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/smp_lock.h>
  15#include <linux/device_cgroup.h>
  16#include <linux/highmem.h>
  17#include <linux/blkdev.h>
  18#include <linux/module.h>
  19#include <linux/blkpg.h>
  20#include <linux/buffer_head.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 <asm/uaccess.h>
  28#include "internal.h"
  29
  30struct bdev_inode {
  31        struct block_device bdev;
  32        struct inode vfs_inode;
  33};
  34
  35static const struct address_space_operations def_blk_aops;
  36
  37static inline struct bdev_inode *BDEV_I(struct inode *inode)
  38{
  39        return container_of(inode, struct bdev_inode, vfs_inode);
  40}
  41
  42inline struct block_device *I_BDEV(struct inode *inode)
  43{
  44        return &BDEV_I(inode)->bdev;
  45}
  46
  47EXPORT_SYMBOL(I_BDEV);
  48
  49static sector_t max_block(struct block_device *bdev)
  50{
  51        sector_t retval = ~((sector_t)0);
  52        loff_t sz = i_size_read(bdev->bd_inode);
  53
  54        if (sz) {
  55                unsigned int size = block_size(bdev);
  56                unsigned int sizebits = blksize_bits(size);
  57                retval = (sz >> sizebits);
  58        }
  59        return retval;
  60}
  61
  62/* Kill _all_ buffers and pagecache , dirty or not.. */
  63static void kill_bdev(struct block_device *bdev)
  64{
  65        if (bdev->bd_inode->i_mapping->nrpages == 0)
  66                return;
  67        invalidate_bh_lrus();
  68        truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
  69}       
  70
  71int set_blocksize(struct block_device *bdev, int size)
  72{
  73        /* Size must be a power of two, and between 512 and PAGE_SIZE */
  74        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
  75                return -EINVAL;
  76
  77        /* Size cannot be smaller than the size supported by the device */
  78        if (size < bdev_hardsect_size(bdev))
  79                return -EINVAL;
  80
  81        /* Don't change the size if it is same as current */
  82        if (bdev->bd_block_size != size) {
  83                sync_blockdev(bdev);
  84                bdev->bd_block_size = size;
  85                bdev->bd_inode->i_blkbits = blksize_bits(size);
  86                kill_bdev(bdev);
  87        }
  88        return 0;
  89}
  90
  91EXPORT_SYMBOL(set_blocksize);
  92
  93int sb_set_blocksize(struct super_block *sb, int size)
  94{
  95        if (set_blocksize(sb->s_bdev, size))
  96                return 0;
  97        /* If we get here, we know size is power of two
  98         * and it's value is between 512 and PAGE_SIZE */
  99        sb->s_blocksize = size;
 100        sb->s_blocksize_bits = blksize_bits(size);
 101        return sb->s_blocksize;
 102}
 103
 104EXPORT_SYMBOL(sb_set_blocksize);
 105
 106int sb_min_blocksize(struct super_block *sb, int size)
 107{
 108        int minsize = bdev_hardsect_size(sb->s_bdev);
 109        if (size < minsize)
 110                size = minsize;
 111        return sb_set_blocksize(sb, size);
 112}
 113
 114EXPORT_SYMBOL(sb_min_blocksize);
 115
 116static int
 117blkdev_get_block(struct inode *inode, sector_t iblock,
 118                struct buffer_head *bh, int create)
 119{
 120        if (iblock >= max_block(I_BDEV(inode))) {
 121                if (create)
 122                        return -EIO;
 123
 124                /*
 125                 * for reads, we're just trying to fill a partial page.
 126                 * return a hole, they will have to call get_block again
 127                 * before they can fill it, and they will get -EIO at that
 128                 * time
 129                 */
 130                return 0;
 131        }
 132        bh->b_bdev = I_BDEV(inode);
 133        bh->b_blocknr = iblock;
 134        set_buffer_mapped(bh);
 135        return 0;
 136}
 137
 138static int
 139blkdev_get_blocks(struct inode *inode, sector_t iblock,
 140                struct buffer_head *bh, int create)
 141{
 142        sector_t end_block = max_block(I_BDEV(inode));
 143        unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
 144
 145        if ((iblock + max_blocks) > end_block) {
 146                max_blocks = end_block - iblock;
 147                if ((long)max_blocks <= 0) {
 148                        if (create)
 149                                return -EIO;    /* write fully beyond EOF */
 150                        /*
 151                         * It is a read which is fully beyond EOF.  We return
 152                         * a !buffer_mapped buffer
 153                         */
 154                        max_blocks = 0;
 155                }
 156        }
 157
 158        bh->b_bdev = I_BDEV(inode);
 159        bh->b_blocknr = iblock;
 160        bh->b_size = max_blocks << inode->i_blkbits;
 161        if (max_blocks)
 162                set_buffer_mapped(bh);
 163        return 0;
 164}
 165
 166static ssize_t
 167blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 168                        loff_t offset, unsigned long nr_segs)
 169{
 170        struct file *file = iocb->ki_filp;
 171        struct inode *inode = file->f_mapping->host;
 172
 173        return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
 174                                iov, offset, nr_segs, blkdev_get_blocks, NULL);
 175}
 176
 177static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 178{
 179        return block_write_full_page(page, blkdev_get_block, wbc);
 180}
 181
 182static int blkdev_readpage(struct file * file, struct page * page)
 183{
 184        return block_read_full_page(page, blkdev_get_block);
 185}
 186
 187static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 188                        loff_t pos, unsigned len, unsigned flags,
 189                        struct page **pagep, void **fsdata)
 190{
 191        *pagep = NULL;
 192        return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
 193                                blkdev_get_block);
 194}
 195
 196static int blkdev_write_end(struct file *file, struct address_space *mapping,
 197                        loff_t pos, unsigned len, unsigned copied,
 198                        struct page *page, void *fsdata)
 199{
 200        int ret;
 201        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 202
 203        unlock_page(page);
 204        page_cache_release(page);
 205
 206        return ret;
 207}
 208
 209/*
 210 * private llseek:
 211 * for a block special file file->f_path.dentry->d_inode->i_size is zero
 212 * so we compute the size by hand (just as in block_read/write above)
 213 */
 214static loff_t block_llseek(struct file *file, loff_t offset, int origin)
 215{
 216        struct inode *bd_inode = file->f_mapping->host;
 217        loff_t size;
 218        loff_t retval;
 219
 220        mutex_lock(&bd_inode->i_mutex);
 221        size = i_size_read(bd_inode);
 222
 223        switch (origin) {
 224                case 2:
 225                        offset += size;
 226                        break;
 227                case 1:
 228                        offset += file->f_pos;
 229        }
 230        retval = -EINVAL;
 231        if (offset >= 0 && offset <= size) {
 232                if (offset != file->f_pos) {
 233                        file->f_pos = offset;
 234                }
 235                retval = offset;
 236        }
 237        mutex_unlock(&bd_inode->i_mutex);
 238        return retval;
 239}
 240        
 241/*
 242 *      Filp is never NULL; the only case when ->fsync() is called with
 243 *      NULL first argument is nfsd_sync_dir() and that's not a directory.
 244 */
 245 
 246static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
 247{
 248        return sync_blockdev(I_BDEV(filp->f_mapping->host));
 249}
 250
 251/*
 252 * pseudo-fs
 253 */
 254
 255static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 256static struct kmem_cache * bdev_cachep __read_mostly;
 257
 258static struct inode *bdev_alloc_inode(struct super_block *sb)
 259{
 260        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 261        if (!ei)
 262                return NULL;
 263        return &ei->vfs_inode;
 264}
 265
 266static void bdev_destroy_inode(struct inode *inode)
 267{
 268        struct bdev_inode *bdi = BDEV_I(inode);
 269
 270        bdi->bdev.bd_inode_backing_dev_info = NULL;
 271        kmem_cache_free(bdev_cachep, bdi);
 272}
 273
 274static void init_once(void *foo)
 275{
 276        struct bdev_inode *ei = (struct bdev_inode *) foo;
 277        struct block_device *bdev = &ei->bdev;
 278
 279        memset(bdev, 0, sizeof(*bdev));
 280        mutex_init(&bdev->bd_mutex);
 281        sema_init(&bdev->bd_mount_sem, 1);
 282        INIT_LIST_HEAD(&bdev->bd_inodes);
 283        INIT_LIST_HEAD(&bdev->bd_list);
 284#ifdef CONFIG_SYSFS
 285        INIT_LIST_HEAD(&bdev->bd_holder_list);
 286#endif
 287        inode_init_once(&ei->vfs_inode);
 288}
 289
 290static inline void __bd_forget(struct inode *inode)
 291{
 292        list_del_init(&inode->i_devices);
 293        inode->i_bdev = NULL;
 294        inode->i_mapping = &inode->i_data;
 295}
 296
 297static void bdev_clear_inode(struct inode *inode)
 298{
 299        struct block_device *bdev = &BDEV_I(inode)->bdev;
 300        struct list_head *p;
 301        spin_lock(&bdev_lock);
 302        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 303                __bd_forget(list_entry(p, struct inode, i_devices));
 304        }
 305        list_del_init(&bdev->bd_list);
 306        spin_unlock(&bdev_lock);
 307}
 308
 309static const struct super_operations bdev_sops = {
 310        .statfs = simple_statfs,
 311        .alloc_inode = bdev_alloc_inode,
 312        .destroy_inode = bdev_destroy_inode,
 313        .drop_inode = generic_delete_inode,
 314        .clear_inode = bdev_clear_inode,
 315};
 316
 317static int bd_get_sb(struct file_system_type *fs_type,
 318        int flags, const char *dev_name, void *data, struct vfsmount *mnt)
 319{
 320        return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
 321}
 322
 323static struct file_system_type bd_type = {
 324        .name           = "bdev",
 325        .get_sb         = bd_get_sb,
 326        .kill_sb        = kill_anon_super,
 327};
 328
 329static struct vfsmount *bd_mnt __read_mostly;
 330struct super_block *blockdev_superblock;
 331
 332void __init bdev_cache_init(void)
 333{
 334        int err;
 335        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 336                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 337                                SLAB_MEM_SPREAD|SLAB_PANIC),
 338                        init_once);
 339        err = register_filesystem(&bd_type);
 340        if (err)
 341                panic("Cannot register bdev pseudo-fs");
 342        bd_mnt = kern_mount(&bd_type);
 343        if (IS_ERR(bd_mnt))
 344                panic("Cannot create bdev pseudo-fs");
 345        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 346}
 347
 348/*
 349 * Most likely _very_ bad one - but then it's hardly critical for small
 350 * /dev and can be fixed when somebody will need really large one.
 351 * Keep in mind that it will be fed through icache hash function too.
 352 */
 353static inline unsigned long hash(dev_t dev)
 354{
 355        return MAJOR(dev)+MINOR(dev);
 356}
 357
 358static int bdev_test(struct inode *inode, void *data)
 359{
 360        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 361}
 362
 363static int bdev_set(struct inode *inode, void *data)
 364{
 365        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 366        return 0;
 367}
 368
 369static LIST_HEAD(all_bdevs);
 370
 371struct block_device *bdget(dev_t dev)
 372{
 373        struct block_device *bdev;
 374        struct inode *inode;
 375
 376        inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
 377                        bdev_test, bdev_set, &dev);
 378
 379        if (!inode)
 380                return NULL;
 381
 382        bdev = &BDEV_I(inode)->bdev;
 383
 384        if (inode->i_state & I_NEW) {
 385                bdev->bd_contains = NULL;
 386                bdev->bd_inode = inode;
 387                bdev->bd_block_size = (1 << inode->i_blkbits);
 388                bdev->bd_part_count = 0;
 389                bdev->bd_invalidated = 0;
 390                inode->i_mode = S_IFBLK;
 391                inode->i_rdev = dev;
 392                inode->i_bdev = bdev;
 393                inode->i_data.a_ops = &def_blk_aops;
 394                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 395                inode->i_data.backing_dev_info = &default_backing_dev_info;
 396                spin_lock(&bdev_lock);
 397                list_add(&bdev->bd_list, &all_bdevs);
 398                spin_unlock(&bdev_lock);
 399                unlock_new_inode(inode);
 400        }
 401        return bdev;
 402}
 403
 404EXPORT_SYMBOL(bdget);
 405
 406long nr_blockdev_pages(void)
 407{
 408        struct block_device *bdev;
 409        long ret = 0;
 410        spin_lock(&bdev_lock);
 411        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 412                ret += bdev->bd_inode->i_mapping->nrpages;
 413        }
 414        spin_unlock(&bdev_lock);
 415        return ret;
 416}
 417
 418void bdput(struct block_device *bdev)
 419{
 420        iput(bdev->bd_inode);
 421}
 422
 423EXPORT_SYMBOL(bdput);
 424 
 425static struct block_device *bd_acquire(struct inode *inode)
 426{
 427        struct block_device *bdev;
 428
 429        spin_lock(&bdev_lock);
 430        bdev = inode->i_bdev;
 431        if (bdev) {
 432                atomic_inc(&bdev->bd_inode->i_count);
 433                spin_unlock(&bdev_lock);
 434                return bdev;
 435        }
 436        spin_unlock(&bdev_lock);
 437
 438        bdev = bdget(inode->i_rdev);
 439        if (bdev) {
 440                spin_lock(&bdev_lock);
 441                if (!inode->i_bdev) {
 442                        /*
 443                         * We take an additional bd_inode->i_count for inode,
 444                         * and it's released in clear_inode() of inode.
 445                         * So, we can access it via ->i_mapping always
 446                         * without igrab().
 447                         */
 448                        atomic_inc(&bdev->bd_inode->i_count);
 449                        inode->i_bdev = bdev;
 450                        inode->i_mapping = bdev->bd_inode->i_mapping;
 451                        list_add(&inode->i_devices, &bdev->bd_inodes);
 452                }
 453                spin_unlock(&bdev_lock);
 454        }
 455        return bdev;
 456}
 457
 458/* Call when you free inode */
 459
 460void bd_forget(struct inode *inode)
 461{
 462        struct block_device *bdev = NULL;
 463
 464        spin_lock(&bdev_lock);
 465        if (inode->i_bdev) {
 466                if (inode->i_sb != blockdev_superblock)
 467                        bdev = inode->i_bdev;
 468                __bd_forget(inode);
 469        }
 470        spin_unlock(&bdev_lock);
 471
 472        if (bdev)
 473                iput(bdev->bd_inode);
 474}
 475
 476int bd_claim(struct block_device *bdev, void *holder)
 477{
 478        int res;
 479        spin_lock(&bdev_lock);
 480
 481        /* first decide result */
 482        if (bdev->bd_holder == holder)
 483                res = 0;         /* already a holder */
 484        else if (bdev->bd_holder != NULL)
 485                res = -EBUSY;    /* held by someone else */
 486        else if (bdev->bd_contains == bdev)
 487                res = 0;         /* is a whole device which isn't held */
 488
 489        else if (bdev->bd_contains->bd_holder == bd_claim)
 490                res = 0;         /* is a partition of a device that is being partitioned */
 491        else if (bdev->bd_contains->bd_holder != NULL)
 492                res = -EBUSY;    /* is a partition of a held device */
 493        else
 494                res = 0;         /* is a partition of an un-held device */
 495
 496        /* now impose change */
 497        if (res==0) {
 498                /* note that for a whole device bd_holders
 499                 * will be incremented twice, and bd_holder will
 500                 * be set to bd_claim before being set to holder
 501                 */
 502                bdev->bd_contains->bd_holders ++;
 503                bdev->bd_contains->bd_holder = bd_claim;
 504                bdev->bd_holders++;
 505                bdev->bd_holder = holder;
 506        }
 507        spin_unlock(&bdev_lock);
 508        return res;
 509}
 510
 511EXPORT_SYMBOL(bd_claim);
 512
 513void bd_release(struct block_device *bdev)
 514{
 515        spin_lock(&bdev_lock);
 516        if (!--bdev->bd_contains->bd_holders)
 517                bdev->bd_contains->bd_holder = NULL;
 518        if (!--bdev->bd_holders)
 519                bdev->bd_holder = NULL;
 520        spin_unlock(&bdev_lock);
 521}
 522
 523EXPORT_SYMBOL(bd_release);
 524
 525#ifdef CONFIG_SYSFS
 526/*
 527 * Functions for bd_claim_by_kobject / bd_release_from_kobject
 528 *
 529 *     If a kobject is passed to bd_claim_by_kobject()
 530 *     and the kobject has a parent directory,
 531 *     following symlinks are created:
 532 *        o from the kobject to the claimed bdev
 533 *        o from "holders" directory of the bdev to the parent of the kobject
 534 *     bd_release_from_kobject() removes these symlinks.
 535 *
 536 *     Example:
 537 *        If /dev/dm-0 maps to /dev/sda, kobject corresponding to
 538 *        /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
 539 *           /sys/block/dm-0/slaves/sda --> /sys/block/sda
 540 *           /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 541 */
 542
 543static int add_symlink(struct kobject *from, struct kobject *to)
 544{
 545        if (!from || !to)
 546                return 0;
 547        return sysfs_create_link(from, to, kobject_name(to));
 548}
 549
 550static void del_symlink(struct kobject *from, struct kobject *to)
 551{
 552        if (!from || !to)
 553                return;
 554        sysfs_remove_link(from, kobject_name(to));
 555}
 556
 557/*
 558 * 'struct bd_holder' contains pointers to kobjects symlinked by
 559 * bd_claim_by_kobject.
 560 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
 561 */
 562struct bd_holder {
 563        struct list_head list;  /* chain of holders of the bdev */
 564        int count;              /* references from the holder */
 565        struct kobject *sdir;   /* holder object, e.g. "/block/dm-0/slaves" */
 566        struct kobject *hdev;   /* e.g. "/block/dm-0" */
 567        struct kobject *hdir;   /* e.g. "/block/sda/holders" */
 568        struct kobject *sdev;   /* e.g. "/block/sda" */
 569};
 570
 571/*
 572 * Get references of related kobjects at once.
 573 * Returns 1 on success. 0 on failure.
 574 *
 575 * Should call bd_holder_release_dirs() after successful use.
 576 */
 577static int bd_holder_grab_dirs(struct block_device *bdev,
 578                        struct bd_holder *bo)
 579{
 580        if (!bdev || !bo)
 581                return 0;
 582
 583        bo->sdir = kobject_get(bo->sdir);
 584        if (!bo->sdir)
 585                return 0;
 586
 587        bo->hdev = kobject_get(bo->sdir->parent);
 588        if (!bo->hdev)
 589                goto fail_put_sdir;
 590
 591        bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
 592        if (!bo->sdev)
 593                goto fail_put_hdev;
 594
 595        bo->hdir = kobject_get(bdev->bd_part->holder_dir);
 596        if (!bo->hdir)
 597                goto fail_put_sdev;
 598
 599        return 1;
 600
 601fail_put_sdev:
 602        kobject_put(bo->sdev);
 603fail_put_hdev:
 604        kobject_put(bo->hdev);
 605fail_put_sdir:
 606        kobject_put(bo->sdir);
 607
 608        return 0;
 609}
 610
 611/* Put references of related kobjects at once. */
 612static void bd_holder_release_dirs(struct bd_holder *bo)
 613{
 614        kobject_put(bo->hdir);
 615        kobject_put(bo->sdev);
 616        kobject_put(bo->hdev);
 617        kobject_put(bo->sdir);
 618}
 619
 620static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
 621{
 622        struct bd_holder *bo;
 623
 624        bo = kzalloc(sizeof(*bo), GFP_KERNEL);
 625        if (!bo)
 626                return NULL;
 627
 628        bo->count = 1;
 629        bo->sdir = kobj;
 630
 631        return bo;
 632}
 633
 634static void free_bd_holder(struct bd_holder *bo)
 635{
 636        kfree(bo);
 637}
 638
 639/**
 640 * find_bd_holder - find matching struct bd_holder from the block device
 641 *
 642 * @bdev:       struct block device to be searched
 643 * @bo:         target struct bd_holder
 644 *
 645 * Returns matching entry with @bo in @bdev->bd_holder_list.
 646 * If found, increment the reference count and return the pointer.
 647 * If not found, returns NULL.
 648 */
 649static struct bd_holder *find_bd_holder(struct block_device *bdev,
 650                                        struct bd_holder *bo)
 651{
 652        struct bd_holder *tmp;
 653
 654        list_for_each_entry(tmp, &bdev->bd_holder_list, list)
 655                if (tmp->sdir == bo->sdir) {
 656                        tmp->count++;
 657                        return tmp;
 658                }
 659
 660        return NULL;
 661}
 662
 663/**
 664 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
 665 *
 666 * @bdev:       block device to be bd_claimed
 667 * @bo:         preallocated and initialized by alloc_bd_holder()
 668 *
 669 * Add @bo to @bdev->bd_holder_list, create symlinks.
 670 *
 671 * Returns 0 if symlinks are created.
 672 * Returns -ve if something fails.
 673 */
 674static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
 675{
 676        int err;
 677
 678        if (!bo)
 679                return -EINVAL;
 680
 681        if (!bd_holder_grab_dirs(bdev, bo))
 682                return -EBUSY;
 683
 684        err = add_symlink(bo->sdir, bo->sdev);
 685        if (err)
 686                return err;
 687
 688        err = add_symlink(bo->hdir, bo->hdev);
 689        if (err) {
 690                del_symlink(bo->sdir, bo->sdev);
 691                return err;
 692        }
 693
 694        list_add_tail(&bo->list, &bdev->bd_holder_list);
 695        return 0;
 696}
 697
 698/**
 699 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
 700 *
 701 * @bdev:       block device to be bd_claimed
 702 * @kobj:       holder's kobject
 703 *
 704 * If there is matching entry with @kobj in @bdev->bd_holder_list
 705 * and no other bd_claim() from the same kobject,
 706 * remove the struct bd_holder from the list, delete symlinks for it.
 707 *
 708 * Returns a pointer to the struct bd_holder when it's removed from the list
 709 * and ready to be freed.
 710 * Returns NULL if matching claim isn't found or there is other bd_claim()
 711 * by the same kobject.
 712 */
 713static struct bd_holder *del_bd_holder(struct block_device *bdev,
 714                                        struct kobject *kobj)
 715{
 716        struct bd_holder *bo;
 717
 718        list_for_each_entry(bo, &bdev->bd_holder_list, list) {
 719                if (bo->sdir == kobj) {
 720                        bo->count--;
 721                        BUG_ON(bo->count < 0);
 722                        if (!bo->count) {
 723                                list_del(&bo->list);
 724                                del_symlink(bo->sdir, bo->sdev);
 725                                del_symlink(bo->hdir, bo->hdev);
 726                                bd_holder_release_dirs(bo);
 727                                return bo;
 728                        }
 729                        break;
 730                }
 731        }
 732
 733        return NULL;
 734}
 735
 736/**
 737 * bd_claim_by_kobject - bd_claim() with additional kobject signature
 738 *
 739 * @bdev:       block device to be claimed
 740 * @holder:     holder's signature
 741 * @kobj:       holder's kobject
 742 *
 743 * Do bd_claim() and if it succeeds, create sysfs symlinks between
 744 * the bdev and the holder's kobject.
 745 * Use bd_release_from_kobject() when relesing the claimed bdev.
 746 *
 747 * Returns 0 on success. (same as bd_claim())
 748 * Returns errno on failure.
 749 */
 750static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
 751                                struct kobject *kobj)
 752{
 753        int err;
 754        struct bd_holder *bo, *found;
 755
 756        if (!kobj)
 757                return -EINVAL;
 758
 759        bo = alloc_bd_holder(kobj);
 760        if (!bo)
 761                return -ENOMEM;
 762
 763        mutex_lock(&bdev->bd_mutex);
 764
 765        err = bd_claim(bdev, holder);
 766        if (err)
 767                goto fail;
 768
 769        found = find_bd_holder(bdev, bo);
 770        if (found)
 771                goto fail;
 772
 773        err = add_bd_holder(bdev, bo);
 774        if (err)
 775                bd_release(bdev);
 776        else
 777                bo = NULL;
 778fail:
 779        mutex_unlock(&bdev->bd_mutex);
 780        free_bd_holder(bo);
 781        return err;
 782}
 783
 784/**
 785 * bd_release_from_kobject - bd_release() with additional kobject signature
 786 *
 787 * @bdev:       block device to be released
 788 * @kobj:       holder's kobject
 789 *
 790 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
 791 */
 792static void bd_release_from_kobject(struct block_device *bdev,
 793                                        struct kobject *kobj)
 794{
 795        if (!kobj)
 796                return;
 797
 798        mutex_lock(&bdev->bd_mutex);
 799        bd_release(bdev);
 800        free_bd_holder(del_bd_holder(bdev, kobj));
 801        mutex_unlock(&bdev->bd_mutex);
 802}
 803
 804/**
 805 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
 806 *
 807 * @bdev:       block device to be claimed
 808 * @holder:     holder's signature
 809 * @disk:       holder's gendisk
 810 *
 811 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
 812 */
 813int bd_claim_by_disk(struct block_device *bdev, void *holder,
 814                        struct gendisk *disk)
 815{
 816        return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
 817}
 818EXPORT_SYMBOL_GPL(bd_claim_by_disk);
 819
 820/**
 821 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
 822 *
 823 * @bdev:       block device to be claimed
 824 * @disk:       holder's gendisk
 825 *
 826 * Call bd_release_from_kobject() and put @disk->slave_dir.
 827 */
 828void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
 829{
 830        bd_release_from_kobject(bdev, disk->slave_dir);
 831        kobject_put(disk->slave_dir);
 832}
 833EXPORT_SYMBOL_GPL(bd_release_from_disk);
 834#endif
 835
 836/*
 837 * Tries to open block device by device number.  Use it ONLY if you
 838 * really do not have anything better - i.e. when you are behind a
 839 * truly sucky interface and all you are given is a device number.  _Never_
 840 * to be used for internal purposes.  If you ever need it - reconsider
 841 * your API.
 842 */
 843struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
 844{
 845        struct block_device *bdev = bdget(dev);
 846        int err = -ENOMEM;
 847        if (bdev)
 848                err = blkdev_get(bdev, mode);
 849        return err ? ERR_PTR(err) : bdev;
 850}
 851
 852EXPORT_SYMBOL(open_by_devnum);
 853
 854/**
 855 * flush_disk - invalidates all buffer-cache entries on a disk
 856 *
 857 * @bdev:      struct block device to be flushed
 858 *
 859 * Invalidates all buffer-cache entries on a disk. It should be called
 860 * when a disk has been changed -- either by a media change or online
 861 * resize.
 862 */
 863static void flush_disk(struct block_device *bdev)
 864{
 865        if (__invalidate_device(bdev)) {
 866                char name[BDEVNAME_SIZE] = "";
 867
 868                if (bdev->bd_disk)
 869                        disk_name(bdev->bd_disk, 0, name);
 870                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 871                       "resized disk %s\n", name);
 872        }
 873
 874        if (!bdev->bd_disk)
 875                return;
 876        if (disk_partitionable(bdev->bd_disk))
 877                bdev->bd_invalidated = 1;
 878}
 879
 880/**
 881 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 882 * @disk: struct gendisk to check
 883 * @bdev: struct bdev to adjust.
 884 *
 885 * This routine checks to see if the bdev size does not match the disk size
 886 * and adjusts it if it differs.
 887 */
 888void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 889{
 890        loff_t disk_size, bdev_size;
 891
 892        disk_size = (loff_t)get_capacity(disk) << 9;
 893        bdev_size = i_size_read(bdev->bd_inode);
 894        if (disk_size != bdev_size) {
 895                char name[BDEVNAME_SIZE];
 896
 897                disk_name(disk, 0, name);
 898                printk(KERN_INFO
 899                       "%s: detected capacity change from %lld to %lld\n",
 900                       name, bdev_size, disk_size);
 901                i_size_write(bdev->bd_inode, disk_size);
 902                flush_disk(bdev);
 903        }
 904}
 905EXPORT_SYMBOL(check_disk_size_change);
 906
 907/**
 908 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 909 * @disk: struct gendisk to be revalidated
 910 *
 911 * This routine is a wrapper for lower-level driver's revalidate_disk
 912 * call-backs.  It is used to do common pre and post operations needed
 913 * for all revalidate_disk operations.
 914 */
 915int revalidate_disk(struct gendisk *disk)
 916{
 917        struct block_device *bdev;
 918        int ret = 0;
 919
 920        if (disk->fops->revalidate_disk)
 921                ret = disk->fops->revalidate_disk(disk);
 922
 923        bdev = bdget_disk(disk, 0);
 924        if (!bdev)
 925                return ret;
 926
 927        mutex_lock(&bdev->bd_mutex);
 928        check_disk_size_change(disk, bdev);
 929        mutex_unlock(&bdev->bd_mutex);
 930        bdput(bdev);
 931        return ret;
 932}
 933EXPORT_SYMBOL(revalidate_disk);
 934
 935/*
 936 * This routine checks whether a removable media has been changed,
 937 * and invalidates all buffer-cache-entries in that case. This
 938 * is a relatively slow routine, so we have to try to minimize using
 939 * it. Thus it is called only upon a 'mount' or 'open'. This
 940 * is the best way of combining speed and utility, I think.
 941 * People changing diskettes in the middle of an operation deserve
 942 * to lose :-)
 943 */
 944int check_disk_change(struct block_device *bdev)
 945{
 946        struct gendisk *disk = bdev->bd_disk;
 947        struct block_device_operations * bdops = disk->fops;
 948
 949        if (!bdops->media_changed)
 950                return 0;
 951        if (!bdops->media_changed(bdev->bd_disk))
 952                return 0;
 953
 954        flush_disk(bdev);
 955        if (bdops->revalidate_disk)
 956                bdops->revalidate_disk(bdev->bd_disk);
 957        return 1;
 958}
 959
 960EXPORT_SYMBOL(check_disk_change);
 961
 962void bd_set_size(struct block_device *bdev, loff_t size)
 963{
 964        unsigned bsize = bdev_hardsect_size(bdev);
 965
 966        bdev->bd_inode->i_size = size;
 967        while (bsize < PAGE_CACHE_SIZE) {
 968                if (size & bsize)
 969                        break;
 970                bsize <<= 1;
 971        }
 972        bdev->bd_block_size = bsize;
 973        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
 974}
 975EXPORT_SYMBOL(bd_set_size);
 976
 977static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
 978
 979/*
 980 * bd_mutex locking:
 981 *
 982 *  mutex_lock(part->bd_mutex)
 983 *    mutex_lock_nested(whole->bd_mutex, 1)
 984 */
 985
 986static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
 987{
 988        struct gendisk *disk;
 989        int ret;
 990        int partno;
 991        int perm = 0;
 992
 993        if (mode & FMODE_READ)
 994                perm |= MAY_READ;
 995        if (mode & FMODE_WRITE)
 996                perm |= MAY_WRITE;
 997        /*
 998         * hooks: /n/, see "layering violations".
 999         */
1000        ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1001        if (ret != 0) {
1002                bdput(bdev);
1003                return ret;
1004        }
1005
1006        lock_kernel();
1007
1008        ret = -ENXIO;
1009        disk = get_gendisk(bdev->bd_dev, &partno);
1010        if (!disk)
1011                goto out_unlock_kernel;
1012
1013        mutex_lock_nested(&bdev->bd_mutex, for_part);
1014        if (!bdev->bd_openers) {
1015                bdev->bd_disk = disk;
1016                bdev->bd_contains = bdev;
1017                if (!partno) {
1018                        struct backing_dev_info *bdi;
1019
1020                        ret = -ENXIO;
1021                        bdev->bd_part = disk_get_part(disk, partno);
1022                        if (!bdev->bd_part)
1023                                goto out_clear;
1024
1025                        if (disk->fops->open) {
1026                                ret = disk->fops->open(bdev, mode);
1027                                if (ret)
1028                                        goto out_clear;
1029                        }
1030                        if (!bdev->bd_openers) {
1031                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1032                                bdi = blk_get_backing_dev_info(bdev);
1033                                if (bdi == NULL)
1034                                        bdi = &default_backing_dev_info;
1035                                bdev->bd_inode->i_data.backing_dev_info = bdi;
1036                        }
1037                        if (bdev->bd_invalidated)
1038                                rescan_partitions(disk, bdev);
1039                } else {
1040                        struct block_device *whole;
1041                        whole = bdget_disk(disk, 0);
1042                        ret = -ENOMEM;
1043                        if (!whole)
1044                                goto out_clear;
1045                        BUG_ON(for_part);
1046                        ret = __blkdev_get(whole, mode, 1);
1047                        if (ret)
1048                                goto out_clear;
1049                        bdev->bd_contains = whole;
1050                        bdev->bd_inode->i_data.backing_dev_info =
1051                           whole->bd_inode->i_data.backing_dev_info;
1052                        bdev->bd_part = disk_get_part(disk, partno);
1053                        if (!(disk->flags & GENHD_FL_UP) ||
1054                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1055                                ret = -ENXIO;
1056                                goto out_clear;
1057                        }
1058                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1059                }
1060        } else {
1061                put_disk(disk);
1062                module_put(disk->fops->owner);
1063                disk = NULL;
1064                if (bdev->bd_contains == bdev) {
1065                        if (bdev->bd_disk->fops->open) {
1066                                ret = bdev->bd_disk->fops->open(bdev, mode);
1067                                if (ret)
1068                                        goto out_unlock_bdev;
1069                        }
1070                        if (bdev->bd_invalidated)
1071                                rescan_partitions(bdev->bd_disk, bdev);
1072                }
1073        }
1074        bdev->bd_openers++;
1075        if (for_part)
1076                bdev->bd_part_count++;
1077        mutex_unlock(&bdev->bd_mutex);
1078        unlock_kernel();
1079        return 0;
1080
1081 out_clear:
1082        disk_put_part(bdev->bd_part);
1083        bdev->bd_disk = NULL;
1084        bdev->bd_part = NULL;
1085        bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1086        if (bdev != bdev->bd_contains)
1087                __blkdev_put(bdev->bd_contains, mode, 1);
1088        bdev->bd_contains = NULL;
1089 out_unlock_bdev:
1090        mutex_unlock(&bdev->bd_mutex);
1091 out_unlock_kernel:
1092        unlock_kernel();
1093
1094        if (disk)
1095                module_put(disk->fops->owner);
1096        put_disk(disk);
1097        bdput(bdev);
1098
1099        return ret;
1100}
1101
1102int blkdev_get(struct block_device *bdev, fmode_t mode)
1103{
1104        return __blkdev_get(bdev, mode, 0);
1105}
1106EXPORT_SYMBOL(blkdev_get);
1107
1108static int blkdev_open(struct inode * inode, struct file * filp)
1109{
1110        struct block_device *bdev;
1111        int res;
1112
1113        /*
1114         * Preserve backwards compatibility and allow large file access
1115         * even if userspace doesn't ask for it explicitly. Some mkfs
1116         * binary needs it. We might want to drop this workaround
1117         * during an unstable branch.
1118         */
1119        filp->f_flags |= O_LARGEFILE;
1120
1121        if (filp->f_flags & O_NDELAY)
1122                filp->f_mode |= FMODE_NDELAY;
1123        if (filp->f_flags & O_EXCL)
1124                filp->f_mode |= FMODE_EXCL;
1125        if ((filp->f_flags & O_ACCMODE) == 3)
1126                filp->f_mode |= FMODE_WRITE_IOCTL;
1127
1128        bdev = bd_acquire(inode);
1129        if (bdev == NULL)
1130                return -ENOMEM;
1131
1132        filp->f_mapping = bdev->bd_inode->i_mapping;
1133
1134        res = blkdev_get(bdev, filp->f_mode);
1135        if (res)
1136                return res;
1137
1138        if (filp->f_mode & FMODE_EXCL) {
1139                res = bd_claim(bdev, filp);
1140                if (res)
1141                        goto out_blkdev_put;
1142        }
1143
1144        return 0;
1145
1146 out_blkdev_put:
1147        blkdev_put(bdev, filp->f_mode);
1148        return res;
1149}
1150
1151static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1152{
1153        int ret = 0;
1154        struct gendisk *disk = bdev->bd_disk;
1155        struct block_device *victim = NULL;
1156
1157        mutex_lock_nested(&bdev->bd_mutex, for_part);
1158        lock_kernel();
1159        if (for_part)
1160                bdev->bd_part_count--;
1161
1162        if (!--bdev->bd_openers) {
1163                sync_blockdev(bdev);
1164                kill_bdev(bdev);
1165        }
1166        if (bdev->bd_contains == bdev) {
1167                if (disk->fops->release)
1168                        ret = disk->fops->release(disk, mode);
1169        }
1170        if (!bdev->bd_openers) {
1171                struct module *owner = disk->fops->owner;
1172
1173                put_disk(disk);
1174                module_put(owner);
1175                disk_put_part(bdev->bd_part);
1176                bdev->bd_part = NULL;
1177                bdev->bd_disk = NULL;
1178                bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1179                if (bdev != bdev->bd_contains)
1180                        victim = bdev->bd_contains;
1181                bdev->bd_contains = NULL;
1182        }
1183        unlock_kernel();
1184        mutex_unlock(&bdev->bd_mutex);
1185        bdput(bdev);
1186        if (victim)
1187                __blkdev_put(victim, mode, 1);
1188        return ret;
1189}
1190
1191int blkdev_put(struct block_device *bdev, fmode_t mode)
1192{
1193        return __blkdev_put(bdev, mode, 0);
1194}
1195EXPORT_SYMBOL(blkdev_put);
1196
1197static int blkdev_close(struct inode * inode, struct file * filp)
1198{
1199        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1200        if (bdev->bd_holder == filp)
1201                bd_release(bdev);
1202        return blkdev_put(bdev, filp->f_mode);
1203}
1204
1205static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1206{
1207        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1208        fmode_t mode = file->f_mode;
1209
1210        /*
1211         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1212         * to updated it before every ioctl.
1213         */
1214        if (file->f_flags & O_NDELAY)
1215                mode |= FMODE_NDELAY;
1216        else
1217                mode &= ~FMODE_NDELAY;
1218
1219        return blkdev_ioctl(bdev, mode, cmd, arg);
1220}
1221
1222static const struct address_space_operations def_blk_aops = {
1223        .readpage       = blkdev_readpage,
1224        .writepage      = blkdev_writepage,
1225        .sync_page      = block_sync_page,
1226        .write_begin    = blkdev_write_begin,
1227        .write_end      = blkdev_write_end,
1228        .writepages     = generic_writepages,
1229        .direct_IO      = blkdev_direct_IO,
1230};
1231
1232const struct file_operations def_blk_fops = {
1233        .open           = blkdev_open,
1234        .release        = blkdev_close,
1235        .llseek         = block_llseek,
1236        .read           = do_sync_read,
1237        .write          = do_sync_write,
1238        .aio_read       = generic_file_aio_read,
1239        .aio_write      = generic_file_aio_write_nolock,
1240        .mmap           = generic_file_mmap,
1241        .fsync          = block_fsync,
1242        .unlocked_ioctl = block_ioctl,
1243#ifdef CONFIG_COMPAT
1244        .compat_ioctl   = compat_blkdev_ioctl,
1245#endif
1246        .splice_read    = generic_file_splice_read,
1247        .splice_write   = generic_file_splice_write,
1248};
1249
1250int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1251{
1252        int res;
1253        mm_segment_t old_fs = get_fs();
1254        set_fs(KERNEL_DS);
1255        res = blkdev_ioctl(bdev, 0, cmd, arg);
1256        set_fs(old_fs);
1257        return res;
1258}
1259
1260EXPORT_SYMBOL(ioctl_by_bdev);
1261
1262/**
1263 * lookup_bdev  - lookup a struct block_device by name
1264 * @path:       special file representing the block device
1265 *
1266 * Get a reference to the blockdevice at @pathname in the current
1267 * namespace if possible and return it.  Return ERR_PTR(error)
1268 * otherwise.
1269 */
1270struct block_device *lookup_bdev(const char *pathname)
1271{
1272        struct block_device *bdev;
1273        struct inode *inode;
1274        struct path path;
1275        int error;
1276
1277        if (!pathname || !*pathname)
1278                return ERR_PTR(-EINVAL);
1279
1280        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1281        if (error)
1282                return ERR_PTR(error);
1283
1284        inode = path.dentry->d_inode;
1285        error = -ENOTBLK;
1286        if (!S_ISBLK(inode->i_mode))
1287                goto fail;
1288        error = -EACCES;
1289        if (path.mnt->mnt_flags & MNT_NODEV)
1290                goto fail;
1291        error = -ENOMEM;
1292        bdev = bd_acquire(inode);
1293        if (!bdev)
1294                goto fail;
1295out:
1296        path_put(&path);
1297        return bdev;
1298fail:
1299        bdev = ERR_PTR(error);
1300        goto out;
1301}
1302EXPORT_SYMBOL(lookup_bdev);
1303
1304/**
1305 * open_bdev_exclusive  -  open a block device by name and set it up for use
1306 *
1307 * @path:       special file representing the block device
1308 * @mode:       FMODE_... combination to pass be used
1309 * @holder:     owner for exclusion
1310 *
1311 * Open the blockdevice described by the special file at @path, claim it
1312 * for the @holder.
1313 */
1314struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1315{
1316        struct block_device *bdev;
1317        int error = 0;
1318
1319        bdev = lookup_bdev(path);
1320        if (IS_ERR(bdev))
1321                return bdev;
1322
1323        error = blkdev_get(bdev, mode);
1324        if (error)
1325                return ERR_PTR(error);
1326        error = -EACCES;
1327        if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1328                goto blkdev_put;
1329        error = bd_claim(bdev, holder);
1330        if (error)
1331                goto blkdev_put;
1332
1333        return bdev;
1334        
1335blkdev_put:
1336        blkdev_put(bdev, mode);
1337        return ERR_PTR(error);
1338}
1339
1340EXPORT_SYMBOL(open_bdev_exclusive);
1341
1342/**
1343 * close_bdev_exclusive  -  close a blockdevice opened by open_bdev_exclusive()
1344 *
1345 * @bdev:       blockdevice to close
1346 * @mode:       mode, must match that used to open.
1347 *
1348 * This is the counterpart to open_bdev_exclusive().
1349 */
1350void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1351{
1352        bd_release(bdev);
1353        blkdev_put(bdev, mode);
1354}
1355
1356EXPORT_SYMBOL(close_bdev_exclusive);
1357
1358int __invalidate_device(struct block_device *bdev)
1359{
1360        struct super_block *sb = get_super(bdev);
1361        int res = 0;
1362
1363        if (sb) {
1364                /*
1365                 * no need to lock the super, get_super holds the
1366                 * read mutex so the filesystem cannot go away
1367                 * under us (->put_super runs with the write lock
1368                 * hold).
1369                 */
1370                shrink_dcache_sb(sb);
1371                res = invalidate_inodes(sb);
1372                drop_super(sb);
1373        }
1374        invalidate_bdev(bdev);
1375        return res;
1376}
1377EXPORT_SYMBOL(__invalidate_device);
1378