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 379                 inode->i_data.backing_dev_info = &default_backing_dev_info;
 380                 spin_lock(&bdev_lock);
 381                 list_add(&bdev->bd_list, &all_bdevs);
 382                 spin_unlock(&bdev_lock);
 383                 unlock_new_inode(inode);
 384         }
 385         return bdev;
 386 }
 387
 388 EXPORT_SYMBOL(bdget);
 389
 390 long nr_blockdev_pages(void)
 391 {
 392         struct list_head *p;
 393         long ret = 0;
 394         spin_lock(&bdev_lock);
 395         list_for_each(p, &all_bdevs) {
 396                 struct block_device *bdev;
 397                 bdev = list_entry(p, struct block_device, bd_list);
 398                 ret += bdev->bd_inode->i_mapping->nrpages;
 399         }
 400         spin_unlock(&bdev_lock);
 401         return ret;
 402 }
 403
 404 void bdput(struct block_device *bdev)
 405 {
 406         iput(bdev->bd_inode);
 407 }
 408
 409 EXPORT_SYMBOL(bdput);
 410
 411 static struct block_device *bd_acquire(struct inode *inode)
 412 {
 413         struct block_device *bdev;
 414         spin_lock(&bdev_lock);
 415         bdev = inode->i_bdev;
 416         if (bdev && igrab(bdev->bd_inode)) {
 417                 spin_unlock(&bdev_lock);
 418                 return bdev;
 419         }
 420         spin_unlock(&bdev_lock);
 421         bdev = bdget(inode->i_rdev);
 422         if (bdev) {
 423                 spin_lock(&bdev_lock);
 424                 if (inode->i_bdev)
 425                         __bd_forget(inode);
 426                 inode->i_bdev = bdev;
 427                 inode->i_mapping = bdev->bd_inode->i_mapping;
 428                 list_add(&inode->i_devices, &bdev->bd_inodes);
 429                 spin_unlock(&bdev_lock);
 430         }
 431         return bdev;
 432 }
 433
 434 /* Call when you free inode */
 435
 436 void bd_forget(struct inode *inode)
 437 {
 438         spin_lock(&bdev_lock);
 439         if (inode->i_bdev)
 440                 __bd_forget(inode);
 441         spin_unlock(&bdev_lock);
 442 }
 443
 444 int bd_claim(struct block_device *bdev, void *holder)
 445 {
 446         int res;
 447         spin_lock(&bdev_lock);
 448
 449         /* first decide result */
 450         if (bdev->bd_holder == holder)
 451                 res = 0;         /* already a holder */
 452         else if (bdev->bd_holder != NULL)
 453                 res = -EBUSY;    /* held by someone else */
 454         else if (bdev->bd_contains == bdev)
 455                 res = 0;         /* is a whole device which isn't held */
 456
 457         else if (bdev->bd_contains->bd_holder == bd_claim)
 458                 res = 0;         /* is a partition of a device that is being partitioned */
 459         else if (bdev->bd_contains->bd_holder != NULL)
 460                 res = -EBUSY;    /* is a partition of a held device */
 461         else
 462                 res = 0;         /* is a partition of an un-held device */
 463
 464         /* now impose change */
 465         if (res==0) {
 466                 /* note that for a whole device bd_holders
 467                  * will be incremented twice, and bd_holder will
 468                  * be set to bd_claim before being set to holder
 469                  */
 470                 bdev->bd_contains->bd_holders ++;
 471                 bdev->bd_contains->bd_holder = bd_claim;
 472                 bdev->bd_holders++;
 473                 bdev->bd_holder = holder;
 474         }
 475         spin_unlock(&bdev_lock);
 476         return res;
 477 }
 478
 479 EXPORT_SYMBOL(bd_claim);
 480
 481 void bd_release(struct block_device *bdev)
 482 {
 483         spin_lock(&bdev_lock);
 484         if (!--bdev->bd_contains->bd_holders)
 485                 bdev->bd_contains->bd_holder = NULL;
 486         if (!--bdev->bd_holders)
 487                 bdev->bd_holder = NULL;
 488         spin_unlock(&bdev_lock);
 489 }
 490
 491 EXPORT_SYMBOL(bd_release);
 492
 493 /*
 494  * Tries to open block device by device number.  Use it ONLY if you
 495  * really do not have anything better - i.e. when you are behind a
 496  * truly sucky interface and all you are given is a device number.  _Never_
 497  * to be used for internal purposes.  If you ever need it - reconsider
 498  * your API.
 499  */
 500 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
 501 {
 502         struct block_device *bdev = bdget(dev);
 503         int err = -ENOMEM;
 504         int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
 505         if (bdev)
 506                 err = blkdev_get(bdev, mode, flags);
 507         return err ? ERR_PTR(err) : bdev;
 508 }
 509
 510 EXPORT_SYMBOL(open_by_devnum);
 511
 512 /*
 513  * This routine checks whether a removable media has been changed,
 514  * and invalidates all buffer-cache-entries in that case. This
 515  * is a relatively slow routine, so we have to try to minimize using
 516  * it. Thus it is called only upon a 'mount' or 'open'. This
 517  * is the best way of combining speed and utility, I think.
 518  * People changing diskettes in the middle of an operation deserve
 519  * to lose :-)
 520  */
 521 int check_disk_change(struct block_device *bdev)
 522 {
 523         struct gendisk *disk = bdev->bd_disk;
 524         struct block_device_operations * bdops = disk->fops;
 525
 526         if (!bdops->media_changed)
 527                 return 0;
 528         if (!bdops->media_changed(bdev->bd_disk))
 529                 return 0;
 530
 531         if (__invalidate_device(bdev))
 532                 printk("VFS: busy inodes on changed media.\n");
 533
 534         if (bdops->revalidate_disk)
 535                 bdops->revalidate_disk(bdev->bd_disk);
 536         if (bdev->bd_disk->minors > 1)
 537                 bdev->bd_invalidated = 1;
 538         return 1;
 539 }
 540
 541 EXPORT_SYMBOL(check_disk_change);
 542
 543 void bd_set_size(struct block_device *bdev, loff_t size)
 544 {
 545         unsigned bsize = bdev_hardsect_size(bdev);
 546
 547         bdev->bd_inode->i_size = size;
 548         while (bsize < PAGE_CACHE_SIZE) {
 549                 if (size & bsize)
 550                         break;
 551                 bsize <<= 1;
 552         }
 553         bdev->bd_block_size = bsize;
 554         bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 555 }
 556 EXPORT_SYMBOL(bd_set_size);
 557
 558 static int do_open(struct block_device *bdev, struct file *file)
 559 {
 560         struct module *owner = NULL;
 561         struct gendisk *disk;
 562         int ret = -ENXIO;
 563         int part;
 564
 565         file->f_mapping = bdev->bd_inode->i_mapping;
 566         lock_kernel();
 567         disk = get_gendisk(bdev->bd_dev, &part);
 568         if (!disk) {
 569                 unlock_kernel();
 570                 bdput(bdev);
 571                 return ret;
 572         }
 573         owner = disk->fops->owner;
 574
 575         mutex_lock(&bdev->bd_mutex);
 576         if (!bdev->bd_openers) {
 577                 bdev->bd_disk = disk;
 578                 bdev->bd_contains = bdev;
 579                 if (!part) {
 580                         struct backing_dev_info *bdi;
 581                         if (disk->fops->open) {
 582                                 ret = disk->fops->open(bdev->bd_inode, file);
 583                                 if (ret)
 584                                         goto out_first;
 585                         }
 586                         if (!bdev->bd_openers) {
 587                                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
 588                                 bdi = blk_get_backing_dev_info(bdev);
 589                                 if (bdi == NULL)
 590                                         bdi = &default_backing_dev_info;
 591                                 bdev->bd_inode->i_data.backing_dev_info = bdi;
 592                         }
 593                         if (bdev->bd_invalidated)
 594                                 rescan_partitions(disk, bdev);
 595                 } else {
 596                         struct hd_struct *p;
 597                         struct block_device *whole;
 598                         whole = bdget_disk(disk, 0);
 599                         ret = -ENOMEM;
 600                         if (!whole)
 601                                 goto out_first;
 602                         ret = blkdev_get(whole, file->f_mode, file->f_flags);
 603                         if (ret)
 604                                 goto out_first;
 605                         bdev->bd_contains = whole;
 606                         mutex_lock(&whole->bd_mutex);
 607                         whole->bd_part_count++;
 608                         p = disk->part[part - 1];
 609                         bdev->bd_inode->i_data.backing_dev_info =
 610                            whole->bd_inode->i_data.backing_dev_info;
 611                         if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
 612                                 whole->bd_part_count--;
 613                                 mutex_unlock(&whole->bd_mutex);
 614                                 ret = -ENXIO;
 615                                 goto out_first;
 616                         }
 617                         kobject_get(&p->kobj);
 618                         bdev->bd_part = p;
 619                         bd_set_size(bdev, (loff_t) p->nr_sects << 9);
 620                         mutex_unlock(&whole->bd_mutex);
 621                 }
 622         } else {
 623                 put_disk(disk);
 624                 module_put(owner);
 625                 if (bdev->bd_contains == bdev) {
 626                         if (bdev->bd_disk->fops->open) {
 627                                 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
 628                                 if (ret)
 629                                         goto out;
 630                         }
 631                         if (bdev->bd_invalidated)
 632                                 rescan_partitions(bdev->bd_disk, bdev);
 633                 } else {
 634                         mutex_lock(&bdev->bd_contains->bd_mutex);
 635                         bdev->bd_contains->bd_part_count++;
 636                         mutex_unlock(&bdev->bd_contains->bd_mutex);
 637                 }
 638         }
 639         bdev->bd_openers++;
 640         mutex_unlock(&bdev->bd_mutex);
 641         unlock_kernel();
 642         return 0;
 643
 644 out_first:
 645         bdev->bd_disk = NULL;
 646         bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 647         if (bdev != bdev->bd_contains)
 648                 blkdev_put(bdev->bd_contains);
 649         bdev->bd_contains = NULL;
 650         put_disk(disk);
 651         module_put(owner);
 652 out:
 653         mutex_unlock(&bdev->bd_mutex);
 654         unlock_kernel();
 655         if (ret)
 656                 bdput(bdev);
 657         return ret;
 658 }
 659
 660 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
 661 {
 662         /*
 663          * This crockload is due to bad choice of ->open() type.
 664          * It will go away.
 665          * For now, block device ->open() routine must _not_
 666          * examine anything in 'inode' argument except ->i_rdev.
 667          */
 668         struct file fake_file = {};
 669         struct dentry fake_dentry = {};
 670         fake_file.f_mode = mode;
 671         fake_file.f_flags = flags;
 672         fake_file.f_dentry = &fake_dentry;
 673         fake_dentry.d_inode = bdev->bd_inode;
 674
 675         return do_open(bdev, &fake_file);
 676 }
 677
 678 EXPORT_SYMBOL(blkdev_get);
 679
 680 static int blkdev_open(struct inode * inode, struct file * filp)
 681 {
 682         struct block_device *bdev;
 683         int res;
 684
 685         /*
 686          * Preserve backwards compatibility and allow large file access
 687          * even if userspace doesn't ask for it explicitly. Some mkfs
 688          * binary needs it. We might want to drop this workaround
 689          * during an unstable branch.
 690          */
 691         filp->f_flags |= O_LARGEFILE;
 692
 693         bdev = bd_acquire(inode);
 694
 695         res = do_open(bdev, filp);
 696         if (res)
 697                 return res;
 698
 699         if (!(filp->f_flags & O_EXCL) )
 700                 return 0;
 701
 702         if (!(res = bd_claim(bdev, filp)))
 703                 return 0;
 704
 705         blkdev_put(bdev);
 706         return res;
 707 }
 708
 709 int blkdev_put(struct block_device *bdev)
 710 {
 711         int ret = 0;
 712         struct inode *bd_inode = bdev->bd_inode;
 713         struct gendisk *disk = bdev->bd_disk;
 714
 715         mutex_lock(&bdev->bd_mutex);
 716         lock_kernel();
 717         if (!--bdev->bd_openers) {
 718                 sync_blockdev(bdev);
 719                 kill_bdev(bdev);
 720         }
 721         if (bdev->bd_contains == bdev) {
 722                 if (disk->fops->release)
 723                         ret = disk->fops->release(bd_inode, NULL);
 724         } else {
 725                 mutex_lock(&bdev->bd_contains->bd_mutex);
 726                 bdev->bd_contains->bd_part_count--;
 727                 mutex_unlock(&bdev->bd_contains->bd_mutex);
 728         }
 729         if (!bdev->bd_openers) {
 730                 struct module *owner = disk->fops->owner;
 731
 732                 put_disk(disk);
 733                 module_put(owner);
 734
 735                 if (bdev->bd_contains != bdev) {
 736                         kobject_put(&bdev->bd_part->kobj);
 737                         bdev->bd_part = NULL;
 738                 }
 739                 bdev->bd_disk = NULL;
 740                 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
 741                 if (bdev != bdev->bd_contains) {
 742                         blkdev_put(bdev->bd_contains);
 743                 }
 744                 bdev->bd_contains = NULL;
 745         }
 746         unlock_kernel();
 747         mutex_unlock(&bdev->bd_mutex);
 748         bdput(bdev);
 749         return ret;
 750 }
 751
 752 EXPORT_SYMBOL(blkdev_put);
 753
 754 static int blkdev_close(struct inode * inode, struct file * filp)
 755 {
 756         struct block_device *bdev = I_BDEV(filp->f_mapping->host);
 757         if (bdev->bd_holder == filp)
 758                 bd_release(bdev);
 759         return blkdev_put(bdev);
 760 }
 761
 762 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
 763                                    size_t count, loff_t *ppos)
 764 {
 765         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 766
 767         return generic_file_write_nolock(file, &local_iov, 1, ppos);
 768 }
 769
 770 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
 771                                    size_t count, loff_t pos)
 772 {
 773         struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
 774
 775         return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
 776 }
 777
 778 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 779 {
 780         return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
 781 }
 782
 783 struct address_space_operations def_blk_aops = {
 784         .readpage       = blkdev_readpage,
 785         .writepage      = blkdev_writepage,
 786         .sync_page      = block_sync_page,
 787         .prepare_write  = blkdev_prepare_write,
 788         .commit_write   = blkdev_commit_write,
 789         .writepages     = generic_writepages,
 790         .direct_IO      = blkdev_direct_IO,
 791 };
 792
 793 struct file_operations def_blk_fops = {
 794         .open           = blkdev_open,
 795         .release        = blkdev_close,
 796         .llseek         = block_llseek,
 797         .read           = generic_file_read,
 798         .write          = blkdev_file_write,
 799         .aio_read       = generic_file_aio_read,
 800         .aio_write      = blkdev_file_aio_write,
 801         .mmap           = generic_file_mmap,
 802         .fsync          = block_fsync,
 803         .unlocked_ioctl = block_ioctl,
 804 #ifdef CONFIG_COMPAT
 805         .compat_ioctl   = compat_blkdev_ioctl,
 806 #endif
 807         .readv          = generic_file_readv,
 808         .writev         = generic_file_write_nolock,
 809         .sendfile       = generic_file_sendfile,
 810 };
 811
 812 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
 813 {
 814         int res;
 815         mm_segment_t old_fs = get_fs();
 816         set_fs(KERNEL_DS);
 817         res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
 818         set_fs(old_fs);
 819         return res;
 820 }
 821
 822 EXPORT_SYMBOL(ioctl_by_bdev);
 823
 824 /**
 825  * lookup_bdev  - lookup a struct block_device by name
 826  *
 827  * @path:       special file representing the block device
 828  *
 829  * Get a reference to the blockdevice at @path in the current
 830  * namespace if possible and return it.  Return ERR_PTR(error)
 831  * otherwise.
 832  */
 833 struct block_device *lookup_bdev(const char *path)
 834 {
 835         struct block_device *bdev;
 836         struct inode *inode;
 837         struct nameidata nd;
 838         int error;
 839
 840         if (!path || !*path)
 841                 return ERR_PTR(-EINVAL);
 842
 843         error = path_lookup(path, LOOKUP_FOLLOW, &nd);
 844         if (error)
 845                 return ERR_PTR(error);
 846
 847         inode = nd.dentry->d_inode;
 848         error = -ENOTBLK;
 849         if (!S_ISBLK(inode->i_mode))
 850                 goto fail;
 851         error = -EACCES;
 852         if (nd.mnt->mnt_flags & MNT_NODEV)
 853                 goto fail;
 854         error = -ENOMEM;
 855         bdev = bd_acquire(inode);
 856         if (!bdev)
 857                 goto fail;
 858 out:
 859         path_release(&nd);
 860         return bdev;
 861 fail:
 862         bdev = ERR_PTR(error);
 863         goto out;
 864 }
 865
 866 /**
 867  * open_bdev_excl  -  open a block device by name and set it up for use
 868  *
 869  * @path:       special file representing the block device
 870  * @flags:      %MS_RDONLY for opening read-only
 871  * @holder:     owner for exclusion
 872  *
 873  * Open the blockdevice described by the special file at @path, claim it
 874  * for the @holder.
 875  */
 876 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
 877 {
 878         struct block_device *bdev;
 879         mode_t mode = FMODE_READ;
 880         int error = 0;
 881
 882         bdev = lookup_bdev(path);
 883         if (IS_ERR(bdev))
 884                 return bdev;
 885
 886         if (!(flags & MS_RDONLY))
 887                 mode |= FMODE_WRITE;
 888         error = blkdev_get(bdev, mode, 0);
 889         if (error)
 890                 return ERR_PTR(error);
 891         error = -EACCES;
 892         if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
 893                 goto blkdev_put;
 894         error = bd_claim(bdev, holder);
 895         if (error)
 896                 goto blkdev_put;
 897
 898         return bdev;
 899
 900 blkdev_put:
 901         blkdev_put(bdev);
 902         return ERR_PTR(error);
 903 }
 904
 905 EXPORT_SYMBOL(open_bdev_excl);
 906
 907 /**
 908  * close_bdev_excl  -  release a blockdevice openen by open_bdev_excl()
 909  *
 910  * @bdev:       blockdevice to close
 911  *
 912  * This is the counterpart to open_bdev_excl().
 913  */
 914 void close_bdev_excl(struct block_device *bdev)
 915 {
 916         bd_release(bdev);
 917         blkdev_put(bdev);
 918 }
 919
 920 EXPORT_SYMBOL(close_bdev_excl);