linux/fs/block_dev.c
<<
>>
Prefs
   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/init.h>
   9#include <linux/mm.h>
  10#include <linux/fcntl.h>
  11#include <linux/slab.h>
  12#include <linux/kmod.h>
  13#include <linux/major.h>
  14#include <linux/device_cgroup.h>
  15#include <linux/highmem.h>
  16#include <linux/blkdev.h>
  17#include <linux/module.h>
  18#include <linux/blkpg.h>
  19#include <linux/buffer_head.h>
  20#include <linux/pagevec.h>
  21#include <linux/writeback.h>
  22#include <linux/mpage.h>
  23#include <linux/mount.h>
  24#include <linux/uio.h>
  25#include <linux/namei.h>
  26#include <linux/log2.h>
  27#include <linux/kmemleak.h>
  28#include <asm/uaccess.h>
  29#include "internal.h"
  30
  31struct bdev_inode {
  32        struct block_device bdev;
  33        struct inode vfs_inode;
  34};
  35
  36static const struct address_space_operations def_blk_aops;
  37
  38static inline struct bdev_inode *BDEV_I(struct inode *inode)
  39{
  40        return container_of(inode, struct bdev_inode, vfs_inode);
  41}
  42
  43inline struct block_device *I_BDEV(struct inode *inode)
  44{
  45        return &BDEV_I(inode)->bdev;
  46}
  47
  48EXPORT_SYMBOL(I_BDEV);
  49
  50/*
  51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
  52 * we need to move it onto the dirty list of @dst so that the inode is always
  53 * on the right list.
  54 */
  55static void bdev_inode_switch_bdi(struct inode *inode,
  56                        struct backing_dev_info *dst)
  57{
  58        spin_lock(&inode_wb_list_lock);
  59        spin_lock(&inode->i_lock);
  60        inode->i_data.backing_dev_info = dst;
  61        if (inode->i_state & I_DIRTY)
  62                list_move(&inode->i_wb_list, &dst->wb.b_dirty);
  63        spin_unlock(&inode->i_lock);
  64        spin_unlock(&inode_wb_list_lock);
  65}
  66
  67static sector_t max_block(struct block_device *bdev)
  68{
  69        sector_t retval = ~((sector_t)0);
  70        loff_t sz = i_size_read(bdev->bd_inode);
  71
  72        if (sz) {
  73                unsigned int size = block_size(bdev);
  74                unsigned int sizebits = blksize_bits(size);
  75                retval = (sz >> sizebits);
  76        }
  77        return retval;
  78}
  79
  80/* Kill _all_ buffers and pagecache , dirty or not.. */
  81static void kill_bdev(struct block_device *bdev)
  82{
  83        if (bdev->bd_inode->i_mapping->nrpages == 0)
  84                return;
  85        invalidate_bh_lrus();
  86        truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  87}       
  88
  89int set_blocksize(struct block_device *bdev, int size)
  90{
  91        /* Size must be a power of two, and between 512 and PAGE_SIZE */
  92        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
  93                return -EINVAL;
  94
  95        /* Size cannot be smaller than the size supported by the device */
  96        if (size < bdev_logical_block_size(bdev))
  97                return -EINVAL;
  98
  99        /* Don't change the size if it is same as current */
 100        if (bdev->bd_block_size != size) {
 101                sync_blockdev(bdev);
 102                bdev->bd_block_size = size;
 103                bdev->bd_inode->i_blkbits = blksize_bits(size);
 104                kill_bdev(bdev);
 105        }
 106        return 0;
 107}
 108
 109EXPORT_SYMBOL(set_blocksize);
 110
 111int sb_set_blocksize(struct super_block *sb, int size)
 112{
 113        if (set_blocksize(sb->s_bdev, size))
 114                return 0;
 115        /* If we get here, we know size is power of two
 116         * and it's value is between 512 and PAGE_SIZE */
 117        sb->s_blocksize = size;
 118        sb->s_blocksize_bits = blksize_bits(size);
 119        return sb->s_blocksize;
 120}
 121
 122EXPORT_SYMBOL(sb_set_blocksize);
 123
 124int sb_min_blocksize(struct super_block *sb, int size)
 125{
 126        int minsize = bdev_logical_block_size(sb->s_bdev);
 127        if (size < minsize)
 128                size = minsize;
 129        return sb_set_blocksize(sb, size);
 130}
 131
 132EXPORT_SYMBOL(sb_min_blocksize);
 133
 134static int
 135blkdev_get_block(struct inode *inode, sector_t iblock,
 136                struct buffer_head *bh, int create)
 137{
 138        if (iblock >= max_block(I_BDEV(inode))) {
 139                if (create)
 140                        return -EIO;
 141
 142                /*
 143                 * for reads, we're just trying to fill a partial page.
 144                 * return a hole, they will have to call get_block again
 145                 * before they can fill it, and they will get -EIO at that
 146                 * time
 147                 */
 148                return 0;
 149        }
 150        bh->b_bdev = I_BDEV(inode);
 151        bh->b_blocknr = iblock;
 152        set_buffer_mapped(bh);
 153        return 0;
 154}
 155
 156static int
 157blkdev_get_blocks(struct inode *inode, sector_t iblock,
 158                struct buffer_head *bh, int create)
 159{
 160        sector_t end_block = max_block(I_BDEV(inode));
 161        unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
 162
 163        if ((iblock + max_blocks) > end_block) {
 164                max_blocks = end_block - iblock;
 165                if ((long)max_blocks <= 0) {
 166                        if (create)
 167                                return -EIO;    /* write fully beyond EOF */
 168                        /*
 169                         * It is a read which is fully beyond EOF.  We return
 170                         * a !buffer_mapped buffer
 171                         */
 172                        max_blocks = 0;
 173                }
 174        }
 175
 176        bh->b_bdev = I_BDEV(inode);
 177        bh->b_blocknr = iblock;
 178        bh->b_size = max_blocks << inode->i_blkbits;
 179        if (max_blocks)
 180                set_buffer_mapped(bh);
 181        return 0;
 182}
 183
 184static ssize_t
 185blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 186                        loff_t offset, unsigned long nr_segs)
 187{
 188        struct file *file = iocb->ki_filp;
 189        struct inode *inode = file->f_mapping->host;
 190
 191        return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 192                                    nr_segs, blkdev_get_blocks, NULL, NULL, 0);
 193}
 194
 195int __sync_blockdev(struct block_device *bdev, int wait)
 196{
 197        if (!bdev)
 198                return 0;
 199        if (!wait)
 200                return filemap_flush(bdev->bd_inode->i_mapping);
 201        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 202}
 203
 204/*
 205 * Write out and wait upon all the dirty data associated with a block
 206 * device via its mapping.  Does not take the superblock lock.
 207 */
 208int sync_blockdev(struct block_device *bdev)
 209{
 210        return __sync_blockdev(bdev, 1);
 211}
 212EXPORT_SYMBOL(sync_blockdev);
 213
 214/*
 215 * Write out and wait upon all dirty data associated with this
 216 * device.   Filesystem data as well as the underlying block
 217 * device.  Takes the superblock lock.
 218 */
 219int fsync_bdev(struct block_device *bdev)
 220{
 221        struct super_block *sb = get_super(bdev);
 222        if (sb) {
 223                int res = sync_filesystem(sb);
 224                drop_super(sb);
 225                return res;
 226        }
 227        return sync_blockdev(bdev);
 228}
 229EXPORT_SYMBOL(fsync_bdev);
 230
 231/**
 232 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 233 * @bdev:       blockdevice to lock
 234 *
 235 * If a superblock is found on this device, we take the s_umount semaphore
 236 * on it to make sure nobody unmounts until the snapshot creation is done.
 237 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 238 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 239 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 240 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 241 * actually.
 242 */
 243struct super_block *freeze_bdev(struct block_device *bdev)
 244{
 245        struct super_block *sb;
 246        int error = 0;
 247
 248        mutex_lock(&bdev->bd_fsfreeze_mutex);
 249        if (++bdev->bd_fsfreeze_count > 1) {
 250                /*
 251                 * We don't even need to grab a reference - the first call
 252                 * to freeze_bdev grab an active reference and only the last
 253                 * thaw_bdev drops it.
 254                 */
 255                sb = get_super(bdev);
 256                drop_super(sb);
 257                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 258                return sb;
 259        }
 260
 261        sb = get_active_super(bdev);
 262        if (!sb)
 263                goto out;
 264        error = freeze_super(sb);
 265        if (error) {
 266                deactivate_super(sb);
 267                bdev->bd_fsfreeze_count--;
 268                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 269                return ERR_PTR(error);
 270        }
 271        deactivate_super(sb);
 272 out:
 273        sync_blockdev(bdev);
 274        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 275        return sb;      /* thaw_bdev releases s->s_umount */
 276}
 277EXPORT_SYMBOL(freeze_bdev);
 278
 279/**
 280 * thaw_bdev  -- unlock filesystem
 281 * @bdev:       blockdevice to unlock
 282 * @sb:         associated superblock
 283 *
 284 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 285 */
 286int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 287{
 288        int error = -EINVAL;
 289
 290        mutex_lock(&bdev->bd_fsfreeze_mutex);
 291        if (!bdev->bd_fsfreeze_count)
 292                goto out;
 293
 294        error = 0;
 295        if (--bdev->bd_fsfreeze_count > 0)
 296                goto out;
 297
 298        if (!sb)
 299                goto out;
 300
 301        error = thaw_super(sb);
 302        if (error) {
 303                bdev->bd_fsfreeze_count++;
 304                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 305                return error;
 306        }
 307out:
 308        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 309        return 0;
 310}
 311EXPORT_SYMBOL(thaw_bdev);
 312
 313static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 314{
 315        return block_write_full_page(page, blkdev_get_block, wbc);
 316}
 317
 318static int blkdev_readpage(struct file * file, struct page * page)
 319{
 320        return block_read_full_page(page, blkdev_get_block);
 321}
 322
 323static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 324                        loff_t pos, unsigned len, unsigned flags,
 325                        struct page **pagep, void **fsdata)
 326{
 327        return block_write_begin(mapping, pos, len, flags, pagep,
 328                                 blkdev_get_block);
 329}
 330
 331static int blkdev_write_end(struct file *file, struct address_space *mapping,
 332                        loff_t pos, unsigned len, unsigned copied,
 333                        struct page *page, void *fsdata)
 334{
 335        int ret;
 336        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 337
 338        unlock_page(page);
 339        page_cache_release(page);
 340
 341        return ret;
 342}
 343
 344/*
 345 * private llseek:
 346 * for a block special file file->f_path.dentry->d_inode->i_size is zero
 347 * so we compute the size by hand (just as in block_read/write above)
 348 */
 349static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 350{
 351        struct inode *bd_inode = file->f_mapping->host;
 352        loff_t size;
 353        loff_t retval;
 354
 355        mutex_lock(&bd_inode->i_mutex);
 356        size = i_size_read(bd_inode);
 357
 358        switch (origin) {
 359                case 2:
 360                        offset += size;
 361                        break;
 362                case 1:
 363                        offset += file->f_pos;
 364        }
 365        retval = -EINVAL;
 366        if (offset >= 0 && offset <= size) {
 367                if (offset != file->f_pos) {
 368                        file->f_pos = offset;
 369                }
 370                retval = offset;
 371        }
 372        mutex_unlock(&bd_inode->i_mutex);
 373        return retval;
 374}
 375        
 376int blkdev_fsync(struct file *filp, int datasync)
 377{
 378        struct inode *bd_inode = filp->f_mapping->host;
 379        struct block_device *bdev = I_BDEV(bd_inode);
 380        int error;
 381
 382        /*
 383         * There is no need to serialise calls to blkdev_issue_flush with
 384         * i_mutex and doing so causes performance issues with concurrent
 385         * O_SYNC writers to a block device.
 386         */
 387        mutex_unlock(&bd_inode->i_mutex);
 388
 389        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 390        if (error == -EOPNOTSUPP)
 391                error = 0;
 392
 393        mutex_lock(&bd_inode->i_mutex);
 394
 395        return error;
 396}
 397EXPORT_SYMBOL(blkdev_fsync);
 398
 399/*
 400 * pseudo-fs
 401 */
 402
 403static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 404static struct kmem_cache * bdev_cachep __read_mostly;
 405
 406static struct inode *bdev_alloc_inode(struct super_block *sb)
 407{
 408        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 409        if (!ei)
 410                return NULL;
 411        return &ei->vfs_inode;
 412}
 413
 414static void bdev_i_callback(struct rcu_head *head)
 415{
 416        struct inode *inode = container_of(head, struct inode, i_rcu);
 417        struct bdev_inode *bdi = BDEV_I(inode);
 418
 419        INIT_LIST_HEAD(&inode->i_dentry);
 420        kmem_cache_free(bdev_cachep, bdi);
 421}
 422
 423static void bdev_destroy_inode(struct inode *inode)
 424{
 425        call_rcu(&inode->i_rcu, bdev_i_callback);
 426}
 427
 428static void init_once(void *foo)
 429{
 430        struct bdev_inode *ei = (struct bdev_inode *) foo;
 431        struct block_device *bdev = &ei->bdev;
 432
 433        memset(bdev, 0, sizeof(*bdev));
 434        mutex_init(&bdev->bd_mutex);
 435        INIT_LIST_HEAD(&bdev->bd_inodes);
 436        INIT_LIST_HEAD(&bdev->bd_list);
 437#ifdef CONFIG_SYSFS
 438        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 439#endif
 440        inode_init_once(&ei->vfs_inode);
 441        /* Initialize mutex for freeze. */
 442        mutex_init(&bdev->bd_fsfreeze_mutex);
 443}
 444
 445static inline void __bd_forget(struct inode *inode)
 446{
 447        list_del_init(&inode->i_devices);
 448        inode->i_bdev = NULL;
 449        inode->i_mapping = &inode->i_data;
 450}
 451
 452static void bdev_evict_inode(struct inode *inode)
 453{
 454        struct block_device *bdev = &BDEV_I(inode)->bdev;
 455        struct list_head *p;
 456        truncate_inode_pages(&inode->i_data, 0);
 457        invalidate_inode_buffers(inode); /* is it needed here? */
 458        end_writeback(inode);
 459        spin_lock(&bdev_lock);
 460        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 461                __bd_forget(list_entry(p, struct inode, i_devices));
 462        }
 463        list_del_init(&bdev->bd_list);
 464        spin_unlock(&bdev_lock);
 465}
 466
 467static const struct super_operations bdev_sops = {
 468        .statfs = simple_statfs,
 469        .alloc_inode = bdev_alloc_inode,
 470        .destroy_inode = bdev_destroy_inode,
 471        .drop_inode = generic_delete_inode,
 472        .evict_inode = bdev_evict_inode,
 473};
 474
 475static struct dentry *bd_mount(struct file_system_type *fs_type,
 476        int flags, const char *dev_name, void *data)
 477{
 478        return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
 479}
 480
 481static struct file_system_type bd_type = {
 482        .name           = "bdev",
 483        .mount          = bd_mount,
 484        .kill_sb        = kill_anon_super,
 485};
 486
 487struct super_block *blockdev_superblock __read_mostly;
 488
 489void __init bdev_cache_init(void)
 490{
 491        int err;
 492        struct vfsmount *bd_mnt;
 493
 494        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 495                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 496                                SLAB_MEM_SPREAD|SLAB_PANIC),
 497                        init_once);
 498        err = register_filesystem(&bd_type);
 499        if (err)
 500                panic("Cannot register bdev pseudo-fs");
 501        bd_mnt = kern_mount(&bd_type);
 502        if (IS_ERR(bd_mnt))
 503                panic("Cannot create bdev pseudo-fs");
 504        /*
 505         * This vfsmount structure is only used to obtain the
 506         * blockdev_superblock, so tell kmemleak not to report it.
 507         */
 508        kmemleak_not_leak(bd_mnt);
 509        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 510}
 511
 512/*
 513 * Most likely _very_ bad one - but then it's hardly critical for small
 514 * /dev and can be fixed when somebody will need really large one.
 515 * Keep in mind that it will be fed through icache hash function too.
 516 */
 517static inline unsigned long hash(dev_t dev)
 518{
 519        return MAJOR(dev)+MINOR(dev);
 520}
 521
 522static int bdev_test(struct inode *inode, void *data)
 523{
 524        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 525}
 526
 527static int bdev_set(struct inode *inode, void *data)
 528{
 529        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 530        return 0;
 531}
 532
 533static LIST_HEAD(all_bdevs);
 534
 535struct block_device *bdget(dev_t dev)
 536{
 537        struct block_device *bdev;
 538        struct inode *inode;
 539
 540        inode = iget5_locked(blockdev_superblock, hash(dev),
 541                        bdev_test, bdev_set, &dev);
 542
 543        if (!inode)
 544                return NULL;
 545
 546        bdev = &BDEV_I(inode)->bdev;
 547
 548        if (inode->i_state & I_NEW) {
 549                bdev->bd_contains = NULL;
 550                bdev->bd_inode = inode;
 551                bdev->bd_block_size = (1 << inode->i_blkbits);
 552                bdev->bd_part_count = 0;
 553                bdev->bd_invalidated = 0;
 554                inode->i_mode = S_IFBLK;
 555                inode->i_rdev = dev;
 556                inode->i_bdev = bdev;
 557                inode->i_data.a_ops = &def_blk_aops;
 558                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 559                inode->i_data.backing_dev_info = &default_backing_dev_info;
 560                spin_lock(&bdev_lock);
 561                list_add(&bdev->bd_list, &all_bdevs);
 562                spin_unlock(&bdev_lock);
 563                unlock_new_inode(inode);
 564        }
 565        return bdev;
 566}
 567
 568EXPORT_SYMBOL(bdget);
 569
 570/**
 571 * bdgrab -- Grab a reference to an already referenced block device
 572 * @bdev:       Block device to grab a reference to.
 573 */
 574struct block_device *bdgrab(struct block_device *bdev)
 575{
 576        ihold(bdev->bd_inode);
 577        return bdev;
 578}
 579
 580long nr_blockdev_pages(void)
 581{
 582        struct block_device *bdev;
 583        long ret = 0;
 584        spin_lock(&bdev_lock);
 585        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 586                ret += bdev->bd_inode->i_mapping->nrpages;
 587        }
 588        spin_unlock(&bdev_lock);
 589        return ret;
 590}
 591
 592void bdput(struct block_device *bdev)
 593{
 594        iput(bdev->bd_inode);
 595}
 596
 597EXPORT_SYMBOL(bdput);
 598 
 599static struct block_device *bd_acquire(struct inode *inode)
 600{
 601        struct block_device *bdev;
 602
 603        spin_lock(&bdev_lock);
 604        bdev = inode->i_bdev;
 605        if (bdev) {
 606                ihold(bdev->bd_inode);
 607                spin_unlock(&bdev_lock);
 608                return bdev;
 609        }
 610        spin_unlock(&bdev_lock);
 611
 612        bdev = bdget(inode->i_rdev);
 613        if (bdev) {
 614                spin_lock(&bdev_lock);
 615                if (!inode->i_bdev) {
 616                        /*
 617                         * We take an additional reference to bd_inode,
 618                         * and it's released in clear_inode() of inode.
 619                         * So, we can access it via ->i_mapping always
 620                         * without igrab().
 621                         */
 622                        ihold(bdev->bd_inode);
 623                        inode->i_bdev = bdev;
 624                        inode->i_mapping = bdev->bd_inode->i_mapping;
 625                        list_add(&inode->i_devices, &bdev->bd_inodes);
 626                }
 627                spin_unlock(&bdev_lock);
 628        }
 629        return bdev;
 630}
 631
 632/* Call when you free inode */
 633
 634void bd_forget(struct inode *inode)
 635{
 636        struct block_device *bdev = NULL;
 637
 638        spin_lock(&bdev_lock);
 639        if (inode->i_bdev) {
 640                if (!sb_is_blkdev_sb(inode->i_sb))
 641                        bdev = inode->i_bdev;
 642                __bd_forget(inode);
 643        }
 644        spin_unlock(&bdev_lock);
 645
 646        if (bdev)
 647                iput(bdev->bd_inode);
 648}
 649
 650/**
 651 * bd_may_claim - test whether a block device can be claimed
 652 * @bdev: block device of interest
 653 * @whole: whole block device containing @bdev, may equal @bdev
 654 * @holder: holder trying to claim @bdev
 655 *
 656 * Test whether @bdev can be claimed by @holder.
 657 *
 658 * CONTEXT:
 659 * spin_lock(&bdev_lock).
 660 *
 661 * RETURNS:
 662 * %true if @bdev can be claimed, %false otherwise.
 663 */
 664static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 665                         void *holder)
 666{
 667        if (bdev->bd_holder == holder)
 668                return true;     /* already a holder */
 669        else if (bdev->bd_holder != NULL)
 670                return false;    /* held by someone else */
 671        else if (bdev->bd_contains == bdev)
 672                return true;     /* is a whole device which isn't held */
 673
 674        else if (whole->bd_holder == bd_may_claim)
 675                return true;     /* is a partition of a device that is being partitioned */
 676        else if (whole->bd_holder != NULL)
 677                return false;    /* is a partition of a held device */
 678        else
 679                return true;     /* is a partition of an un-held device */
 680}
 681
 682/**
 683 * bd_prepare_to_claim - prepare to claim a block device
 684 * @bdev: block device of interest
 685 * @whole: the whole device containing @bdev, may equal @bdev
 686 * @holder: holder trying to claim @bdev
 687 *
 688 * Prepare to claim @bdev.  This function fails if @bdev is already
 689 * claimed by another holder and waits if another claiming is in
 690 * progress.  This function doesn't actually claim.  On successful
 691 * return, the caller has ownership of bd_claiming and bd_holder[s].
 692 *
 693 * CONTEXT:
 694 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 695 * it multiple times.
 696 *
 697 * RETURNS:
 698 * 0 if @bdev can be claimed, -EBUSY otherwise.
 699 */
 700static int bd_prepare_to_claim(struct block_device *bdev,
 701                               struct block_device *whole, void *holder)
 702{
 703retry:
 704        /* if someone else claimed, fail */
 705        if (!bd_may_claim(bdev, whole, holder))
 706                return -EBUSY;
 707
 708        /* if claiming is already in progress, wait for it to finish */
 709        if (whole->bd_claiming) {
 710                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 711                DEFINE_WAIT(wait);
 712
 713                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 714                spin_unlock(&bdev_lock);
 715                schedule();
 716                finish_wait(wq, &wait);
 717                spin_lock(&bdev_lock);
 718                goto retry;
 719        }
 720
 721        /* yay, all mine */
 722        return 0;
 723}
 724
 725/**
 726 * bd_start_claiming - start claiming a block device
 727 * @bdev: block device of interest
 728 * @holder: holder trying to claim @bdev
 729 *
 730 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 731 * exclusively and mark that an exclusive open is in progress.  Each
 732 * successful call to this function must be matched with a call to
 733 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 734 * fail).
 735 *
 736 * This function is used to gain exclusive access to the block device
 737 * without actually causing other exclusive open attempts to fail. It
 738 * should be used when the open sequence itself requires exclusive
 739 * access but may subsequently fail.
 740 *
 741 * CONTEXT:
 742 * Might sleep.
 743 *
 744 * RETURNS:
 745 * Pointer to the block device containing @bdev on success, ERR_PTR()
 746 * value on failure.
 747 */
 748static struct block_device *bd_start_claiming(struct block_device *bdev,
 749                                              void *holder)
 750{
 751        struct gendisk *disk;
 752        struct block_device *whole;
 753        int partno, err;
 754
 755        might_sleep();
 756
 757        /*
 758         * @bdev might not have been initialized properly yet, look up
 759         * and grab the outer block device the hard way.
 760         */
 761        disk = get_gendisk(bdev->bd_dev, &partno);
 762        if (!disk)
 763                return ERR_PTR(-ENXIO);
 764
 765        whole = bdget_disk(disk, 0);
 766        module_put(disk->fops->owner);
 767        put_disk(disk);
 768        if (!whole)
 769                return ERR_PTR(-ENOMEM);
 770
 771        /* prepare to claim, if successful, mark claiming in progress */
 772        spin_lock(&bdev_lock);
 773
 774        err = bd_prepare_to_claim(bdev, whole, holder);
 775        if (err == 0) {
 776                whole->bd_claiming = holder;
 777                spin_unlock(&bdev_lock);
 778                return whole;
 779        } else {
 780                spin_unlock(&bdev_lock);
 781                bdput(whole);
 782                return ERR_PTR(err);
 783        }
 784}
 785
 786#ifdef CONFIG_SYSFS
 787struct bd_holder_disk {
 788        struct list_head        list;
 789        struct gendisk          *disk;
 790        int                     refcnt;
 791};
 792
 793static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 794                                                  struct gendisk *disk)
 795{
 796        struct bd_holder_disk *holder;
 797
 798        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 799                if (holder->disk == disk)
 800                        return holder;
 801        return NULL;
 802}
 803
 804static int add_symlink(struct kobject *from, struct kobject *to)
 805{
 806        return sysfs_create_link(from, to, kobject_name(to));
 807}
 808
 809static void del_symlink(struct kobject *from, struct kobject *to)
 810{
 811        sysfs_remove_link(from, kobject_name(to));
 812}
 813
 814/**
 815 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 816 * @bdev: the claimed slave bdev
 817 * @disk: the holding disk
 818 *
 819 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 820 *
 821 * This functions creates the following sysfs symlinks.
 822 *
 823 * - from "slaves" directory of the holder @disk to the claimed @bdev
 824 * - from "holders" directory of the @bdev to the holder @disk
 825 *
 826 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
 827 * passed to bd_link_disk_holder(), then:
 828 *
 829 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
 830 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 831 *
 832 * The caller must have claimed @bdev before calling this function and
 833 * ensure that both @bdev and @disk are valid during the creation and
 834 * lifetime of these symlinks.
 835 *
 836 * CONTEXT:
 837 * Might sleep.
 838 *
 839 * RETURNS:
 840 * 0 on success, -errno on failure.
 841 */
 842int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 843{
 844        struct bd_holder_disk *holder;
 845        int ret = 0;
 846
 847        mutex_lock(&bdev->bd_mutex);
 848
 849        WARN_ON_ONCE(!bdev->bd_holder);
 850
 851        /* FIXME: remove the following once add_disk() handles errors */
 852        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
 853                goto out_unlock;
 854
 855        holder = bd_find_holder_disk(bdev, disk);
 856        if (holder) {
 857                holder->refcnt++;
 858                goto out_unlock;
 859        }
 860
 861        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 862        if (!holder) {
 863                ret = -ENOMEM;
 864                goto out_unlock;
 865        }
 866
 867        INIT_LIST_HEAD(&holder->list);
 868        holder->disk = disk;
 869        holder->refcnt = 1;
 870
 871        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 872        if (ret)
 873                goto out_free;
 874
 875        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
 876        if (ret)
 877                goto out_del;
 878        /*
 879         * bdev could be deleted beneath us which would implicitly destroy
 880         * the holder directory.  Hold on to it.
 881         */
 882        kobject_get(bdev->bd_part->holder_dir);
 883
 884        list_add(&holder->list, &bdev->bd_holder_disks);
 885        goto out_unlock;
 886
 887out_del:
 888        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 889out_free:
 890        kfree(holder);
 891out_unlock:
 892        mutex_unlock(&bdev->bd_mutex);
 893        return ret;
 894}
 895EXPORT_SYMBOL_GPL(bd_link_disk_holder);
 896
 897/**
 898 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
 899 * @bdev: the calimed slave bdev
 900 * @disk: the holding disk
 901 *
 902 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 903 *
 904 * CONTEXT:
 905 * Might sleep.
 906 */
 907void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 908{
 909        struct bd_holder_disk *holder;
 910
 911        mutex_lock(&bdev->bd_mutex);
 912
 913        holder = bd_find_holder_disk(bdev, disk);
 914
 915        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 916                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 917                del_symlink(bdev->bd_part->holder_dir,
 918                            &disk_to_dev(disk)->kobj);
 919                kobject_put(bdev->bd_part->holder_dir);
 920                list_del_init(&holder->list);
 921                kfree(holder);
 922        }
 923
 924        mutex_unlock(&bdev->bd_mutex);
 925}
 926EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
 927#endif
 928
 929/**
 930 * flush_disk - invalidates all buffer-cache entries on a disk
 931 *
 932 * @bdev:      struct block device to be flushed
 933 * @kill_dirty: flag to guide handling of dirty inodes
 934 *
 935 * Invalidates all buffer-cache entries on a disk. It should be called
 936 * when a disk has been changed -- either by a media change or online
 937 * resize.
 938 */
 939static void flush_disk(struct block_device *bdev, bool kill_dirty)
 940{
 941        if (__invalidate_device(bdev, kill_dirty)) {
 942                char name[BDEVNAME_SIZE] = "";
 943
 944                if (bdev->bd_disk)
 945                        disk_name(bdev->bd_disk, 0, name);
 946                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 947                       "resized disk %s\n", name);
 948        }
 949
 950        if (!bdev->bd_disk)
 951                return;
 952        if (disk_partitionable(bdev->bd_disk))
 953                bdev->bd_invalidated = 1;
 954}
 955
 956/**
 957 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 958 * @disk: struct gendisk to check
 959 * @bdev: struct bdev to adjust.
 960 *
 961 * This routine checks to see if the bdev size does not match the disk size
 962 * and adjusts it if it differs.
 963 */
 964void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 965{
 966        loff_t disk_size, bdev_size;
 967
 968        disk_size = (loff_t)get_capacity(disk) << 9;
 969        bdev_size = i_size_read(bdev->bd_inode);
 970        if (disk_size != bdev_size) {
 971                char name[BDEVNAME_SIZE];
 972
 973                disk_name(disk, 0, name);
 974                printk(KERN_INFO
 975                       "%s: detected capacity change from %lld to %lld\n",
 976                       name, bdev_size, disk_size);
 977                i_size_write(bdev->bd_inode, disk_size);
 978                flush_disk(bdev, false);
 979        }
 980}
 981EXPORT_SYMBOL(check_disk_size_change);
 982
 983/**
 984 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 985 * @disk: struct gendisk to be revalidated
 986 *
 987 * This routine is a wrapper for lower-level driver's revalidate_disk
 988 * call-backs.  It is used to do common pre and post operations needed
 989 * for all revalidate_disk operations.
 990 */
 991int revalidate_disk(struct gendisk *disk)
 992{
 993        struct block_device *bdev;
 994        int ret = 0;
 995
 996        if (disk->fops->revalidate_disk)
 997                ret = disk->fops->revalidate_disk(disk);
 998
 999        bdev = bdget_disk(disk, 0);
1000        if (!bdev)
1001                return ret;
1002
1003        mutex_lock(&bdev->bd_mutex);
1004        check_disk_size_change(disk, bdev);
1005        mutex_unlock(&bdev->bd_mutex);
1006        bdput(bdev);
1007        return ret;
1008}
1009EXPORT_SYMBOL(revalidate_disk);
1010
1011/*
1012 * This routine checks whether a removable media has been changed,
1013 * and invalidates all buffer-cache-entries in that case. This
1014 * is a relatively slow routine, so we have to try to minimize using
1015 * it. Thus it is called only upon a 'mount' or 'open'. This
1016 * is the best way of combining speed and utility, I think.
1017 * People changing diskettes in the middle of an operation deserve
1018 * to lose :-)
1019 */
1020int check_disk_change(struct block_device *bdev)
1021{
1022        struct gendisk *disk = bdev->bd_disk;
1023        const struct block_device_operations *bdops = disk->fops;
1024        unsigned int events;
1025
1026        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1027                                   DISK_EVENT_EJECT_REQUEST);
1028        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1029                return 0;
1030
1031        flush_disk(bdev, true);
1032        if (bdops->revalidate_disk)
1033                bdops->revalidate_disk(bdev->bd_disk);
1034        return 1;
1035}
1036
1037EXPORT_SYMBOL(check_disk_change);
1038
1039void bd_set_size(struct block_device *bdev, loff_t size)
1040{
1041        unsigned bsize = bdev_logical_block_size(bdev);
1042
1043        bdev->bd_inode->i_size = size;
1044        while (bsize < PAGE_CACHE_SIZE) {
1045                if (size & bsize)
1046                        break;
1047                bsize <<= 1;
1048        }
1049        bdev->bd_block_size = bsize;
1050        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1051}
1052EXPORT_SYMBOL(bd_set_size);
1053
1054static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1055
1056/*
1057 * bd_mutex locking:
1058 *
1059 *  mutex_lock(part->bd_mutex)
1060 *    mutex_lock_nested(whole->bd_mutex, 1)
1061 */
1062
1063static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1064{
1065        struct gendisk *disk;
1066        int ret;
1067        int partno;
1068        int perm = 0;
1069
1070        if (mode & FMODE_READ)
1071                perm |= MAY_READ;
1072        if (mode & FMODE_WRITE)
1073                perm |= MAY_WRITE;
1074        /*
1075         * hooks: /n/, see "layering violations".
1076         */
1077        if (!for_part) {
1078                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1079                if (ret != 0) {
1080                        bdput(bdev);
1081                        return ret;
1082                }
1083        }
1084
1085 restart:
1086
1087        ret = -ENXIO;
1088        disk = get_gendisk(bdev->bd_dev, &partno);
1089        if (!disk)
1090                goto out;
1091
1092        disk_block_events(disk);
1093        mutex_lock_nested(&bdev->bd_mutex, for_part);
1094        if (!bdev->bd_openers) {
1095                bdev->bd_disk = disk;
1096                bdev->bd_contains = bdev;
1097                if (!partno) {
1098                        struct backing_dev_info *bdi;
1099
1100                        ret = -ENXIO;
1101                        bdev->bd_part = disk_get_part(disk, partno);
1102                        if (!bdev->bd_part)
1103                                goto out_clear;
1104
1105                        ret = 0;
1106                        if (disk->fops->open) {
1107                                ret = disk->fops->open(bdev, mode);
1108                                if (ret == -ERESTARTSYS) {
1109                                        /* Lost a race with 'disk' being
1110                                         * deleted, try again.
1111                                         * See md.c
1112                                         */
1113                                        disk_put_part(bdev->bd_part);
1114                                        bdev->bd_part = NULL;
1115                                        bdev->bd_disk = NULL;
1116                                        mutex_unlock(&bdev->bd_mutex);
1117                                        disk_unblock_events(disk);
1118                                        module_put(disk->fops->owner);
1119                                        put_disk(disk);
1120                                        goto restart;
1121                                }
1122                        }
1123                        /*
1124                         * If the device is invalidated, rescan partition
1125                         * if open succeeded or failed with -ENOMEDIUM.
1126                         * The latter is necessary to prevent ghost
1127                         * partitions on a removed medium.
1128                         */
1129                        if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1130                                rescan_partitions(disk, bdev);
1131                        if (ret)
1132                                goto out_clear;
1133
1134                        if (!bdev->bd_openers) {
1135                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1136                                bdi = blk_get_backing_dev_info(bdev);
1137                                if (bdi == NULL)
1138                                        bdi = &default_backing_dev_info;
1139                                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1140                        }
1141                } else {
1142                        struct block_device *whole;
1143                        whole = bdget_disk(disk, 0);
1144                        ret = -ENOMEM;
1145                        if (!whole)
1146                                goto out_clear;
1147                        BUG_ON(for_part);
1148                        ret = __blkdev_get(whole, mode, 1);
1149                        if (ret)
1150                                goto out_clear;
1151                        bdev->bd_contains = whole;
1152                        bdev_inode_switch_bdi(bdev->bd_inode,
1153                                whole->bd_inode->i_data.backing_dev_info);
1154                        bdev->bd_part = disk_get_part(disk, partno);
1155                        if (!(disk->flags & GENHD_FL_UP) ||
1156                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1157                                ret = -ENXIO;
1158                                goto out_clear;
1159                        }
1160                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1161                }
1162        } else {
1163                if (bdev->bd_contains == bdev) {
1164                        ret = 0;
1165                        if (bdev->bd_disk->fops->open)
1166                                ret = bdev->bd_disk->fops->open(bdev, mode);
1167                        /* the same as first opener case, read comment there */
1168                        if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1169                                rescan_partitions(bdev->bd_disk, bdev);
1170                        if (ret)
1171                                goto out_unlock_bdev;
1172                }
1173                /* only one opener holds refs to the module and disk */
1174                module_put(disk->fops->owner);
1175                put_disk(disk);
1176        }
1177        bdev->bd_openers++;
1178        if (for_part)
1179                bdev->bd_part_count++;
1180        mutex_unlock(&bdev->bd_mutex);
1181        disk_unblock_events(disk);
1182        return 0;
1183
1184 out_clear:
1185        disk_put_part(bdev->bd_part);
1186        bdev->bd_disk = NULL;
1187        bdev->bd_part = NULL;
1188        bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1189        if (bdev != bdev->bd_contains)
1190                __blkdev_put(bdev->bd_contains, mode, 1);
1191        bdev->bd_contains = NULL;
1192 out_unlock_bdev:
1193        mutex_unlock(&bdev->bd_mutex);
1194        disk_unblock_events(disk);
1195        module_put(disk->fops->owner);
1196        put_disk(disk);
1197 out:
1198        bdput(bdev);
1199
1200        return ret;
1201}
1202
1203/**
1204 * blkdev_get - open a block device
1205 * @bdev: block_device to open
1206 * @mode: FMODE_* mask
1207 * @holder: exclusive holder identifier
1208 *
1209 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1210 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1211 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1212 *
1213 * On success, the reference count of @bdev is unchanged.  On failure,
1214 * @bdev is put.
1215 *
1216 * CONTEXT:
1217 * Might sleep.
1218 *
1219 * RETURNS:
1220 * 0 on success, -errno on failure.
1221 */
1222int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1223{
1224        struct block_device *whole = NULL;
1225        int res;
1226
1227        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1228
1229        if ((mode & FMODE_EXCL) && holder) {
1230                whole = bd_start_claiming(bdev, holder);
1231                if (IS_ERR(whole)) {
1232                        bdput(bdev);
1233                        return PTR_ERR(whole);
1234                }
1235        }
1236
1237        res = __blkdev_get(bdev, mode, 0);
1238
1239        if (whole) {
1240                /* finish claiming */
1241                mutex_lock(&bdev->bd_mutex);
1242                spin_lock(&bdev_lock);
1243
1244                if (!res) {
1245                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1246                        /*
1247                         * Note that for a whole device bd_holders
1248                         * will be incremented twice, and bd_holder
1249                         * will be set to bd_may_claim before being
1250                         * set to holder
1251                         */
1252                        whole->bd_holders++;
1253                        whole->bd_holder = bd_may_claim;
1254                        bdev->bd_holders++;
1255                        bdev->bd_holder = holder;
1256                }
1257
1258                /* tell others that we're done */
1259                BUG_ON(whole->bd_claiming != holder);
1260                whole->bd_claiming = NULL;
1261                wake_up_bit(&whole->bd_claiming, 0);
1262
1263                spin_unlock(&bdev_lock);
1264
1265                /*
1266                 * Block event polling for write claims.  Any write
1267                 * holder makes the write_holder state stick until all
1268                 * are released.  This is good enough and tracking
1269                 * individual writeable reference is too fragile given
1270                 * the way @mode is used in blkdev_get/put().
1271                 */
1272                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
1273                        bdev->bd_write_holder = true;
1274                        disk_block_events(bdev->bd_disk);
1275                }
1276
1277                mutex_unlock(&bdev->bd_mutex);
1278                bdput(whole);
1279        }
1280
1281        return res;
1282}
1283EXPORT_SYMBOL(blkdev_get);
1284
1285/**
1286 * blkdev_get_by_path - open a block device by name
1287 * @path: path to the block device to open
1288 * @mode: FMODE_* mask
1289 * @holder: exclusive holder identifier
1290 *
1291 * Open the blockdevice described by the device file at @path.  @mode
1292 * and @holder are identical to blkdev_get().
1293 *
1294 * On success, the returned block_device has reference count of one.
1295 *
1296 * CONTEXT:
1297 * Might sleep.
1298 *
1299 * RETURNS:
1300 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1301 */
1302struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1303                                        void *holder)
1304{
1305        struct block_device *bdev;
1306        int err;
1307
1308        bdev = lookup_bdev(path);
1309        if (IS_ERR(bdev))
1310                return bdev;
1311
1312        err = blkdev_get(bdev, mode, holder);
1313        if (err)
1314                return ERR_PTR(err);
1315
1316        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1317                blkdev_put(bdev, mode);
1318                return ERR_PTR(-EACCES);
1319        }
1320
1321        return bdev;
1322}
1323EXPORT_SYMBOL(blkdev_get_by_path);
1324
1325/**
1326 * blkdev_get_by_dev - open a block device by device number
1327 * @dev: device number of block device to open
1328 * @mode: FMODE_* mask
1329 * @holder: exclusive holder identifier
1330 *
1331 * Open the blockdevice described by device number @dev.  @mode and
1332 * @holder are identical to blkdev_get().
1333 *
1334 * Use it ONLY if you really do not have anything better - i.e. when
1335 * you are behind a truly sucky interface and all you are given is a
1336 * device number.  _Never_ to be used for internal purposes.  If you
1337 * ever need it - reconsider your API.
1338 *
1339 * On success, the returned block_device has reference count of one.
1340 *
1341 * CONTEXT:
1342 * Might sleep.
1343 *
1344 * RETURNS:
1345 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1346 */
1347struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1348{
1349        struct block_device *bdev;
1350        int err;
1351
1352        bdev = bdget(dev);
1353        if (!bdev)
1354                return ERR_PTR(-ENOMEM);
1355
1356        err = blkdev_get(bdev, mode, holder);
1357        if (err)
1358                return ERR_PTR(err);
1359
1360        return bdev;
1361}
1362EXPORT_SYMBOL(blkdev_get_by_dev);
1363
1364static int blkdev_open(struct inode * inode, struct file * filp)
1365{
1366        struct block_device *bdev;
1367
1368        /*
1369         * Preserve backwards compatibility and allow large file access
1370         * even if userspace doesn't ask for it explicitly. Some mkfs
1371         * binary needs it. We might want to drop this workaround
1372         * during an unstable branch.
1373         */
1374        filp->f_flags |= O_LARGEFILE;
1375
1376        if (filp->f_flags & O_NDELAY)
1377                filp->f_mode |= FMODE_NDELAY;
1378        if (filp->f_flags & O_EXCL)
1379                filp->f_mode |= FMODE_EXCL;
1380        if ((filp->f_flags & O_ACCMODE) == 3)
1381                filp->f_mode |= FMODE_WRITE_IOCTL;
1382
1383        bdev = bd_acquire(inode);
1384        if (bdev == NULL)
1385                return -ENOMEM;
1386
1387        filp->f_mapping = bdev->bd_inode->i_mapping;
1388
1389        return blkdev_get(bdev, filp->f_mode, filp);
1390}
1391
1392static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1393{
1394        int ret = 0;
1395        struct gendisk *disk = bdev->bd_disk;
1396        struct block_device *victim = NULL;
1397
1398        mutex_lock_nested(&bdev->bd_mutex, for_part);
1399        if (for_part)
1400                bdev->bd_part_count--;
1401
1402        if (!--bdev->bd_openers) {
1403                WARN_ON_ONCE(bdev->bd_holders);
1404                sync_blockdev(bdev);
1405                kill_bdev(bdev);
1406        }
1407        if (bdev->bd_contains == bdev) {
1408                if (disk->fops->release)
1409                        ret = disk->fops->release(disk, mode);
1410        }
1411        if (!bdev->bd_openers) {
1412                struct module *owner = disk->fops->owner;
1413
1414                put_disk(disk);
1415                module_put(owner);
1416                disk_put_part(bdev->bd_part);
1417                bdev->bd_part = NULL;
1418                bdev->bd_disk = NULL;
1419                bdev_inode_switch_bdi(bdev->bd_inode,
1420                                        &default_backing_dev_info);
1421                if (bdev != bdev->bd_contains)
1422                        victim = bdev->bd_contains;
1423                bdev->bd_contains = NULL;
1424        }
1425        mutex_unlock(&bdev->bd_mutex);
1426        bdput(bdev);
1427        if (victim)
1428                __blkdev_put(victim, mode, 1);
1429        return ret;
1430}
1431
1432int blkdev_put(struct block_device *bdev, fmode_t mode)
1433{
1434        if (mode & FMODE_EXCL) {
1435                bool bdev_free;
1436
1437                /*
1438                 * Release a claim on the device.  The holder fields
1439                 * are protected with bdev_lock.  bd_mutex is to
1440                 * synchronize disk_holder unlinking.
1441                 */
1442                mutex_lock(&bdev->bd_mutex);
1443                spin_lock(&bdev_lock);
1444
1445                WARN_ON_ONCE(--bdev->bd_holders < 0);
1446                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1447
1448                /* bd_contains might point to self, check in a separate step */
1449                if ((bdev_free = !bdev->bd_holders))
1450                        bdev->bd_holder = NULL;
1451                if (!bdev->bd_contains->bd_holders)
1452                        bdev->bd_contains->bd_holder = NULL;
1453
1454                spin_unlock(&bdev_lock);
1455
1456                /*
1457                 * If this was the last claim, remove holder link and
1458                 * unblock evpoll if it was a write holder.
1459                 */
1460                if (bdev_free) {
1461                        if (bdev->bd_write_holder) {
1462                                disk_unblock_events(bdev->bd_disk);
1463                                disk_check_events(bdev->bd_disk);
1464                                bdev->bd_write_holder = false;
1465                        }
1466                }
1467
1468                mutex_unlock(&bdev->bd_mutex);
1469        }
1470
1471        return __blkdev_put(bdev, mode, 0);
1472}
1473EXPORT_SYMBOL(blkdev_put);
1474
1475static int blkdev_close(struct inode * inode, struct file * filp)
1476{
1477        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1478
1479        return blkdev_put(bdev, filp->f_mode);
1480}
1481
1482static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1483{
1484        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1485        fmode_t mode = file->f_mode;
1486
1487        /*
1488         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1489         * to updated it before every ioctl.
1490         */
1491        if (file->f_flags & O_NDELAY)
1492                mode |= FMODE_NDELAY;
1493        else
1494                mode &= ~FMODE_NDELAY;
1495
1496        return blkdev_ioctl(bdev, mode, cmd, arg);
1497}
1498
1499/*
1500 * Write data to the block device.  Only intended for the block device itself
1501 * and the raw driver which basically is a fake block device.
1502 *
1503 * Does not take i_mutex for the write and thus is not for general purpose
1504 * use.
1505 */
1506ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1507                         unsigned long nr_segs, loff_t pos)
1508{
1509        struct file *file = iocb->ki_filp;
1510        ssize_t ret;
1511
1512        BUG_ON(iocb->ki_pos != pos);
1513
1514        ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1515        if (ret > 0 || ret == -EIOCBQUEUED) {
1516                ssize_t err;
1517
1518                err = generic_write_sync(file, pos, ret);
1519                if (err < 0 && ret > 0)
1520                        ret = err;
1521        }
1522        return ret;
1523}
1524EXPORT_SYMBOL_GPL(blkdev_aio_write);
1525
1526/*
1527 * Try to release a page associated with block device when the system
1528 * is under memory pressure.
1529 */
1530static int blkdev_releasepage(struct page *page, gfp_t wait)
1531{
1532        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1533
1534        if (super && super->s_op->bdev_try_to_free_page)
1535                return super->s_op->bdev_try_to_free_page(super, page, wait);
1536
1537        return try_to_free_buffers(page);
1538}
1539
1540static const struct address_space_operations def_blk_aops = {
1541        .readpage       = blkdev_readpage,
1542        .writepage      = blkdev_writepage,
1543        .write_begin    = blkdev_write_begin,
1544        .write_end      = blkdev_write_end,
1545        .writepages     = generic_writepages,
1546        .releasepage    = blkdev_releasepage,
1547        .direct_IO      = blkdev_direct_IO,
1548};
1549
1550const struct file_operations def_blk_fops = {
1551        .open           = blkdev_open,
1552        .release        = blkdev_close,
1553        .llseek         = block_llseek,
1554        .read           = do_sync_read,
1555        .write          = do_sync_write,
1556        .aio_read       = generic_file_aio_read,
1557        .aio_write      = blkdev_aio_write,
1558        .mmap           = generic_file_mmap,
1559        .fsync          = blkdev_fsync,
1560        .unlocked_ioctl = block_ioctl,
1561#ifdef CONFIG_COMPAT
1562        .compat_ioctl   = compat_blkdev_ioctl,
1563#endif
1564        .splice_read    = generic_file_splice_read,
1565        .splice_write   = generic_file_splice_write,
1566};
1567
1568int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1569{
1570        int res;
1571        mm_segment_t old_fs = get_fs();
1572        set_fs(KERNEL_DS);
1573        res = blkdev_ioctl(bdev, 0, cmd, arg);
1574        set_fs(old_fs);
1575        return res;
1576}
1577
1578EXPORT_SYMBOL(ioctl_by_bdev);
1579
1580/**
1581 * lookup_bdev  - lookup a struct block_device by name
1582 * @pathname:   special file representing the block device
1583 *
1584 * Get a reference to the blockdevice at @pathname in the current
1585 * namespace if possible and return it.  Return ERR_PTR(error)
1586 * otherwise.
1587 */
1588struct block_device *lookup_bdev(const char *pathname)
1589{
1590        struct block_device *bdev;
1591        struct inode *inode;
1592        struct path path;
1593        int error;
1594
1595        if (!pathname || !*pathname)
1596                return ERR_PTR(-EINVAL);
1597
1598        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1599        if (error)
1600                return ERR_PTR(error);
1601
1602        inode = path.dentry->d_inode;
1603        error = -ENOTBLK;
1604        if (!S_ISBLK(inode->i_mode))
1605                goto fail;
1606        error = -EACCES;
1607        if (path.mnt->mnt_flags & MNT_NODEV)
1608                goto fail;
1609        error = -ENOMEM;
1610        bdev = bd_acquire(inode);
1611        if (!bdev)
1612                goto fail;
1613out:
1614        path_put(&path);
1615        return bdev;
1616fail:
1617        bdev = ERR_PTR(error);
1618        goto out;
1619}
1620EXPORT_SYMBOL(lookup_bdev);
1621
1622int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1623{
1624        struct super_block *sb = get_super(bdev);
1625        int res = 0;
1626
1627        if (sb) {
1628                /*
1629                 * no need to lock the super, get_super holds the
1630                 * read mutex so the filesystem cannot go away
1631                 * under us (->put_super runs with the write lock
1632                 * hold).
1633                 */
1634                shrink_dcache_sb(sb);
1635                res = invalidate_inodes(sb, kill_dirty);
1636                drop_super(sb);
1637        }
1638        invalidate_bdev(bdev);
1639        return res;
1640}
1641EXPORT_SYMBOL(__invalidate_device);
1642