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}
  47EXPORT_SYMBOL(I_BDEV);
  48
  49/*
  50 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
  51 * need to move it onto the dirty list of @dst so that the inode is always on
  52 * the right list.
  53 */
  54static void bdev_inode_switch_bdi(struct inode *inode,
  55                        struct backing_dev_info *dst)
  56{
  57        struct backing_dev_info *old = inode->i_data.backing_dev_info;
  58
  59        if (unlikely(dst == old))               /* deadlock avoidance */
  60                return;
  61        bdi_lock_two(&old->wb, &dst->wb);
  62        spin_lock(&inode->i_lock);
  63        inode->i_data.backing_dev_info = dst;
  64        if (inode->i_state & I_DIRTY)
  65                list_move(&inode->i_wb_list, &dst->wb.b_dirty);
  66        spin_unlock(&inode->i_lock);
  67        spin_unlock(&old->wb.list_lock);
  68        spin_unlock(&dst->wb.list_lock);
  69}
  70
  71static sector_t max_block(struct block_device *bdev)
  72{
  73        sector_t retval = ~((sector_t)0);
  74        loff_t sz = i_size_read(bdev->bd_inode);
  75
  76        if (sz) {
  77                unsigned int size = block_size(bdev);
  78                unsigned int sizebits = blksize_bits(size);
  79                retval = (sz >> sizebits);
  80        }
  81        return retval;
  82}
  83
  84/* Kill _all_ buffers and pagecache , dirty or not.. */
  85static void kill_bdev(struct block_device *bdev)
  86{
  87        if (bdev->bd_inode->i_mapping->nrpages == 0)
  88                return;
  89        invalidate_bh_lrus();
  90        truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  91}       
  92
  93int set_blocksize(struct block_device *bdev, int size)
  94{
  95        /* Size must be a power of two, and between 512 and PAGE_SIZE */
  96        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
  97                return -EINVAL;
  98
  99        /* Size cannot be smaller than the size supported by the device */
 100        if (size < bdev_logical_block_size(bdev))
 101                return -EINVAL;
 102
 103        /* Don't change the size if it is same as current */
 104        if (bdev->bd_block_size != size) {
 105                sync_blockdev(bdev);
 106                bdev->bd_block_size = size;
 107                bdev->bd_inode->i_blkbits = blksize_bits(size);
 108                kill_bdev(bdev);
 109        }
 110        return 0;
 111}
 112
 113EXPORT_SYMBOL(set_blocksize);
 114
 115int sb_set_blocksize(struct super_block *sb, int size)
 116{
 117        if (set_blocksize(sb->s_bdev, size))
 118                return 0;
 119        /* If we get here, we know size is power of two
 120         * and it's value is between 512 and PAGE_SIZE */
 121        sb->s_blocksize = size;
 122        sb->s_blocksize_bits = blksize_bits(size);
 123        return sb->s_blocksize;
 124}
 125
 126EXPORT_SYMBOL(sb_set_blocksize);
 127
 128int sb_min_blocksize(struct super_block *sb, int size)
 129{
 130        int minsize = bdev_logical_block_size(sb->s_bdev);
 131        if (size < minsize)
 132                size = minsize;
 133        return sb_set_blocksize(sb, size);
 134}
 135
 136EXPORT_SYMBOL(sb_min_blocksize);
 137
 138static int
 139blkdev_get_block(struct inode *inode, sector_t iblock,
 140                struct buffer_head *bh, int create)
 141{
 142        if (iblock >= max_block(I_BDEV(inode))) {
 143                if (create)
 144                        return -EIO;
 145
 146                /*
 147                 * for reads, we're just trying to fill a partial page.
 148                 * return a hole, they will have to call get_block again
 149                 * before they can fill it, and they will get -EIO at that
 150                 * time
 151                 */
 152                return 0;
 153        }
 154        bh->b_bdev = I_BDEV(inode);
 155        bh->b_blocknr = iblock;
 156        set_buffer_mapped(bh);
 157        return 0;
 158}
 159
 160static int
 161blkdev_get_blocks(struct inode *inode, sector_t iblock,
 162                struct buffer_head *bh, int create)
 163{
 164        sector_t end_block = max_block(I_BDEV(inode));
 165        unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
 166
 167        if ((iblock + max_blocks) > end_block) {
 168                max_blocks = end_block - iblock;
 169                if ((long)max_blocks <= 0) {
 170                        if (create)
 171                                return -EIO;    /* write fully beyond EOF */
 172                        /*
 173                         * It is a read which is fully beyond EOF.  We return
 174                         * a !buffer_mapped buffer
 175                         */
 176                        max_blocks = 0;
 177                }
 178        }
 179
 180        bh->b_bdev = I_BDEV(inode);
 181        bh->b_blocknr = iblock;
 182        bh->b_size = max_blocks << inode->i_blkbits;
 183        if (max_blocks)
 184                set_buffer_mapped(bh);
 185        return 0;
 186}
 187
 188static ssize_t
 189blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 190                        loff_t offset, unsigned long nr_segs)
 191{
 192        struct file *file = iocb->ki_filp;
 193        struct inode *inode = file->f_mapping->host;
 194
 195        return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 196                                    nr_segs, blkdev_get_blocks, NULL, NULL, 0);
 197}
 198
 199int __sync_blockdev(struct block_device *bdev, int wait)
 200{
 201        if (!bdev)
 202                return 0;
 203        if (!wait)
 204                return filemap_flush(bdev->bd_inode->i_mapping);
 205        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 206}
 207
 208/*
 209 * Write out and wait upon all the dirty data associated with a block
 210 * device via its mapping.  Does not take the superblock lock.
 211 */
 212int sync_blockdev(struct block_device *bdev)
 213{
 214        return __sync_blockdev(bdev, 1);
 215}
 216EXPORT_SYMBOL(sync_blockdev);
 217
 218/*
 219 * Write out and wait upon all dirty data associated with this
 220 * device.   Filesystem data as well as the underlying block
 221 * device.  Takes the superblock lock.
 222 */
 223int fsync_bdev(struct block_device *bdev)
 224{
 225        struct super_block *sb = get_super(bdev);
 226        if (sb) {
 227                int res = sync_filesystem(sb);
 228                drop_super(sb);
 229                return res;
 230        }
 231        return sync_blockdev(bdev);
 232}
 233EXPORT_SYMBOL(fsync_bdev);
 234
 235/**
 236 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 237 * @bdev:       blockdevice to lock
 238 *
 239 * If a superblock is found on this device, we take the s_umount semaphore
 240 * on it to make sure nobody unmounts until the snapshot creation is done.
 241 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 242 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 243 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 244 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 245 * actually.
 246 */
 247struct super_block *freeze_bdev(struct block_device *bdev)
 248{
 249        struct super_block *sb;
 250        int error = 0;
 251
 252        mutex_lock(&bdev->bd_fsfreeze_mutex);
 253        if (++bdev->bd_fsfreeze_count > 1) {
 254                /*
 255                 * We don't even need to grab a reference - the first call
 256                 * to freeze_bdev grab an active reference and only the last
 257                 * thaw_bdev drops it.
 258                 */
 259                sb = get_super(bdev);
 260                drop_super(sb);
 261                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 262                return sb;
 263        }
 264
 265        sb = get_active_super(bdev);
 266        if (!sb)
 267                goto out;
 268        error = freeze_super(sb);
 269        if (error) {
 270                deactivate_super(sb);
 271                bdev->bd_fsfreeze_count--;
 272                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 273                return ERR_PTR(error);
 274        }
 275        deactivate_super(sb);
 276 out:
 277        sync_blockdev(bdev);
 278        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 279        return sb;      /* thaw_bdev releases s->s_umount */
 280}
 281EXPORT_SYMBOL(freeze_bdev);
 282
 283/**
 284 * thaw_bdev  -- unlock filesystem
 285 * @bdev:       blockdevice to unlock
 286 * @sb:         associated superblock
 287 *
 288 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 289 */
 290int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 291{
 292        int error = -EINVAL;
 293
 294        mutex_lock(&bdev->bd_fsfreeze_mutex);
 295        if (!bdev->bd_fsfreeze_count)
 296                goto out;
 297
 298        error = 0;
 299        if (--bdev->bd_fsfreeze_count > 0)
 300                goto out;
 301
 302        if (!sb)
 303                goto out;
 304
 305        error = thaw_super(sb);
 306        if (error) {
 307                bdev->bd_fsfreeze_count++;
 308                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 309                return error;
 310        }
 311out:
 312        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 313        return 0;
 314}
 315EXPORT_SYMBOL(thaw_bdev);
 316
 317static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 318{
 319        return block_write_full_page(page, blkdev_get_block, wbc);
 320}
 321
 322static int blkdev_readpage(struct file * file, struct page * page)
 323{
 324        return block_read_full_page(page, blkdev_get_block);
 325}
 326
 327static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 328                        loff_t pos, unsigned len, unsigned flags,
 329                        struct page **pagep, void **fsdata)
 330{
 331        return block_write_begin(mapping, pos, len, flags, pagep,
 332                                 blkdev_get_block);
 333}
 334
 335static int blkdev_write_end(struct file *file, struct address_space *mapping,
 336                        loff_t pos, unsigned len, unsigned copied,
 337                        struct page *page, void *fsdata)
 338{
 339        int ret;
 340        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 341
 342        unlock_page(page);
 343        page_cache_release(page);
 344
 345        return ret;
 346}
 347
 348/*
 349 * private llseek:
 350 * for a block special file file->f_path.dentry->d_inode->i_size is zero
 351 * so we compute the size by hand (just as in block_read/write above)
 352 */
 353static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 354{
 355        struct inode *bd_inode = file->f_mapping->host;
 356        loff_t size;
 357        loff_t retval;
 358
 359        mutex_lock(&bd_inode->i_mutex);
 360        size = i_size_read(bd_inode);
 361
 362        retval = -EINVAL;
 363        switch (origin) {
 364                case SEEK_END:
 365                        offset += size;
 366                        break;
 367                case SEEK_CUR:
 368                        offset += file->f_pos;
 369                case SEEK_SET:
 370                        break;
 371                default:
 372                        goto out;
 373        }
 374        if (offset >= 0 && offset <= size) {
 375                if (offset != file->f_pos) {
 376                        file->f_pos = offset;
 377                }
 378                retval = offset;
 379        }
 380out:
 381        mutex_unlock(&bd_inode->i_mutex);
 382        return retval;
 383}
 384        
 385int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 386{
 387        struct inode *bd_inode = filp->f_mapping->host;
 388        struct block_device *bdev = I_BDEV(bd_inode);
 389        int error;
 390        
 391        error = filemap_write_and_wait_range(filp->f_mapping, start, end);
 392        if (error)
 393                return error;
 394
 395        /*
 396         * There is no need to serialise calls to blkdev_issue_flush with
 397         * i_mutex and doing so causes performance issues with concurrent
 398         * O_SYNC writers to a block device.
 399         */
 400        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 401        if (error == -EOPNOTSUPP)
 402                error = 0;
 403
 404        return error;
 405}
 406EXPORT_SYMBOL(blkdev_fsync);
 407
 408/*
 409 * pseudo-fs
 410 */
 411
 412static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 413static struct kmem_cache * bdev_cachep __read_mostly;
 414
 415static struct inode *bdev_alloc_inode(struct super_block *sb)
 416{
 417        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 418        if (!ei)
 419                return NULL;
 420        return &ei->vfs_inode;
 421}
 422
 423static void bdev_i_callback(struct rcu_head *head)
 424{
 425        struct inode *inode = container_of(head, struct inode, i_rcu);
 426        struct bdev_inode *bdi = BDEV_I(inode);
 427
 428        INIT_LIST_HEAD(&inode->i_dentry);
 429        kmem_cache_free(bdev_cachep, bdi);
 430}
 431
 432static void bdev_destroy_inode(struct inode *inode)
 433{
 434        call_rcu(&inode->i_rcu, bdev_i_callback);
 435}
 436
 437static void init_once(void *foo)
 438{
 439        struct bdev_inode *ei = (struct bdev_inode *) foo;
 440        struct block_device *bdev = &ei->bdev;
 441
 442        memset(bdev, 0, sizeof(*bdev));
 443        mutex_init(&bdev->bd_mutex);
 444        INIT_LIST_HEAD(&bdev->bd_inodes);
 445        INIT_LIST_HEAD(&bdev->bd_list);
 446#ifdef CONFIG_SYSFS
 447        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 448#endif
 449        inode_init_once(&ei->vfs_inode);
 450        /* Initialize mutex for freeze. */
 451        mutex_init(&bdev->bd_fsfreeze_mutex);
 452}
 453
 454static inline void __bd_forget(struct inode *inode)
 455{
 456        list_del_init(&inode->i_devices);
 457        inode->i_bdev = NULL;
 458        inode->i_mapping = &inode->i_data;
 459}
 460
 461static void bdev_evict_inode(struct inode *inode)
 462{
 463        struct block_device *bdev = &BDEV_I(inode)->bdev;
 464        struct list_head *p;
 465        truncate_inode_pages(&inode->i_data, 0);
 466        invalidate_inode_buffers(inode); /* is it needed here? */
 467        end_writeback(inode);
 468        spin_lock(&bdev_lock);
 469        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 470                __bd_forget(list_entry(p, struct inode, i_devices));
 471        }
 472        list_del_init(&bdev->bd_list);
 473        spin_unlock(&bdev_lock);
 474}
 475
 476static const struct super_operations bdev_sops = {
 477        .statfs = simple_statfs,
 478        .alloc_inode = bdev_alloc_inode,
 479        .destroy_inode = bdev_destroy_inode,
 480        .drop_inode = generic_delete_inode,
 481        .evict_inode = bdev_evict_inode,
 482};
 483
 484static struct dentry *bd_mount(struct file_system_type *fs_type,
 485        int flags, const char *dev_name, void *data)
 486{
 487        return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
 488}
 489
 490static struct file_system_type bd_type = {
 491        .name           = "bdev",
 492        .mount          = bd_mount,
 493        .kill_sb        = kill_anon_super,
 494};
 495
 496struct super_block *blockdev_superblock __read_mostly;
 497
 498void __init bdev_cache_init(void)
 499{
 500        int err;
 501        struct vfsmount *bd_mnt;
 502
 503        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 504                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 505                                SLAB_MEM_SPREAD|SLAB_PANIC),
 506                        init_once);
 507        err = register_filesystem(&bd_type);
 508        if (err)
 509                panic("Cannot register bdev pseudo-fs");
 510        bd_mnt = kern_mount(&bd_type);
 511        if (IS_ERR(bd_mnt))
 512                panic("Cannot create bdev pseudo-fs");
 513        /*
 514         * This vfsmount structure is only used to obtain the
 515         * blockdev_superblock, so tell kmemleak not to report it.
 516         */
 517        kmemleak_not_leak(bd_mnt);
 518        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 519}
 520
 521/*
 522 * Most likely _very_ bad one - but then it's hardly critical for small
 523 * /dev and can be fixed when somebody will need really large one.
 524 * Keep in mind that it will be fed through icache hash function too.
 525 */
 526static inline unsigned long hash(dev_t dev)
 527{
 528        return MAJOR(dev)+MINOR(dev);
 529}
 530
 531static int bdev_test(struct inode *inode, void *data)
 532{
 533        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 534}
 535
 536static int bdev_set(struct inode *inode, void *data)
 537{
 538        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 539        return 0;
 540}
 541
 542static LIST_HEAD(all_bdevs);
 543
 544struct block_device *bdget(dev_t dev)
 545{
 546        struct block_device *bdev;
 547        struct inode *inode;
 548
 549        inode = iget5_locked(blockdev_superblock, hash(dev),
 550                        bdev_test, bdev_set, &dev);
 551
 552        if (!inode)
 553                return NULL;
 554
 555        bdev = &BDEV_I(inode)->bdev;
 556
 557        if (inode->i_state & I_NEW) {
 558                bdev->bd_contains = NULL;
 559                bdev->bd_super = NULL;
 560                bdev->bd_inode = inode;
 561                bdev->bd_block_size = (1 << inode->i_blkbits);
 562                bdev->bd_part_count = 0;
 563                bdev->bd_invalidated = 0;
 564                inode->i_mode = S_IFBLK;
 565                inode->i_rdev = dev;
 566                inode->i_bdev = bdev;
 567                inode->i_data.a_ops = &def_blk_aops;
 568                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 569                inode->i_data.backing_dev_info = &default_backing_dev_info;
 570                spin_lock(&bdev_lock);
 571                list_add(&bdev->bd_list, &all_bdevs);
 572                spin_unlock(&bdev_lock);
 573                unlock_new_inode(inode);
 574        }
 575        return bdev;
 576}
 577
 578EXPORT_SYMBOL(bdget);
 579
 580/**
 581 * bdgrab -- Grab a reference to an already referenced block device
 582 * @bdev:       Block device to grab a reference to.
 583 */
 584struct block_device *bdgrab(struct block_device *bdev)
 585{
 586        ihold(bdev->bd_inode);
 587        return bdev;
 588}
 589
 590long nr_blockdev_pages(void)
 591{
 592        struct block_device *bdev;
 593        long ret = 0;
 594        spin_lock(&bdev_lock);
 595        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 596                ret += bdev->bd_inode->i_mapping->nrpages;
 597        }
 598        spin_unlock(&bdev_lock);
 599        return ret;
 600}
 601
 602void bdput(struct block_device *bdev)
 603{
 604        iput(bdev->bd_inode);
 605}
 606
 607EXPORT_SYMBOL(bdput);
 608 
 609static struct block_device *bd_acquire(struct inode *inode)
 610{
 611        struct block_device *bdev;
 612
 613        spin_lock(&bdev_lock);
 614        bdev = inode->i_bdev;
 615        if (bdev) {
 616                ihold(bdev->bd_inode);
 617                spin_unlock(&bdev_lock);
 618                return bdev;
 619        }
 620        spin_unlock(&bdev_lock);
 621
 622        bdev = bdget(inode->i_rdev);
 623        if (bdev) {
 624                spin_lock(&bdev_lock);
 625                if (!inode->i_bdev) {
 626                        /*
 627                         * We take an additional reference to bd_inode,
 628                         * and it's released in clear_inode() of inode.
 629                         * So, we can access it via ->i_mapping always
 630                         * without igrab().
 631                         */
 632                        ihold(bdev->bd_inode);
 633                        inode->i_bdev = bdev;
 634                        inode->i_mapping = bdev->bd_inode->i_mapping;
 635                        list_add(&inode->i_devices, &bdev->bd_inodes);
 636                }
 637                spin_unlock(&bdev_lock);
 638        }
 639        return bdev;
 640}
 641
 642/* Call when you free inode */
 643
 644void bd_forget(struct inode *inode)
 645{
 646        struct block_device *bdev = NULL;
 647
 648        spin_lock(&bdev_lock);
 649        if (inode->i_bdev) {
 650                if (!sb_is_blkdev_sb(inode->i_sb))
 651                        bdev = inode->i_bdev;
 652                __bd_forget(inode);
 653        }
 654        spin_unlock(&bdev_lock);
 655
 656        if (bdev)
 657                iput(bdev->bd_inode);
 658}
 659
 660/**
 661 * bd_may_claim - test whether a block device can be claimed
 662 * @bdev: block device of interest
 663 * @whole: whole block device containing @bdev, may equal @bdev
 664 * @holder: holder trying to claim @bdev
 665 *
 666 * Test whether @bdev can be claimed by @holder.
 667 *
 668 * CONTEXT:
 669 * spin_lock(&bdev_lock).
 670 *
 671 * RETURNS:
 672 * %true if @bdev can be claimed, %false otherwise.
 673 */
 674static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 675                         void *holder)
 676{
 677        if (bdev->bd_holder == holder)
 678                return true;     /* already a holder */
 679        else if (bdev->bd_holder != NULL)
 680                return false;    /* held by someone else */
 681        else if (bdev->bd_contains == bdev)
 682                return true;     /* is a whole device which isn't held */
 683
 684        else if (whole->bd_holder == bd_may_claim)
 685                return true;     /* is a partition of a device that is being partitioned */
 686        else if (whole->bd_holder != NULL)
 687                return false;    /* is a partition of a held device */
 688        else
 689                return true;     /* is a partition of an un-held device */
 690}
 691
 692/**
 693 * bd_prepare_to_claim - prepare to claim a block device
 694 * @bdev: block device of interest
 695 * @whole: the whole device containing @bdev, may equal @bdev
 696 * @holder: holder trying to claim @bdev
 697 *
 698 * Prepare to claim @bdev.  This function fails if @bdev is already
 699 * claimed by another holder and waits if another claiming is in
 700 * progress.  This function doesn't actually claim.  On successful
 701 * return, the caller has ownership of bd_claiming and bd_holder[s].
 702 *
 703 * CONTEXT:
 704 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 705 * it multiple times.
 706 *
 707 * RETURNS:
 708 * 0 if @bdev can be claimed, -EBUSY otherwise.
 709 */
 710static int bd_prepare_to_claim(struct block_device *bdev,
 711                               struct block_device *whole, void *holder)
 712{
 713retry:
 714        /* if someone else claimed, fail */
 715        if (!bd_may_claim(bdev, whole, holder))
 716                return -EBUSY;
 717
 718        /* if claiming is already in progress, wait for it to finish */
 719        if (whole->bd_claiming) {
 720                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 721                DEFINE_WAIT(wait);
 722
 723                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 724                spin_unlock(&bdev_lock);
 725                schedule();
 726                finish_wait(wq, &wait);
 727                spin_lock(&bdev_lock);
 728                goto retry;
 729        }
 730
 731        /* yay, all mine */
 732        return 0;
 733}
 734
 735/**
 736 * bd_start_claiming - start claiming a block device
 737 * @bdev: block device of interest
 738 * @holder: holder trying to claim @bdev
 739 *
 740 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 741 * exclusively and mark that an exclusive open is in progress.  Each
 742 * successful call to this function must be matched with a call to
 743 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 744 * fail).
 745 *
 746 * This function is used to gain exclusive access to the block device
 747 * without actually causing other exclusive open attempts to fail. It
 748 * should be used when the open sequence itself requires exclusive
 749 * access but may subsequently fail.
 750 *
 751 * CONTEXT:
 752 * Might sleep.
 753 *
 754 * RETURNS:
 755 * Pointer to the block device containing @bdev on success, ERR_PTR()
 756 * value on failure.
 757 */
 758static struct block_device *bd_start_claiming(struct block_device *bdev,
 759                                              void *holder)
 760{
 761        struct gendisk *disk;
 762        struct block_device *whole;
 763        int partno, err;
 764
 765        might_sleep();
 766
 767        /*
 768         * @bdev might not have been initialized properly yet, look up
 769         * and grab the outer block device the hard way.
 770         */
 771        disk = get_gendisk(bdev->bd_dev, &partno);
 772        if (!disk)
 773                return ERR_PTR(-ENXIO);
 774
 775        /*
 776         * Normally, @bdev should equal what's returned from bdget_disk()
 777         * if partno is 0; however, some drivers (floppy) use multiple
 778         * bdev's for the same physical device and @bdev may be one of the
 779         * aliases.  Keep @bdev if partno is 0.  This means claimer
 780         * tracking is broken for those devices but it has always been that
 781         * way.
 782         */
 783        if (partno)
 784                whole = bdget_disk(disk, 0);
 785        else
 786                whole = bdgrab(bdev);
 787
 788        module_put(disk->fops->owner);
 789        put_disk(disk);
 790        if (!whole)
 791                return ERR_PTR(-ENOMEM);
 792
 793        /* prepare to claim, if successful, mark claiming in progress */
 794        spin_lock(&bdev_lock);
 795
 796        err = bd_prepare_to_claim(bdev, whole, holder);
 797        if (err == 0) {
 798                whole->bd_claiming = holder;
 799                spin_unlock(&bdev_lock);
 800                return whole;
 801        } else {
 802                spin_unlock(&bdev_lock);
 803                bdput(whole);
 804                return ERR_PTR(err);
 805        }
 806}
 807
 808#ifdef CONFIG_SYSFS
 809struct bd_holder_disk {
 810        struct list_head        list;
 811        struct gendisk          *disk;
 812        int                     refcnt;
 813};
 814
 815static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 816                                                  struct gendisk *disk)
 817{
 818        struct bd_holder_disk *holder;
 819
 820        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 821                if (holder->disk == disk)
 822                        return holder;
 823        return NULL;
 824}
 825
 826static int add_symlink(struct kobject *from, struct kobject *to)
 827{
 828        return sysfs_create_link(from, to, kobject_name(to));
 829}
 830
 831static void del_symlink(struct kobject *from, struct kobject *to)
 832{
 833        sysfs_remove_link(from, kobject_name(to));
 834}
 835
 836/**
 837 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 838 * @bdev: the claimed slave bdev
 839 * @disk: the holding disk
 840 *
 841 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 842 *
 843 * This functions creates the following sysfs symlinks.
 844 *
 845 * - from "slaves" directory of the holder @disk to the claimed @bdev
 846 * - from "holders" directory of the @bdev to the holder @disk
 847 *
 848 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
 849 * passed to bd_link_disk_holder(), then:
 850 *
 851 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
 852 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 853 *
 854 * The caller must have claimed @bdev before calling this function and
 855 * ensure that both @bdev and @disk are valid during the creation and
 856 * lifetime of these symlinks.
 857 *
 858 * CONTEXT:
 859 * Might sleep.
 860 *
 861 * RETURNS:
 862 * 0 on success, -errno on failure.
 863 */
 864int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 865{
 866        struct bd_holder_disk *holder;
 867        int ret = 0;
 868
 869        mutex_lock(&bdev->bd_mutex);
 870
 871        WARN_ON_ONCE(!bdev->bd_holder);
 872
 873        /* FIXME: remove the following once add_disk() handles errors */
 874        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
 875                goto out_unlock;
 876
 877        holder = bd_find_holder_disk(bdev, disk);
 878        if (holder) {
 879                holder->refcnt++;
 880                goto out_unlock;
 881        }
 882
 883        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 884        if (!holder) {
 885                ret = -ENOMEM;
 886                goto out_unlock;
 887        }
 888
 889        INIT_LIST_HEAD(&holder->list);
 890        holder->disk = disk;
 891        holder->refcnt = 1;
 892
 893        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 894        if (ret)
 895                goto out_free;
 896
 897        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
 898        if (ret)
 899                goto out_del;
 900        /*
 901         * bdev could be deleted beneath us which would implicitly destroy
 902         * the holder directory.  Hold on to it.
 903         */
 904        kobject_get(bdev->bd_part->holder_dir);
 905
 906        list_add(&holder->list, &bdev->bd_holder_disks);
 907        goto out_unlock;
 908
 909out_del:
 910        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 911out_free:
 912        kfree(holder);
 913out_unlock:
 914        mutex_unlock(&bdev->bd_mutex);
 915        return ret;
 916}
 917EXPORT_SYMBOL_GPL(bd_link_disk_holder);
 918
 919/**
 920 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
 921 * @bdev: the calimed slave bdev
 922 * @disk: the holding disk
 923 *
 924 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 925 *
 926 * CONTEXT:
 927 * Might sleep.
 928 */
 929void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 930{
 931        struct bd_holder_disk *holder;
 932
 933        mutex_lock(&bdev->bd_mutex);
 934
 935        holder = bd_find_holder_disk(bdev, disk);
 936
 937        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 938                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 939                del_symlink(bdev->bd_part->holder_dir,
 940                            &disk_to_dev(disk)->kobj);
 941                kobject_put(bdev->bd_part->holder_dir);
 942                list_del_init(&holder->list);
 943                kfree(holder);
 944        }
 945
 946        mutex_unlock(&bdev->bd_mutex);
 947}
 948EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
 949#endif
 950
 951/**
 952 * flush_disk - invalidates all buffer-cache entries on a disk
 953 *
 954 * @bdev:      struct block device to be flushed
 955 * @kill_dirty: flag to guide handling of dirty inodes
 956 *
 957 * Invalidates all buffer-cache entries on a disk. It should be called
 958 * when a disk has been changed -- either by a media change or online
 959 * resize.
 960 */
 961static void flush_disk(struct block_device *bdev, bool kill_dirty)
 962{
 963        if (__invalidate_device(bdev, kill_dirty)) {
 964                char name[BDEVNAME_SIZE] = "";
 965
 966                if (bdev->bd_disk)
 967                        disk_name(bdev->bd_disk, 0, name);
 968                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 969                       "resized disk %s\n", name);
 970        }
 971
 972        if (!bdev->bd_disk)
 973                return;
 974        if (disk_partitionable(bdev->bd_disk))
 975                bdev->bd_invalidated = 1;
 976}
 977
 978/**
 979 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 980 * @disk: struct gendisk to check
 981 * @bdev: struct bdev to adjust.
 982 *
 983 * This routine checks to see if the bdev size does not match the disk size
 984 * and adjusts it if it differs.
 985 */
 986void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 987{
 988        loff_t disk_size, bdev_size;
 989
 990        disk_size = (loff_t)get_capacity(disk) << 9;
 991        bdev_size = i_size_read(bdev->bd_inode);
 992        if (disk_size != bdev_size) {
 993                char name[BDEVNAME_SIZE];
 994
 995                disk_name(disk, 0, name);
 996                printk(KERN_INFO
 997                       "%s: detected capacity change from %lld to %lld\n",
 998                       name, bdev_size, disk_size);
 999                i_size_write(bdev->bd_inode, disk_size);
1000                flush_disk(bdev, false);
1001        }
1002}
1003EXPORT_SYMBOL(check_disk_size_change);
1004
1005/**
1006 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1007 * @disk: struct gendisk to be revalidated
1008 *
1009 * This routine is a wrapper for lower-level driver's revalidate_disk
1010 * call-backs.  It is used to do common pre and post operations needed
1011 * for all revalidate_disk operations.
1012 */
1013int revalidate_disk(struct gendisk *disk)
1014{
1015        struct block_device *bdev;
1016        int ret = 0;
1017
1018        if (disk->fops->revalidate_disk)
1019                ret = disk->fops->revalidate_disk(disk);
1020
1021        bdev = bdget_disk(disk, 0);
1022        if (!bdev)
1023                return ret;
1024
1025        mutex_lock(&bdev->bd_mutex);
1026        check_disk_size_change(disk, bdev);
1027        mutex_unlock(&bdev->bd_mutex);
1028        bdput(bdev);
1029        return ret;
1030}
1031EXPORT_SYMBOL(revalidate_disk);
1032
1033/*
1034 * This routine checks whether a removable media has been changed,
1035 * and invalidates all buffer-cache-entries in that case. This
1036 * is a relatively slow routine, so we have to try to minimize using
1037 * it. Thus it is called only upon a 'mount' or 'open'. This
1038 * is the best way of combining speed and utility, I think.
1039 * People changing diskettes in the middle of an operation deserve
1040 * to lose :-)
1041 */
1042int check_disk_change(struct block_device *bdev)
1043{
1044        struct gendisk *disk = bdev->bd_disk;
1045        const struct block_device_operations *bdops = disk->fops;
1046        unsigned int events;
1047
1048        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1049                                   DISK_EVENT_EJECT_REQUEST);
1050        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1051                return 0;
1052
1053        flush_disk(bdev, true);
1054        if (bdops->revalidate_disk)
1055                bdops->revalidate_disk(bdev->bd_disk);
1056        return 1;
1057}
1058
1059EXPORT_SYMBOL(check_disk_change);
1060
1061void bd_set_size(struct block_device *bdev, loff_t size)
1062{
1063        unsigned bsize = bdev_logical_block_size(bdev);
1064
1065        bdev->bd_inode->i_size = size;
1066        while (bsize < PAGE_CACHE_SIZE) {
1067                if (size & bsize)
1068                        break;
1069                bsize <<= 1;
1070        }
1071        bdev->bd_block_size = bsize;
1072        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1073}
1074EXPORT_SYMBOL(bd_set_size);
1075
1076static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1077
1078/*
1079 * bd_mutex locking:
1080 *
1081 *  mutex_lock(part->bd_mutex)
1082 *    mutex_lock_nested(whole->bd_mutex, 1)
1083 */
1084
1085static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1086{
1087        struct gendisk *disk;
1088        struct module *owner;
1089        int ret;
1090        int partno;
1091        int perm = 0;
1092
1093        if (mode & FMODE_READ)
1094                perm |= MAY_READ;
1095        if (mode & FMODE_WRITE)
1096                perm |= MAY_WRITE;
1097        /*
1098         * hooks: /n/, see "layering violations".
1099         */
1100        if (!for_part) {
1101                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1102                if (ret != 0) {
1103                        bdput(bdev);
1104                        return ret;
1105                }
1106        }
1107
1108 restart:
1109
1110        ret = -ENXIO;
1111        disk = get_gendisk(bdev->bd_dev, &partno);
1112        if (!disk)
1113                goto out;
1114        owner = disk->fops->owner;
1115
1116        disk_block_events(disk);
1117        mutex_lock_nested(&bdev->bd_mutex, for_part);
1118        if (!bdev->bd_openers) {
1119                bdev->bd_disk = disk;
1120                bdev->bd_contains = bdev;
1121                if (!partno) {
1122                        struct backing_dev_info *bdi;
1123
1124                        ret = -ENXIO;
1125                        bdev->bd_part = disk_get_part(disk, partno);
1126                        if (!bdev->bd_part)
1127                                goto out_clear;
1128
1129                        ret = 0;
1130                        if (disk->fops->open) {
1131                                ret = disk->fops->open(bdev, mode);
1132                                if (ret == -ERESTARTSYS) {
1133                                        /* Lost a race with 'disk' being
1134                                         * deleted, try again.
1135                                         * See md.c
1136                                         */
1137                                        disk_put_part(bdev->bd_part);
1138                                        bdev->bd_part = NULL;
1139                                        bdev->bd_disk = NULL;
1140                                        mutex_unlock(&bdev->bd_mutex);
1141                                        disk_unblock_events(disk);
1142                                        put_disk(disk);
1143                                        module_put(owner);
1144                                        goto restart;
1145                                }
1146                        }
1147
1148                        if (!ret && !bdev->bd_openers) {
1149                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1150                                bdi = blk_get_backing_dev_info(bdev);
1151                                if (bdi == NULL)
1152                                        bdi = &default_backing_dev_info;
1153                                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1154                        }
1155
1156                        /*
1157                         * If the device is invalidated, rescan partition
1158                         * if open succeeded or failed with -ENOMEDIUM.
1159                         * The latter is necessary to prevent ghost
1160                         * partitions on a removed medium.
1161                         */
1162                        if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1163                                rescan_partitions(disk, bdev);
1164                        if (ret)
1165                                goto out_clear;
1166                } else {
1167                        struct block_device *whole;
1168                        whole = bdget_disk(disk, 0);
1169                        ret = -ENOMEM;
1170                        if (!whole)
1171                                goto out_clear;
1172                        BUG_ON(for_part);
1173                        ret = __blkdev_get(whole, mode, 1);
1174                        if (ret)
1175                                goto out_clear;
1176                        bdev->bd_contains = whole;
1177                        bdev_inode_switch_bdi(bdev->bd_inode,
1178                                whole->bd_inode->i_data.backing_dev_info);
1179                        bdev->bd_part = disk_get_part(disk, partno);
1180                        if (!(disk->flags & GENHD_FL_UP) ||
1181                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1182                                ret = -ENXIO;
1183                                goto out_clear;
1184                        }
1185                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1186                }
1187        } else {
1188                if (bdev->bd_contains == bdev) {
1189                        ret = 0;
1190                        if (bdev->bd_disk->fops->open)
1191                                ret = bdev->bd_disk->fops->open(bdev, mode);
1192                        /* the same as first opener case, read comment there */
1193                        if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1194                                rescan_partitions(bdev->bd_disk, bdev);
1195                        if (ret)
1196                                goto out_unlock_bdev;
1197                }
1198                /* only one opener holds refs to the module and disk */
1199                put_disk(disk);
1200                module_put(owner);
1201        }
1202        bdev->bd_openers++;
1203        if (for_part)
1204                bdev->bd_part_count++;
1205        mutex_unlock(&bdev->bd_mutex);
1206        disk_unblock_events(disk);
1207        return 0;
1208
1209 out_clear:
1210        disk_put_part(bdev->bd_part);
1211        bdev->bd_disk = NULL;
1212        bdev->bd_part = NULL;
1213        bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1214        if (bdev != bdev->bd_contains)
1215                __blkdev_put(bdev->bd_contains, mode, 1);
1216        bdev->bd_contains = NULL;
1217 out_unlock_bdev:
1218        mutex_unlock(&bdev->bd_mutex);
1219        disk_unblock_events(disk);
1220        put_disk(disk);
1221        module_put(owner);
1222 out:
1223        bdput(bdev);
1224
1225        return ret;
1226}
1227
1228/**
1229 * blkdev_get - open a block device
1230 * @bdev: block_device to open
1231 * @mode: FMODE_* mask
1232 * @holder: exclusive holder identifier
1233 *
1234 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1235 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1236 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1237 *
1238 * On success, the reference count of @bdev is unchanged.  On failure,
1239 * @bdev is put.
1240 *
1241 * CONTEXT:
1242 * Might sleep.
1243 *
1244 * RETURNS:
1245 * 0 on success, -errno on failure.
1246 */
1247int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1248{
1249        struct block_device *whole = NULL;
1250        int res;
1251
1252        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1253
1254        if ((mode & FMODE_EXCL) && holder) {
1255                whole = bd_start_claiming(bdev, holder);
1256                if (IS_ERR(whole)) {
1257                        bdput(bdev);
1258                        return PTR_ERR(whole);
1259                }
1260        }
1261
1262        res = __blkdev_get(bdev, mode, 0);
1263
1264        if (whole) {
1265                struct gendisk *disk = whole->bd_disk;
1266
1267                /* finish claiming */
1268                mutex_lock(&bdev->bd_mutex);
1269                spin_lock(&bdev_lock);
1270
1271                if (!res) {
1272                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1273                        /*
1274                         * Note that for a whole device bd_holders
1275                         * will be incremented twice, and bd_holder
1276                         * will be set to bd_may_claim before being
1277                         * set to holder
1278                         */
1279                        whole->bd_holders++;
1280                        whole->bd_holder = bd_may_claim;
1281                        bdev->bd_holders++;
1282                        bdev->bd_holder = holder;
1283                }
1284
1285                /* tell others that we're done */
1286                BUG_ON(whole->bd_claiming != holder);
1287                whole->bd_claiming = NULL;
1288                wake_up_bit(&whole->bd_claiming, 0);
1289
1290                spin_unlock(&bdev_lock);
1291
1292                /*
1293                 * Block event polling for write claims if requested.  Any
1294                 * write holder makes the write_holder state stick until
1295                 * all are released.  This is good enough and tracking
1296                 * individual writeable reference is too fragile given the
1297                 * way @mode is used in blkdev_get/put().
1298                 */
1299                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1300                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1301                        bdev->bd_write_holder = true;
1302                        disk_block_events(disk);
1303                }
1304
1305                mutex_unlock(&bdev->bd_mutex);
1306                bdput(whole);
1307        }
1308
1309        return res;
1310}
1311EXPORT_SYMBOL(blkdev_get);
1312
1313/**
1314 * blkdev_get_by_path - open a block device by name
1315 * @path: path to the block device to open
1316 * @mode: FMODE_* mask
1317 * @holder: exclusive holder identifier
1318 *
1319 * Open the blockdevice described by the device file at @path.  @mode
1320 * and @holder are identical to blkdev_get().
1321 *
1322 * On success, the returned block_device has reference count of one.
1323 *
1324 * CONTEXT:
1325 * Might sleep.
1326 *
1327 * RETURNS:
1328 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1329 */
1330struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1331                                        void *holder)
1332{
1333        struct block_device *bdev;
1334        int err;
1335
1336        bdev = lookup_bdev(path);
1337        if (IS_ERR(bdev))
1338                return bdev;
1339
1340        err = blkdev_get(bdev, mode, holder);
1341        if (err)
1342                return ERR_PTR(err);
1343
1344        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1345                blkdev_put(bdev, mode);
1346                return ERR_PTR(-EACCES);
1347        }
1348
1349        return bdev;
1350}
1351EXPORT_SYMBOL(blkdev_get_by_path);
1352
1353/**
1354 * blkdev_get_by_dev - open a block device by device number
1355 * @dev: device number of block device to open
1356 * @mode: FMODE_* mask
1357 * @holder: exclusive holder identifier
1358 *
1359 * Open the blockdevice described by device number @dev.  @mode and
1360 * @holder are identical to blkdev_get().
1361 *
1362 * Use it ONLY if you really do not have anything better - i.e. when
1363 * you are behind a truly sucky interface and all you are given is a
1364 * device number.  _Never_ to be used for internal purposes.  If you
1365 * ever need it - reconsider your API.
1366 *
1367 * On success, the returned block_device has reference count of one.
1368 *
1369 * CONTEXT:
1370 * Might sleep.
1371 *
1372 * RETURNS:
1373 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1374 */
1375struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1376{
1377        struct block_device *bdev;
1378        int err;
1379
1380        bdev = bdget(dev);
1381        if (!bdev)
1382                return ERR_PTR(-ENOMEM);
1383
1384        err = blkdev_get(bdev, mode, holder);
1385        if (err)
1386                return ERR_PTR(err);
1387
1388        return bdev;
1389}
1390EXPORT_SYMBOL(blkdev_get_by_dev);
1391
1392static int blkdev_open(struct inode * inode, struct file * filp)
1393{
1394        struct block_device *bdev;
1395
1396        /*
1397         * Preserve backwards compatibility and allow large file access
1398         * even if userspace doesn't ask for it explicitly. Some mkfs
1399         * binary needs it. We might want to drop this workaround
1400         * during an unstable branch.
1401         */
1402        filp->f_flags |= O_LARGEFILE;
1403
1404        if (filp->f_flags & O_NDELAY)
1405                filp->f_mode |= FMODE_NDELAY;
1406        if (filp->f_flags & O_EXCL)
1407                filp->f_mode |= FMODE_EXCL;
1408        if ((filp->f_flags & O_ACCMODE) == 3)
1409                filp->f_mode |= FMODE_WRITE_IOCTL;
1410
1411        bdev = bd_acquire(inode);
1412        if (bdev == NULL)
1413                return -ENOMEM;
1414
1415        filp->f_mapping = bdev->bd_inode->i_mapping;
1416
1417        return blkdev_get(bdev, filp->f_mode, filp);
1418}
1419
1420static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1421{
1422        int ret = 0;
1423        struct gendisk *disk = bdev->bd_disk;
1424        struct block_device *victim = NULL;
1425
1426        mutex_lock_nested(&bdev->bd_mutex, for_part);
1427        if (for_part)
1428                bdev->bd_part_count--;
1429
1430        if (!--bdev->bd_openers) {
1431                WARN_ON_ONCE(bdev->bd_holders);
1432                sync_blockdev(bdev);
1433                kill_bdev(bdev);
1434                /* ->release can cause the old bdi to disappear,
1435                 * so must switch it out first
1436                 */
1437                bdev_inode_switch_bdi(bdev->bd_inode,
1438                                        &default_backing_dev_info);
1439        }
1440        if (bdev->bd_contains == bdev) {
1441                if (disk->fops->release)
1442                        ret = disk->fops->release(disk, mode);
1443        }
1444        if (!bdev->bd_openers) {
1445                struct module *owner = disk->fops->owner;
1446
1447                disk_put_part(bdev->bd_part);
1448                bdev->bd_part = NULL;
1449                bdev->bd_disk = NULL;
1450                if (bdev != bdev->bd_contains)
1451                        victim = bdev->bd_contains;
1452                bdev->bd_contains = NULL;
1453
1454                put_disk(disk);
1455                module_put(owner);
1456        }
1457        mutex_unlock(&bdev->bd_mutex);
1458        bdput(bdev);
1459        if (victim)
1460                __blkdev_put(victim, mode, 1);
1461        return ret;
1462}
1463
1464int blkdev_put(struct block_device *bdev, fmode_t mode)
1465{
1466        mutex_lock(&bdev->bd_mutex);
1467
1468        if (mode & FMODE_EXCL) {
1469                bool bdev_free;
1470
1471                /*
1472                 * Release a claim on the device.  The holder fields
1473                 * are protected with bdev_lock.  bd_mutex is to
1474                 * synchronize disk_holder unlinking.
1475                 */
1476                spin_lock(&bdev_lock);
1477
1478                WARN_ON_ONCE(--bdev->bd_holders < 0);
1479                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1480
1481                /* bd_contains might point to self, check in a separate step */
1482                if ((bdev_free = !bdev->bd_holders))
1483                        bdev->bd_holder = NULL;
1484                if (!bdev->bd_contains->bd_holders)
1485                        bdev->bd_contains->bd_holder = NULL;
1486
1487                spin_unlock(&bdev_lock);
1488
1489                /*
1490                 * If this was the last claim, remove holder link and
1491                 * unblock evpoll if it was a write holder.
1492                 */
1493                if (bdev_free && bdev->bd_write_holder) {
1494                        disk_unblock_events(bdev->bd_disk);
1495                        bdev->bd_write_holder = false;
1496                }
1497        }
1498
1499        /*
1500         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1501         * event.  This is to ensure detection of media removal commanded
1502         * from userland - e.g. eject(1).
1503         */
1504        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1505
1506        mutex_unlock(&bdev->bd_mutex);
1507
1508        return __blkdev_put(bdev, mode, 0);
1509}
1510EXPORT_SYMBOL(blkdev_put);
1511
1512static int blkdev_close(struct inode * inode, struct file * filp)
1513{
1514        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1515
1516        return blkdev_put(bdev, filp->f_mode);
1517}
1518
1519static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1520{
1521        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1522        fmode_t mode = file->f_mode;
1523
1524        /*
1525         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1526         * to updated it before every ioctl.
1527         */
1528        if (file->f_flags & O_NDELAY)
1529                mode |= FMODE_NDELAY;
1530        else
1531                mode &= ~FMODE_NDELAY;
1532
1533        return blkdev_ioctl(bdev, mode, cmd, arg);
1534}
1535
1536/*
1537 * Write data to the block device.  Only intended for the block device itself
1538 * and the raw driver which basically is a fake block device.
1539 *
1540 * Does not take i_mutex for the write and thus is not for general purpose
1541 * use.
1542 */
1543ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1544                         unsigned long nr_segs, loff_t pos)
1545{
1546        struct file *file = iocb->ki_filp;
1547        ssize_t ret;
1548
1549        BUG_ON(iocb->ki_pos != pos);
1550
1551        ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1552        if (ret > 0 || ret == -EIOCBQUEUED) {
1553                ssize_t err;
1554
1555                err = generic_write_sync(file, pos, ret);
1556                if (err < 0 && ret > 0)
1557                        ret = err;
1558        }
1559        return ret;
1560}
1561EXPORT_SYMBOL_GPL(blkdev_aio_write);
1562
1563/*
1564 * Try to release a page associated with block device when the system
1565 * is under memory pressure.
1566 */
1567static int blkdev_releasepage(struct page *page, gfp_t wait)
1568{
1569        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1570
1571        if (super && super->s_op->bdev_try_to_free_page)
1572                return super->s_op->bdev_try_to_free_page(super, page, wait);
1573
1574        return try_to_free_buffers(page);
1575}
1576
1577static const struct address_space_operations def_blk_aops = {
1578        .readpage       = blkdev_readpage,
1579        .writepage      = blkdev_writepage,
1580        .write_begin    = blkdev_write_begin,
1581        .write_end      = blkdev_write_end,
1582        .writepages     = generic_writepages,
1583        .releasepage    = blkdev_releasepage,
1584        .direct_IO      = blkdev_direct_IO,
1585};
1586
1587const struct file_operations def_blk_fops = {
1588        .open           = blkdev_open,
1589        .release        = blkdev_close,
1590        .llseek         = block_llseek,
1591        .read           = do_sync_read,
1592        .write          = do_sync_write,
1593        .aio_read       = generic_file_aio_read,
1594        .aio_write      = blkdev_aio_write,
1595        .mmap           = generic_file_mmap,
1596        .fsync          = blkdev_fsync,
1597        .unlocked_ioctl = block_ioctl,
1598#ifdef CONFIG_COMPAT
1599        .compat_ioctl   = compat_blkdev_ioctl,
1600#endif
1601        .splice_read    = generic_file_splice_read,
1602        .splice_write   = generic_file_splice_write,
1603};
1604
1605int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1606{
1607        int res;
1608        mm_segment_t old_fs = get_fs();
1609        set_fs(KERNEL_DS);
1610        res = blkdev_ioctl(bdev, 0, cmd, arg);
1611        set_fs(old_fs);
1612        return res;
1613}
1614
1615EXPORT_SYMBOL(ioctl_by_bdev);
1616
1617/**
1618 * lookup_bdev  - lookup a struct block_device by name
1619 * @pathname:   special file representing the block device
1620 *
1621 * Get a reference to the blockdevice at @pathname in the current
1622 * namespace if possible and return it.  Return ERR_PTR(error)
1623 * otherwise.
1624 */
1625struct block_device *lookup_bdev(const char *pathname)
1626{
1627        struct block_device *bdev;
1628        struct inode *inode;
1629        struct path path;
1630        int error;
1631
1632        if (!pathname || !*pathname)
1633                return ERR_PTR(-EINVAL);
1634
1635        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1636        if (error)
1637                return ERR_PTR(error);
1638
1639        inode = path.dentry->d_inode;
1640        error = -ENOTBLK;
1641        if (!S_ISBLK(inode->i_mode))
1642                goto fail;
1643        error = -EACCES;
1644        if (path.mnt->mnt_flags & MNT_NODEV)
1645                goto fail;
1646        error = -ENOMEM;
1647        bdev = bd_acquire(inode);
1648        if (!bdev)
1649                goto fail;
1650out:
1651        path_put(&path);
1652        return bdev;
1653fail:
1654        bdev = ERR_PTR(error);
1655        goto out;
1656}
1657EXPORT_SYMBOL(lookup_bdev);
1658
1659int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1660{
1661        struct super_block *sb = get_super(bdev);
1662        int res = 0;
1663
1664        if (sb) {
1665                /*
1666                 * no need to lock the super, get_super holds the
1667                 * read mutex so the filesystem cannot go away
1668                 * under us (->put_super runs with the write lock
1669                 * hold).
1670                 */
1671                shrink_dcache_sb(sb);
1672                res = invalidate_inodes(sb, kill_dirty);
1673                drop_super(sb);
1674        }
1675        invalidate_bdev(bdev);
1676        return res;
1677}
1678EXPORT_SYMBOL(__invalidate_device);
1679
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.