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/magic.h>
  20#include <linux/buffer_head.h>
  21#include <linux/swap.h>
  22#include <linux/pagevec.h>
  23#include <linux/writeback.h>
  24#include <linux/mpage.h>
  25#include <linux/mount.h>
  26#include <linux/uio.h>
  27#include <linux/namei.h>
  28#include <linux/log2.h>
  29#include <linux/cleancache.h>
  30#include <linux/aio.h>
  31#include <asm/uaccess.h>
  32#include "internal.h"
  33
  34struct bdev_inode {
  35        struct block_device bdev;
  36        struct inode vfs_inode;
  37};
  38
  39static const struct address_space_operations def_blk_aops;
  40
  41static inline struct bdev_inode *BDEV_I(struct inode *inode)
  42{
  43        return container_of(inode, struct bdev_inode, vfs_inode);
  44}
  45
  46inline struct block_device *I_BDEV(struct inode *inode)
  47{
  48        return &BDEV_I(inode)->bdev;
  49}
  50EXPORT_SYMBOL(I_BDEV);
  51
  52/*
  53 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
  54 * need to move it onto the dirty list of @dst so that the inode is always on
  55 * the right list.
  56 */
  57static void bdev_inode_switch_bdi(struct inode *inode,
  58                        struct backing_dev_info *dst)
  59{
  60        struct backing_dev_info *old = inode->i_data.backing_dev_info;
  61        bool wakeup_bdi = false;
  62
  63        if (unlikely(dst == old))               /* deadlock avoidance */
  64                return;
  65        bdi_lock_two(&old->wb, &dst->wb);
  66        spin_lock(&inode->i_lock);
  67        inode->i_data.backing_dev_info = dst;
  68        if (inode->i_state & I_DIRTY) {
  69                if (bdi_cap_writeback_dirty(dst) && !wb_has_dirty_io(&dst->wb))
  70                        wakeup_bdi = true;
  71                list_move(&inode->i_wb_list, &dst->wb.b_dirty);
  72        }
  73        spin_unlock(&inode->i_lock);
  74        spin_unlock(&old->wb.list_lock);
  75        spin_unlock(&dst->wb.list_lock);
  76
  77        if (wakeup_bdi)
  78                bdi_wakeup_thread_delayed(dst);
  79}
  80
  81/* Kill _all_ buffers and pagecache , dirty or not.. */
  82void kill_bdev(struct block_device *bdev)
  83{
  84        struct address_space *mapping = bdev->bd_inode->i_mapping;
  85
  86        if (mapping->nrpages == 0)
  87                return;
  88
  89        invalidate_bh_lrus();
  90        truncate_inode_pages(mapping, 0);
  91}       
  92EXPORT_SYMBOL(kill_bdev);
  93
  94/* Invalidate clean unused buffers and pagecache. */
  95void invalidate_bdev(struct block_device *bdev)
  96{
  97        struct address_space *mapping = bdev->bd_inode->i_mapping;
  98
  99        if (mapping->nrpages == 0)
 100                return;
 101
 102        invalidate_bh_lrus();
 103        lru_add_drain_all();    /* make sure all lru add caches are flushed */
 104        invalidate_mapping_pages(mapping, 0, -1);
 105        /* 99% of the time, we don't need to flush the cleancache on the bdev.
 106         * But, for the strange corners, lets be cautious
 107         */
 108        cleancache_invalidate_inode(mapping);
 109}
 110EXPORT_SYMBOL(invalidate_bdev);
 111
 112int set_blocksize(struct block_device *bdev, int size)
 113{
 114        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 115        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 116                return -EINVAL;
 117
 118        /* Size cannot be smaller than the size supported by the device */
 119        if (size < bdev_logical_block_size(bdev))
 120                return -EINVAL;
 121
 122        /* Don't change the size if it is same as current */
 123        if (bdev->bd_block_size != size) {
 124                sync_blockdev(bdev);
 125                bdev->bd_block_size = size;
 126                bdev->bd_inode->i_blkbits = blksize_bits(size);
 127                kill_bdev(bdev);
 128        }
 129        return 0;
 130}
 131
 132EXPORT_SYMBOL(set_blocksize);
 133
 134int sb_set_blocksize(struct super_block *sb, int size)
 135{
 136        if (set_blocksize(sb->s_bdev, size))
 137                return 0;
 138        /* If we get here, we know size is power of two
 139         * and it's value is between 512 and PAGE_SIZE */
 140        sb->s_blocksize = size;
 141        sb->s_blocksize_bits = blksize_bits(size);
 142        return sb->s_blocksize;
 143}
 144
 145EXPORT_SYMBOL(sb_set_blocksize);
 146
 147int sb_min_blocksize(struct super_block *sb, int size)
 148{
 149        int minsize = bdev_logical_block_size(sb->s_bdev);
 150        if (size < minsize)
 151                size = minsize;
 152        return sb_set_blocksize(sb, size);
 153}
 154
 155EXPORT_SYMBOL(sb_min_blocksize);
 156
 157static int
 158blkdev_get_block(struct inode *inode, sector_t iblock,
 159                struct buffer_head *bh, int create)
 160{
 161        bh->b_bdev = I_BDEV(inode);
 162        bh->b_blocknr = iblock;
 163        set_buffer_mapped(bh);
 164        return 0;
 165}
 166
 167static ssize_t
 168blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 169                        loff_t offset, unsigned long nr_segs)
 170{
 171        struct file *file = iocb->ki_filp;
 172        struct inode *inode = file->f_mapping->host;
 173
 174        return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 175                                    nr_segs, blkdev_get_block, NULL, NULL, 0);
 176}
 177
 178int __sync_blockdev(struct block_device *bdev, int wait)
 179{
 180        if (!bdev)
 181                return 0;
 182        if (!wait)
 183                return filemap_flush(bdev->bd_inode->i_mapping);
 184        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 185}
 186
 187/*
 188 * Write out and wait upon all the dirty data associated with a block
 189 * device via its mapping.  Does not take the superblock lock.
 190 */
 191int sync_blockdev(struct block_device *bdev)
 192{
 193        return __sync_blockdev(bdev, 1);
 194}
 195EXPORT_SYMBOL(sync_blockdev);
 196
 197/*
 198 * Write out and wait upon all dirty data associated with this
 199 * device.   Filesystem data as well as the underlying block
 200 * device.  Takes the superblock lock.
 201 */
 202int fsync_bdev(struct block_device *bdev)
 203{
 204        struct super_block *sb = get_super(bdev);
 205        if (sb) {
 206                int res = sync_filesystem(sb);
 207                drop_super(sb);
 208                return res;
 209        }
 210        return sync_blockdev(bdev);
 211}
 212EXPORT_SYMBOL(fsync_bdev);
 213
 214/**
 215 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 216 * @bdev:       blockdevice to lock
 217 *
 218 * If a superblock is found on this device, we take the s_umount semaphore
 219 * on it to make sure nobody unmounts until the snapshot creation is done.
 220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 224 * actually.
 225 */
 226struct super_block *freeze_bdev(struct block_device *bdev)
 227{
 228        struct super_block *sb;
 229        int error = 0;
 230
 231        mutex_lock(&bdev->bd_fsfreeze_mutex);
 232        if (++bdev->bd_fsfreeze_count > 1) {
 233                /*
 234                 * We don't even need to grab a reference - the first call
 235                 * to freeze_bdev grab an active reference and only the last
 236                 * thaw_bdev drops it.
 237                 */
 238                sb = get_super(bdev);
 239                drop_super(sb);
 240                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 241                return sb;
 242        }
 243
 244        sb = get_active_super(bdev);
 245        if (!sb)
 246                goto out;
 247        error = freeze_super(sb);
 248        if (error) {
 249                deactivate_super(sb);
 250                bdev->bd_fsfreeze_count--;
 251                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 252                return ERR_PTR(error);
 253        }
 254        deactivate_super(sb);
 255 out:
 256        sync_blockdev(bdev);
 257        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 258        return sb;      /* thaw_bdev releases s->s_umount */
 259}
 260EXPORT_SYMBOL(freeze_bdev);
 261
 262/**
 263 * thaw_bdev  -- unlock filesystem
 264 * @bdev:       blockdevice to unlock
 265 * @sb:         associated superblock
 266 *
 267 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 268 */
 269int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 270{
 271        int error = -EINVAL;
 272
 273        mutex_lock(&bdev->bd_fsfreeze_mutex);
 274        if (!bdev->bd_fsfreeze_count)
 275                goto out;
 276
 277        error = 0;
 278        if (--bdev->bd_fsfreeze_count > 0)
 279                goto out;
 280
 281        if (!sb)
 282                goto out;
 283
 284        error = thaw_super(sb);
 285        if (error) {
 286                bdev->bd_fsfreeze_count++;
 287                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 288                return error;
 289        }
 290out:
 291        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 292        return 0;
 293}
 294EXPORT_SYMBOL(thaw_bdev);
 295
 296static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 297{
 298        return block_write_full_page(page, blkdev_get_block, wbc);
 299}
 300
 301static int blkdev_readpage(struct file * file, struct page * page)
 302{
 303        return block_read_full_page(page, blkdev_get_block);
 304}
 305
 306static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 307                        loff_t pos, unsigned len, unsigned flags,
 308                        struct page **pagep, void **fsdata)
 309{
 310        return block_write_begin(mapping, pos, len, flags, pagep,
 311                                 blkdev_get_block);
 312}
 313
 314static int blkdev_write_end(struct file *file, struct address_space *mapping,
 315                        loff_t pos, unsigned len, unsigned copied,
 316                        struct page *page, void *fsdata)
 317{
 318        int ret;
 319        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 320
 321        unlock_page(page);
 322        page_cache_release(page);
 323
 324        return ret;
 325}
 326
 327/*
 328 * private llseek:
 329 * for a block special file file_inode(file)->i_size is zero
 330 * so we compute the size by hand (just as in block_read/write above)
 331 */
 332static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 333{
 334        struct inode *bd_inode = file->f_mapping->host;
 335        loff_t retval;
 336
 337        mutex_lock(&bd_inode->i_mutex);
 338        retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 339        mutex_unlock(&bd_inode->i_mutex);
 340        return retval;
 341}
 342        
 343int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 344{
 345        struct inode *bd_inode = filp->f_mapping->host;
 346        struct block_device *bdev = I_BDEV(bd_inode);
 347        int error;
 348        
 349        error = filemap_write_and_wait_range(filp->f_mapping, start, end);
 350        if (error)
 351                return error;
 352
 353        /*
 354         * There is no need to serialise calls to blkdev_issue_flush with
 355         * i_mutex and doing so causes performance issues with concurrent
 356         * O_SYNC writers to a block device.
 357         */
 358        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 359        if (error == -EOPNOTSUPP)
 360                error = 0;
 361
 362        return error;
 363}
 364EXPORT_SYMBOL(blkdev_fsync);
 365
 366/*
 367 * pseudo-fs
 368 */
 369
 370static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 371static struct kmem_cache * bdev_cachep __read_mostly;
 372
 373static struct inode *bdev_alloc_inode(struct super_block *sb)
 374{
 375        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 376        if (!ei)
 377                return NULL;
 378        return &ei->vfs_inode;
 379}
 380
 381static void bdev_i_callback(struct rcu_head *head)
 382{
 383        struct inode *inode = container_of(head, struct inode, i_rcu);
 384        struct bdev_inode *bdi = BDEV_I(inode);
 385
 386        kmem_cache_free(bdev_cachep, bdi);
 387}
 388
 389static void bdev_destroy_inode(struct inode *inode)
 390{
 391        call_rcu(&inode->i_rcu, bdev_i_callback);
 392}
 393
 394static void init_once(void *foo)
 395{
 396        struct bdev_inode *ei = (struct bdev_inode *) foo;
 397        struct block_device *bdev = &ei->bdev;
 398
 399        memset(bdev, 0, sizeof(*bdev));
 400        mutex_init(&bdev->bd_mutex);
 401        INIT_LIST_HEAD(&bdev->bd_inodes);
 402        INIT_LIST_HEAD(&bdev->bd_list);
 403#ifdef CONFIG_SYSFS
 404        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 405#endif
 406        inode_init_once(&ei->vfs_inode);
 407        /* Initialize mutex for freeze. */
 408        mutex_init(&bdev->bd_fsfreeze_mutex);
 409}
 410
 411static inline void __bd_forget(struct inode *inode)
 412{
 413        list_del_init(&inode->i_devices);
 414        inode->i_bdev = NULL;
 415        inode->i_mapping = &inode->i_data;
 416}
 417
 418static void bdev_evict_inode(struct inode *inode)
 419{
 420        struct block_device *bdev = &BDEV_I(inode)->bdev;
 421        struct list_head *p;
 422        truncate_inode_pages(&inode->i_data, 0);
 423        invalidate_inode_buffers(inode); /* is it needed here? */
 424        clear_inode(inode);
 425        spin_lock(&bdev_lock);
 426        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 427                __bd_forget(list_entry(p, struct inode, i_devices));
 428        }
 429        list_del_init(&bdev->bd_list);
 430        spin_unlock(&bdev_lock);
 431}
 432
 433static const struct super_operations bdev_sops = {
 434        .statfs = simple_statfs,
 435        .alloc_inode = bdev_alloc_inode,
 436        .destroy_inode = bdev_destroy_inode,
 437        .drop_inode = generic_delete_inode,
 438        .evict_inode = bdev_evict_inode,
 439};
 440
 441static struct dentry *bd_mount(struct file_system_type *fs_type,
 442        int flags, const char *dev_name, void *data)
 443{
 444        return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
 445}
 446
 447static struct file_system_type bd_type = {
 448        .name           = "bdev",
 449        .mount          = bd_mount,
 450        .kill_sb        = kill_anon_super,
 451};
 452
 453static struct super_block *blockdev_superblock __read_mostly;
 454
 455void __init bdev_cache_init(void)
 456{
 457        int err;
 458        static struct vfsmount *bd_mnt;
 459
 460        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 461                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 462                                SLAB_MEM_SPREAD|SLAB_PANIC),
 463                        init_once);
 464        err = register_filesystem(&bd_type);
 465        if (err)
 466                panic("Cannot register bdev pseudo-fs");
 467        bd_mnt = kern_mount(&bd_type);
 468        if (IS_ERR(bd_mnt))
 469                panic("Cannot create bdev pseudo-fs");
 470        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 471}
 472
 473/*
 474 * Most likely _very_ bad one - but then it's hardly critical for small
 475 * /dev and can be fixed when somebody will need really large one.
 476 * Keep in mind that it will be fed through icache hash function too.
 477 */
 478static inline unsigned long hash(dev_t dev)
 479{
 480        return MAJOR(dev)+MINOR(dev);
 481}
 482
 483static int bdev_test(struct inode *inode, void *data)
 484{
 485        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 486}
 487
 488static int bdev_set(struct inode *inode, void *data)
 489{
 490        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 491        return 0;
 492}
 493
 494static LIST_HEAD(all_bdevs);
 495
 496struct block_device *bdget(dev_t dev)
 497{
 498        struct block_device *bdev;
 499        struct inode *inode;
 500
 501        inode = iget5_locked(blockdev_superblock, hash(dev),
 502                        bdev_test, bdev_set, &dev);
 503
 504        if (!inode)
 505                return NULL;
 506
 507        bdev = &BDEV_I(inode)->bdev;
 508
 509        if (inode->i_state & I_NEW) {
 510                bdev->bd_contains = NULL;
 511                bdev->bd_super = NULL;
 512                bdev->bd_inode = inode;
 513                bdev->bd_block_size = (1 << inode->i_blkbits);
 514                bdev->bd_part_count = 0;
 515                bdev->bd_invalidated = 0;
 516                inode->i_mode = S_IFBLK;
 517                inode->i_rdev = dev;
 518                inode->i_bdev = bdev;
 519                inode->i_data.a_ops = &def_blk_aops;
 520                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 521                inode->i_data.backing_dev_info = &default_backing_dev_info;
 522                spin_lock(&bdev_lock);
 523                list_add(&bdev->bd_list, &all_bdevs);
 524                spin_unlock(&bdev_lock);
 525                unlock_new_inode(inode);
 526        }
 527        return bdev;
 528}
 529
 530EXPORT_SYMBOL(bdget);
 531
 532/**
 533 * bdgrab -- Grab a reference to an already referenced block device
 534 * @bdev:       Block device to grab a reference to.
 535 */
 536struct block_device *bdgrab(struct block_device *bdev)
 537{
 538        ihold(bdev->bd_inode);
 539        return bdev;
 540}
 541EXPORT_SYMBOL(bdgrab);
 542
 543long nr_blockdev_pages(void)
 544{
 545        struct block_device *bdev;
 546        long ret = 0;
 547        spin_lock(&bdev_lock);
 548        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 549                ret += bdev->bd_inode->i_mapping->nrpages;
 550        }
 551        spin_unlock(&bdev_lock);
 552        return ret;
 553}
 554
 555void bdput(struct block_device *bdev)
 556{
 557        iput(bdev->bd_inode);
 558}
 559
 560EXPORT_SYMBOL(bdput);
 561 
 562static struct block_device *bd_acquire(struct inode *inode)
 563{
 564        struct block_device *bdev;
 565
 566        spin_lock(&bdev_lock);
 567        bdev = inode->i_bdev;
 568        if (bdev) {
 569                ihold(bdev->bd_inode);
 570                spin_unlock(&bdev_lock);
 571                return bdev;
 572        }
 573        spin_unlock(&bdev_lock);
 574
 575        bdev = bdget(inode->i_rdev);
 576        if (bdev) {
 577                spin_lock(&bdev_lock);
 578                if (!inode->i_bdev) {
 579                        /*
 580                         * We take an additional reference to bd_inode,
 581                         * and it's released in clear_inode() of inode.
 582                         * So, we can access it via ->i_mapping always
 583                         * without igrab().
 584                         */
 585                        ihold(bdev->bd_inode);
 586                        inode->i_bdev = bdev;
 587                        inode->i_mapping = bdev->bd_inode->i_mapping;
 588                        list_add(&inode->i_devices, &bdev->bd_inodes);
 589                }
 590                spin_unlock(&bdev_lock);
 591        }
 592        return bdev;
 593}
 594
 595int sb_is_blkdev_sb(struct super_block *sb)
 596{
 597        return sb == blockdev_superblock;
 598}
 599
 600/* Call when you free inode */
 601
 602void bd_forget(struct inode *inode)
 603{
 604        struct block_device *bdev = NULL;
 605
 606        spin_lock(&bdev_lock);
 607        if (!sb_is_blkdev_sb(inode->i_sb))
 608                bdev = inode->i_bdev;
 609        __bd_forget(inode);
 610        spin_unlock(&bdev_lock);
 611
 612        if (bdev)
 613                iput(bdev->bd_inode);
 614}
 615
 616/**
 617 * bd_may_claim - test whether a block device can be claimed
 618 * @bdev: block device of interest
 619 * @whole: whole block device containing @bdev, may equal @bdev
 620 * @holder: holder trying to claim @bdev
 621 *
 622 * Test whether @bdev can be claimed by @holder.
 623 *
 624 * CONTEXT:
 625 * spin_lock(&bdev_lock).
 626 *
 627 * RETURNS:
 628 * %true if @bdev can be claimed, %false otherwise.
 629 */
 630static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 631                         void *holder)
 632{
 633        if (bdev->bd_holder == holder)
 634                return true;     /* already a holder */
 635        else if (bdev->bd_holder != NULL)
 636                return false;    /* held by someone else */
 637        else if (bdev->bd_contains == bdev)
 638                return true;     /* is a whole device which isn't held */
 639
 640        else if (whole->bd_holder == bd_may_claim)
 641                return true;     /* is a partition of a device that is being partitioned */
 642        else if (whole->bd_holder != NULL)
 643                return false;    /* is a partition of a held device */
 644        else
 645                return true;     /* is a partition of an un-held device */
 646}
 647
 648/**
 649 * bd_prepare_to_claim - prepare to claim a block device
 650 * @bdev: block device of interest
 651 * @whole: the whole device containing @bdev, may equal @bdev
 652 * @holder: holder trying to claim @bdev
 653 *
 654 * Prepare to claim @bdev.  This function fails if @bdev is already
 655 * claimed by another holder and waits if another claiming is in
 656 * progress.  This function doesn't actually claim.  On successful
 657 * return, the caller has ownership of bd_claiming and bd_holder[s].
 658 *
 659 * CONTEXT:
 660 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 661 * it multiple times.
 662 *
 663 * RETURNS:
 664 * 0 if @bdev can be claimed, -EBUSY otherwise.
 665 */
 666static int bd_prepare_to_claim(struct block_device *bdev,
 667                               struct block_device *whole, void *holder)
 668{
 669retry:
 670        /* if someone else claimed, fail */
 671        if (!bd_may_claim(bdev, whole, holder))
 672                return -EBUSY;
 673
 674        /* if claiming is already in progress, wait for it to finish */
 675        if (whole->bd_claiming) {
 676                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 677                DEFINE_WAIT(wait);
 678
 679                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 680                spin_unlock(&bdev_lock);
 681                schedule();
 682                finish_wait(wq, &wait);
 683                spin_lock(&bdev_lock);
 684                goto retry;
 685        }
 686
 687        /* yay, all mine */
 688        return 0;
 689}
 690
 691/**
 692 * bd_start_claiming - start claiming a block device
 693 * @bdev: block device of interest
 694 * @holder: holder trying to claim @bdev
 695 *
 696 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 697 * exclusively and mark that an exclusive open is in progress.  Each
 698 * successful call to this function must be matched with a call to
 699 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 700 * fail).
 701 *
 702 * This function is used to gain exclusive access to the block device
 703 * without actually causing other exclusive open attempts to fail. It
 704 * should be used when the open sequence itself requires exclusive
 705 * access but may subsequently fail.
 706 *
 707 * CONTEXT:
 708 * Might sleep.
 709 *
 710 * RETURNS:
 711 * Pointer to the block device containing @bdev on success, ERR_PTR()
 712 * value on failure.
 713 */
 714static struct block_device *bd_start_claiming(struct block_device *bdev,
 715                                              void *holder)
 716{
 717        struct gendisk *disk;
 718        struct block_device *whole;
 719        int partno, err;
 720
 721        might_sleep();
 722
 723        /*
 724         * @bdev might not have been initialized properly yet, look up
 725         * and grab the outer block device the hard way.
 726         */
 727        disk = get_gendisk(bdev->bd_dev, &partno);
 728        if (!disk)
 729                return ERR_PTR(-ENXIO);
 730
 731        /*
 732         * Normally, @bdev should equal what's returned from bdget_disk()
 733         * if partno is 0; however, some drivers (floppy) use multiple
 734         * bdev's for the same physical device and @bdev may be one of the
 735         * aliases.  Keep @bdev if partno is 0.  This means claimer
 736         * tracking is broken for those devices but it has always been that
 737         * way.
 738         */
 739        if (partno)
 740                whole = bdget_disk(disk, 0);
 741        else
 742                whole = bdgrab(bdev);
 743
 744        module_put(disk->fops->owner);
 745        put_disk(disk);
 746        if (!whole)
 747                return ERR_PTR(-ENOMEM);
 748
 749        /* prepare to claim, if successful, mark claiming in progress */
 750        spin_lock(&bdev_lock);
 751
 752        err = bd_prepare_to_claim(bdev, whole, holder);
 753        if (err == 0) {
 754                whole->bd_claiming = holder;
 755                spin_unlock(&bdev_lock);
 756                return whole;
 757        } else {
 758                spin_unlock(&bdev_lock);
 759                bdput(whole);
 760                return ERR_PTR(err);
 761        }
 762}
 763
 764#ifdef CONFIG_SYSFS
 765struct bd_holder_disk {
 766        struct list_head        list;
 767        struct gendisk          *disk;
 768        int                     refcnt;
 769};
 770
 771static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 772                                                  struct gendisk *disk)
 773{
 774        struct bd_holder_disk *holder;
 775
 776        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 777                if (holder->disk == disk)
 778                        return holder;
 779        return NULL;
 780}
 781
 782static int add_symlink(struct kobject *from, struct kobject *to)
 783{
 784        return sysfs_create_link(from, to, kobject_name(to));
 785}
 786
 787static void del_symlink(struct kobject *from, struct kobject *to)
 788{
 789        sysfs_remove_link(from, kobject_name(to));
 790}
 791
 792/**
 793 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 794 * @bdev: the claimed slave bdev
 795 * @disk: the holding disk
 796 *
 797 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 798 *
 799 * This functions creates the following sysfs symlinks.
 800 *
 801 * - from "slaves" directory of the holder @disk to the claimed @bdev
 802 * - from "holders" directory of the @bdev to the holder @disk
 803 *
 804 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
 805 * passed to bd_link_disk_holder(), then:
 806 *
 807 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
 808 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 809 *
 810 * The caller must have claimed @bdev before calling this function and
 811 * ensure that both @bdev and @disk are valid during the creation and
 812 * lifetime of these symlinks.
 813 *
 814 * CONTEXT:
 815 * Might sleep.
 816 *
 817 * RETURNS:
 818 * 0 on success, -errno on failure.
 819 */
 820int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 821{
 822        struct bd_holder_disk *holder;
 823        int ret = 0;
 824
 825        mutex_lock(&bdev->bd_mutex);
 826
 827        WARN_ON_ONCE(!bdev->bd_holder);
 828
 829        /* FIXME: remove the following once add_disk() handles errors */
 830        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
 831                goto out_unlock;
 832
 833        holder = bd_find_holder_disk(bdev, disk);
 834        if (holder) {
 835                holder->refcnt++;
 836                goto out_unlock;
 837        }
 838
 839        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 840        if (!holder) {
 841                ret = -ENOMEM;
 842                goto out_unlock;
 843        }
 844
 845        INIT_LIST_HEAD(&holder->list);
 846        holder->disk = disk;
 847        holder->refcnt = 1;
 848
 849        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 850        if (ret)
 851                goto out_free;
 852
 853        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
 854        if (ret)
 855                goto out_del;
 856        /*
 857         * bdev could be deleted beneath us which would implicitly destroy
 858         * the holder directory.  Hold on to it.
 859         */
 860        kobject_get(bdev->bd_part->holder_dir);
 861
 862        list_add(&holder->list, &bdev->bd_holder_disks);
 863        goto out_unlock;
 864
 865out_del:
 866        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 867out_free:
 868        kfree(holder);
 869out_unlock:
 870        mutex_unlock(&bdev->bd_mutex);
 871        return ret;
 872}
 873EXPORT_SYMBOL_GPL(bd_link_disk_holder);
 874
 875/**
 876 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
 877 * @bdev: the calimed slave bdev
 878 * @disk: the holding disk
 879 *
 880 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 881 *
 882 * CONTEXT:
 883 * Might sleep.
 884 */
 885void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 886{
 887        struct bd_holder_disk *holder;
 888
 889        mutex_lock(&bdev->bd_mutex);
 890
 891        holder = bd_find_holder_disk(bdev, disk);
 892
 893        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 894                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 895                del_symlink(bdev->bd_part->holder_dir,
 896                            &disk_to_dev(disk)->kobj);
 897                kobject_put(bdev->bd_part->holder_dir);
 898                list_del_init(&holder->list);
 899                kfree(holder);
 900        }
 901
 902        mutex_unlock(&bdev->bd_mutex);
 903}
 904EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
 905#endif
 906
 907/**
 908 * flush_disk - invalidates all buffer-cache entries on a disk
 909 *
 910 * @bdev:      struct block device to be flushed
 911 * @kill_dirty: flag to guide handling of dirty inodes
 912 *
 913 * Invalidates all buffer-cache entries on a disk. It should be called
 914 * when a disk has been changed -- either by a media change or online
 915 * resize.
 916 */
 917static void flush_disk(struct block_device *bdev, bool kill_dirty)
 918{
 919        if (__invalidate_device(bdev, kill_dirty)) {
 920                char name[BDEVNAME_SIZE] = "";
 921
 922                if (bdev->bd_disk)
 923                        disk_name(bdev->bd_disk, 0, name);
 924                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 925                       "resized disk %s\n", name);
 926        }
 927
 928        if (!bdev->bd_disk)
 929                return;
 930        if (disk_part_scan_enabled(bdev->bd_disk))
 931                bdev->bd_invalidated = 1;
 932}
 933
 934/**
 935 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 936 * @disk: struct gendisk to check
 937 * @bdev: struct bdev to adjust.
 938 *
 939 * This routine checks to see if the bdev size does not match the disk size
 940 * and adjusts it if it differs.
 941 */
 942void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 943{
 944        loff_t disk_size, bdev_size;
 945
 946        disk_size = (loff_t)get_capacity(disk) << 9;
 947        bdev_size = i_size_read(bdev->bd_inode);
 948        if (disk_size != bdev_size) {
 949                char name[BDEVNAME_SIZE];
 950
 951                disk_name(disk, 0, name);
 952                printk(KERN_INFO
 953                       "%s: detected capacity change from %lld to %lld\n",
 954                       name, bdev_size, disk_size);
 955                i_size_write(bdev->bd_inode, disk_size);
 956                flush_disk(bdev, false);
 957        }
 958}
 959EXPORT_SYMBOL(check_disk_size_change);
 960
 961/**
 962 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 963 * @disk: struct gendisk to be revalidated
 964 *
 965 * This routine is a wrapper for lower-level driver's revalidate_disk
 966 * call-backs.  It is used to do common pre and post operations needed
 967 * for all revalidate_disk operations.
 968 */
 969int revalidate_disk(struct gendisk *disk)
 970{
 971        struct block_device *bdev;
 972        int ret = 0;
 973
 974        if (disk->fops->revalidate_disk)
 975                ret = disk->fops->revalidate_disk(disk);
 976
 977        bdev = bdget_disk(disk, 0);
 978        if (!bdev)
 979                return ret;
 980
 981        mutex_lock(&bdev->bd_mutex);
 982        check_disk_size_change(disk, bdev);
 983        bdev->bd_invalidated = 0;
 984        mutex_unlock(&bdev->bd_mutex);
 985        bdput(bdev);
 986        return ret;
 987}
 988EXPORT_SYMBOL(revalidate_disk);
 989
 990/*
 991 * This routine checks whether a removable media has been changed,
 992 * and invalidates all buffer-cache-entries in that case. This
 993 * is a relatively slow routine, so we have to try to minimize using
 994 * it. Thus it is called only upon a 'mount' or 'open'. This
 995 * is the best way of combining speed and utility, I think.
 996 * People changing diskettes in the middle of an operation deserve
 997 * to lose :-)
 998 */
 999int check_disk_change(struct block_device *bdev)
1000{
1001        struct gendisk *disk = bdev->bd_disk;
1002        const struct block_device_operations *bdops = disk->fops;
1003        unsigned int events;
1004
1005        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1006                                   DISK_EVENT_EJECT_REQUEST);
1007        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1008                return 0;
1009
1010        flush_disk(bdev, true);
1011        if (bdops->revalidate_disk)
1012                bdops->revalidate_disk(bdev->bd_disk);
1013        return 1;
1014}
1015
1016EXPORT_SYMBOL(check_disk_change);
1017
1018void bd_set_size(struct block_device *bdev, loff_t size)
1019{
1020        unsigned bsize = bdev_logical_block_size(bdev);
1021
1022        mutex_lock(&bdev->bd_inode->i_mutex);
1023        i_size_write(bdev->bd_inode, size);
1024        mutex_unlock(&bdev->bd_inode->i_mutex);
1025        while (bsize < PAGE_CACHE_SIZE) {
1026                if (size & bsize)
1027                        break;
1028                bsize <<= 1;
1029        }
1030        bdev->bd_block_size = bsize;
1031        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1032}
1033EXPORT_SYMBOL(bd_set_size);
1034
1035static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1036
1037/*
1038 * bd_mutex locking:
1039 *
1040 *  mutex_lock(part->bd_mutex)
1041 *    mutex_lock_nested(whole->bd_mutex, 1)
1042 */
1043
1044static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1045{
1046        struct gendisk *disk;
1047        struct module *owner;
1048        int ret;
1049        int partno;
1050        int perm = 0;
1051
1052        if (mode & FMODE_READ)
1053                perm |= MAY_READ;
1054        if (mode & FMODE_WRITE)
1055                perm |= MAY_WRITE;
1056        /*
1057         * hooks: /n/, see "layering violations".
1058         */
1059        if (!for_part) {
1060                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1061                if (ret != 0) {
1062                        bdput(bdev);
1063                        return ret;
1064                }
1065        }
1066
1067 restart:
1068
1069        ret = -ENXIO;
1070        disk = get_gendisk(bdev->bd_dev, &partno);
1071        if (!disk)
1072                goto out;
1073        owner = disk->fops->owner;
1074
1075        disk_block_events(disk);
1076        mutex_lock_nested(&bdev->bd_mutex, for_part);
1077        if (!bdev->bd_openers) {
1078                bdev->bd_disk = disk;
1079                bdev->bd_queue = disk->queue;
1080                bdev->bd_contains = bdev;
1081                if (!partno) {
1082                        struct backing_dev_info *bdi;
1083
1084                        ret = -ENXIO;
1085                        bdev->bd_part = disk_get_part(disk, partno);
1086                        if (!bdev->bd_part)
1087                                goto out_clear;
1088
1089                        ret = 0;
1090                        if (disk->fops->open) {
1091                                ret = disk->fops->open(bdev, mode);
1092                                if (ret == -ERESTARTSYS) {
1093                                        /* Lost a race with 'disk' being
1094                                         * deleted, try again.
1095                                         * See md.c
1096                                         */
1097                                        disk_put_part(bdev->bd_part);
1098                                        bdev->bd_part = NULL;
1099                                        bdev->bd_disk = NULL;
1100                                        bdev->bd_queue = NULL;
1101                                        mutex_unlock(&bdev->bd_mutex);
1102                                        disk_unblock_events(disk);
1103                                        put_disk(disk);
1104                                        module_put(owner);
1105                                        goto restart;
1106                                }
1107                        }
1108
1109                        if (!ret) {
1110                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1111                                bdi = blk_get_backing_dev_info(bdev);
1112                                if (bdi == NULL)
1113                                        bdi = &default_backing_dev_info;
1114                                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1115                        }
1116
1117                        /*
1118                         * If the device is invalidated, rescan partition
1119                         * if open succeeded or failed with -ENOMEDIUM.
1120                         * The latter is necessary to prevent ghost
1121                         * partitions on a removed medium.
1122                         */
1123                        if (bdev->bd_invalidated) {
1124                                if (!ret)
1125                                        rescan_partitions(disk, bdev);
1126                                else if (ret == -ENOMEDIUM)
1127                                        invalidate_partitions(disk, bdev);
1128                        }
1129                        if (ret)
1130                                goto out_clear;
1131                } else {
1132                        struct block_device *whole;
1133                        whole = bdget_disk(disk, 0);
1134                        ret = -ENOMEM;
1135                        if (!whole)
1136                                goto out_clear;
1137                        BUG_ON(for_part);
1138                        ret = __blkdev_get(whole, mode, 1);
1139                        if (ret)
1140                                goto out_clear;
1141                        bdev->bd_contains = whole;
1142                        bdev_inode_switch_bdi(bdev->bd_inode,
1143                                whole->bd_inode->i_data.backing_dev_info);
1144                        bdev->bd_part = disk_get_part(disk, partno);
1145                        if (!(disk->flags & GENHD_FL_UP) ||
1146                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1147                                ret = -ENXIO;
1148                                goto out_clear;
1149                        }
1150                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1151                }
1152        } else {
1153                if (bdev->bd_contains == bdev) {
1154                        ret = 0;
1155                        if (bdev->bd_disk->fops->open)
1156                                ret = bdev->bd_disk->fops->open(bdev, mode);
1157                        /* the same as first opener case, read comment there */
1158                        if (bdev->bd_invalidated) {
1159                                if (!ret)
1160                                        rescan_partitions(bdev->bd_disk, bdev);
1161                                else if (ret == -ENOMEDIUM)
1162                                        invalidate_partitions(bdev->bd_disk, bdev);
1163                        }
1164                        if (ret)
1165                                goto out_unlock_bdev;
1166                }
1167                /* only one opener holds refs to the module and disk */
1168                put_disk(disk);
1169                module_put(owner);
1170        }
1171        bdev->bd_openers++;
1172        if (for_part)
1173                bdev->bd_part_count++;
1174        mutex_unlock(&bdev->bd_mutex);
1175        disk_unblock_events(disk);
1176        return 0;
1177
1178 out_clear:
1179        disk_put_part(bdev->bd_part);
1180        bdev->bd_disk = NULL;
1181        bdev->bd_part = NULL;
1182        bdev->bd_queue = NULL;
1183        bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1184        if (bdev != bdev->bd_contains)
1185                __blkdev_put(bdev->bd_contains, mode, 1);
1186        bdev->bd_contains = NULL;
1187 out_unlock_bdev:
1188        mutex_unlock(&bdev->bd_mutex);
1189        disk_unblock_events(disk);
1190        put_disk(disk);
1191        module_put(owner);
1192 out:
1193        bdput(bdev);
1194
1195        return ret;
1196}
1197
1198/**
1199 * blkdev_get - open a block device
1200 * @bdev: block_device to open
1201 * @mode: FMODE_* mask
1202 * @holder: exclusive holder identifier
1203 *
1204 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1205 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1206 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1207 *
1208 * On success, the reference count of @bdev is unchanged.  On failure,
1209 * @bdev is put.
1210 *
1211 * CONTEXT:
1212 * Might sleep.
1213 *
1214 * RETURNS:
1215 * 0 on success, -errno on failure.
1216 */
1217int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1218{
1219        struct block_device *whole = NULL;
1220        int res;
1221
1222        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1223
1224        if ((mode & FMODE_EXCL) && holder) {
1225                whole = bd_start_claiming(bdev, holder);
1226                if (IS_ERR(whole)) {
1227                        bdput(bdev);
1228                        return PTR_ERR(whole);
1229                }
1230        }
1231
1232        res = __blkdev_get(bdev, mode, 0);
1233
1234        if (whole) {
1235                struct gendisk *disk = whole->bd_disk;
1236
1237                /* finish claiming */
1238                mutex_lock(&bdev->bd_mutex);
1239                spin_lock(&bdev_lock);
1240
1241                if (!res) {
1242                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1243                        /*
1244                         * Note that for a whole device bd_holders
1245                         * will be incremented twice, and bd_holder
1246                         * will be set to bd_may_claim before being
1247                         * set to holder
1248                         */
1249                        whole->bd_holders++;
1250                        whole->bd_holder = bd_may_claim;
1251                        bdev->bd_holders++;
1252                        bdev->bd_holder = holder;
1253                }
1254
1255                /* tell others that we're done */
1256                BUG_ON(whole->bd_claiming != holder);
1257                whole->bd_claiming = NULL;
1258                wake_up_bit(&whole->bd_claiming, 0);
1259
1260                spin_unlock(&bdev_lock);
1261
1262                /*
1263                 * Block event polling for write claims if requested.  Any
1264                 * write holder makes the write_holder state stick until
1265                 * all are released.  This is good enough and tracking
1266                 * individual writeable reference is too fragile given the
1267                 * way @mode is used in blkdev_get/put().
1268                 */
1269                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1270                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1271                        bdev->bd_write_holder = true;
1272                        disk_block_events(disk);
1273                }
1274
1275                mutex_unlock(&bdev->bd_mutex);
1276                bdput(whole);
1277        }
1278
1279        return res;
1280}
1281EXPORT_SYMBOL(blkdev_get);
1282
1283/**
1284 * blkdev_get_by_path - open a block device by name
1285 * @path: path to the block device to open
1286 * @mode: FMODE_* mask
1287 * @holder: exclusive holder identifier
1288 *
1289 * Open the blockdevice described by the device file at @path.  @mode
1290 * and @holder are identical to blkdev_get().
1291 *
1292 * On success, the returned block_device has reference count of one.
1293 *
1294 * CONTEXT:
1295 * Might sleep.
1296 *
1297 * RETURNS:
1298 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1299 */
1300struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1301                                        void *holder)
1302{
1303        struct block_device *bdev;
1304        int err;
1305
1306        bdev = lookup_bdev(path);
1307        if (IS_ERR(bdev))
1308                return bdev;
1309
1310        err = blkdev_get(bdev, mode, holder);
1311        if (err)
1312                return ERR_PTR(err);
1313
1314        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1315                blkdev_put(bdev, mode);
1316                return ERR_PTR(-EACCES);
1317        }
1318
1319        return bdev;
1320}
1321EXPORT_SYMBOL(blkdev_get_by_path);
1322
1323/**
1324 * blkdev_get_by_dev - open a block device by device number
1325 * @dev: device number of block device to open
1326 * @mode: FMODE_* mask
1327 * @holder: exclusive holder identifier
1328 *
1329 * Open the blockdevice described by device number @dev.  @mode and
1330 * @holder are identical to blkdev_get().
1331 *
1332 * Use it ONLY if you really do not have anything better - i.e. when
1333 * you are behind a truly sucky interface and all you are given is a
1334 * device number.  _Never_ to be used for internal purposes.  If you
1335 * ever need it - reconsider your API.
1336 *
1337 * On success, the returned block_device has reference count of one.
1338 *
1339 * CONTEXT:
1340 * Might sleep.
1341 *
1342 * RETURNS:
1343 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1344 */
1345struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1346{
1347        struct block_device *bdev;
1348        int err;
1349
1350        bdev = bdget(dev);
1351        if (!bdev)
1352                return ERR_PTR(-ENOMEM);
1353
1354        err = blkdev_get(bdev, mode, holder);
1355        if (err)
1356                return ERR_PTR(err);
1357
1358        return bdev;
1359}
1360EXPORT_SYMBOL(blkdev_get_by_dev);
1361
1362static int blkdev_open(struct inode * inode, struct file * filp)
1363{
1364        struct block_device *bdev;
1365
1366        /*
1367         * Preserve backwards compatibility and allow large file access
1368         * even if userspace doesn't ask for it explicitly. Some mkfs
1369         * binary needs it. We might want to drop this workaround
1370         * during an unstable branch.
1371         */
1372        filp->f_flags |= O_LARGEFILE;
1373
1374        if (filp->f_flags & O_NDELAY)
1375                filp->f_mode |= FMODE_NDELAY;
1376        if (filp->f_flags & O_EXCL)
1377                filp->f_mode |= FMODE_EXCL;
1378        if ((filp->f_flags & O_ACCMODE) == 3)
1379                filp->f_mode |= FMODE_WRITE_IOCTL;
1380
1381        bdev = bd_acquire(inode);
1382        if (bdev == NULL)
1383                return -ENOMEM;
1384
1385        filp->f_mapping = bdev->bd_inode->i_mapping;
1386
1387        return blkdev_get(bdev, filp->f_mode, filp);
1388}
1389
1390static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1391{
1392        struct gendisk *disk = bdev->bd_disk;
1393        struct block_device *victim = NULL;
1394
1395        mutex_lock_nested(&bdev->bd_mutex, for_part);
1396        if (for_part)
1397                bdev->bd_part_count--;
1398
1399        if (!--bdev->bd_openers) {
1400                WARN_ON_ONCE(bdev->bd_holders);
1401                sync_blockdev(bdev);
1402                kill_bdev(bdev);
1403                /* ->release can cause the old bdi to disappear,
1404                 * so must switch it out first
1405                 */
1406                bdev_inode_switch_bdi(bdev->bd_inode,
1407                                        &default_backing_dev_info);
1408        }
1409        if (bdev->bd_contains == bdev) {
1410                if (disk->fops->release)
1411                        disk->fops->release(disk, mode);
1412        }
1413        if (!bdev->bd_openers) {
1414                struct module *owner = disk->fops->owner;
1415
1416                disk_put_part(bdev->bd_part);
1417                bdev->bd_part = NULL;
1418                bdev->bd_disk = NULL;
1419                if (bdev != bdev->bd_contains)
1420                        victim = bdev->bd_contains;
1421                bdev->bd_contains = NULL;
1422
1423                put_disk(disk);
1424                module_put(owner);
1425        }
1426        mutex_unlock(&bdev->bd_mutex);
1427        bdput(bdev);
1428        if (victim)
1429                __blkdev_put(victim, mode, 1);
1430}
1431
1432void blkdev_put(struct block_device *bdev, fmode_t mode)
1433{
1434        mutex_lock(&bdev->bd_mutex);
1435
1436        if (mode & FMODE_EXCL) {
1437                bool bdev_free;
1438
1439                /*
1440                 * Release a claim on the device.  The holder fields
1441                 * are protected with bdev_lock.  bd_mutex is to
1442                 * synchronize disk_holder unlinking.
1443                 */
1444                spin_lock(&bdev_lock);
1445
1446                WARN_ON_ONCE(--bdev->bd_holders < 0);
1447                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1448
1449                /* bd_contains might point to self, check in a separate step */
1450                if ((bdev_free = !bdev->bd_holders))
1451                        bdev->bd_holder = NULL;
1452                if (!bdev->bd_contains->bd_holders)
1453                        bdev->bd_contains->bd_holder = NULL;
1454
1455                spin_unlock(&bdev_lock);
1456
1457                /*
1458                 * If this was the last claim, remove holder link and
1459                 * unblock evpoll if it was a write holder.
1460                 */
1461                if (bdev_free && bdev->bd_write_holder) {
1462                        disk_unblock_events(bdev->bd_disk);
1463                        bdev->bd_write_holder = false;
1464                }
1465        }
1466
1467        /*
1468         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1469         * event.  This is to ensure detection of media removal commanded
1470         * from userland - e.g. eject(1).
1471         */
1472        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1473
1474        mutex_unlock(&bdev->bd_mutex);
1475
1476        __blkdev_put(bdev, mode, 0);
1477}
1478EXPORT_SYMBOL(blkdev_put);
1479
1480static int blkdev_close(struct inode * inode, struct file * filp)
1481{
1482        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1483        blkdev_put(bdev, filp->f_mode);
1484        return 0;
1485}
1486
1487static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1488{
1489        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1490        fmode_t mode = file->f_mode;
1491
1492        /*
1493         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1494         * to updated it before every ioctl.
1495         */
1496        if (file->f_flags & O_NDELAY)
1497                mode |= FMODE_NDELAY;
1498        else
1499                mode &= ~FMODE_NDELAY;
1500
1501        return blkdev_ioctl(bdev, mode, cmd, arg);
1502}
1503
1504/*
1505 * Write data to the block device.  Only intended for the block device itself
1506 * and the raw driver which basically is a fake block device.
1507 *
1508 * Does not take i_mutex for the write and thus is not for general purpose
1509 * use.
1510 */
1511ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1512                         unsigned long nr_segs, loff_t pos)
1513{
1514        struct file *file = iocb->ki_filp;
1515        struct blk_plug plug;
1516        ssize_t ret;
1517
1518        BUG_ON(iocb->ki_pos != pos);
1519
1520        blk_start_plug(&plug);
1521        ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1522        if (ret > 0) {
1523                ssize_t err;
1524
1525                err = generic_write_sync(file, pos, ret);
1526                if (err < 0 && ret > 0)
1527                        ret = err;
1528        }
1529        blk_finish_plug(&plug);
1530        return ret;
1531}
1532EXPORT_SYMBOL_GPL(blkdev_aio_write);
1533
1534static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1535                         unsigned long nr_segs, loff_t pos)
1536{
1537        struct file *file = iocb->ki_filp;
1538        struct inode *bd_inode = file->f_mapping->host;
1539        loff_t size = i_size_read(bd_inode);
1540
1541        if (pos >= size)
1542                return 0;
1543
1544        size -= pos;
1545        if (size < iocb->ki_nbytes)
1546                nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1547        return generic_file_aio_read(iocb, iov, nr_segs, pos);
1548}
1549
1550/*
1551 * Try to release a page associated with block device when the system
1552 * is under memory pressure.
1553 */
1554static int blkdev_releasepage(struct page *page, gfp_t wait)
1555{
1556        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1557
1558        if (super && super->s_op->bdev_try_to_free_page)
1559                return super->s_op->bdev_try_to_free_page(super, page, wait);
1560
1561        return try_to_free_buffers(page);
1562}
1563
1564static const struct address_space_operations def_blk_aops = {
1565        .readpage       = blkdev_readpage,
1566        .writepage      = blkdev_writepage,
1567        .write_begin    = blkdev_write_begin,
1568        .write_end      = blkdev_write_end,
1569        .writepages     = generic_writepages,
1570        .releasepage    = blkdev_releasepage,
1571        .direct_IO      = blkdev_direct_IO,
1572        .is_dirty_writeback = buffer_check_dirty_writeback,
1573};
1574
1575const struct file_operations def_blk_fops = {
1576        .open           = blkdev_open,
1577        .release        = blkdev_close,
1578        .llseek         = block_llseek,
1579        .read           = do_sync_read,
1580        .write          = do_sync_write,
1581        .aio_read       = blkdev_aio_read,
1582        .aio_write      = blkdev_aio_write,
1583        .mmap           = generic_file_mmap,
1584        .fsync          = blkdev_fsync,
1585        .unlocked_ioctl = block_ioctl,
1586#ifdef CONFIG_COMPAT
1587        .compat_ioctl   = compat_blkdev_ioctl,
1588#endif
1589        .splice_read    = generic_file_splice_read,
1590        .splice_write   = generic_file_splice_write,
1591};
1592
1593int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1594{
1595        int res;
1596        mm_segment_t old_fs = get_fs();
1597        set_fs(KERNEL_DS);
1598        res = blkdev_ioctl(bdev, 0, cmd, arg);
1599        set_fs(old_fs);
1600        return res;
1601}
1602
1603EXPORT_SYMBOL(ioctl_by_bdev);
1604
1605/**
1606 * lookup_bdev  - lookup a struct block_device by name
1607 * @pathname:   special file representing the block device
1608 *
1609 * Get a reference to the blockdevice at @pathname in the current
1610 * namespace if possible and return it.  Return ERR_PTR(error)
1611 * otherwise.
1612 */
1613struct block_device *lookup_bdev(const char *pathname)
1614{
1615        struct block_device *bdev;
1616        struct inode *inode;
1617        struct path path;
1618        int error;
1619
1620        if (!pathname || !*pathname)
1621                return ERR_PTR(-EINVAL);
1622
1623        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1624        if (error)
1625                return ERR_PTR(error);
1626
1627        inode = path.dentry->d_inode;
1628        error = -ENOTBLK;
1629        if (!S_ISBLK(inode->i_mode))
1630                goto fail;
1631        error = -EACCES;
1632        if (path.mnt->mnt_flags & MNT_NODEV)
1633                goto fail;
1634        error = -ENOMEM;
1635        bdev = bd_acquire(inode);
1636        if (!bdev)
1637                goto fail;
1638out:
1639        path_put(&path);
1640        return bdev;
1641fail:
1642        bdev = ERR_PTR(error);
1643        goto out;
1644}
1645EXPORT_SYMBOL(lookup_bdev);
1646
1647int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1648{
1649        struct super_block *sb = get_super(bdev);
1650        int res = 0;
1651
1652        if (sb) {
1653                /*
1654                 * no need to lock the super, get_super holds the
1655                 * read mutex so the filesystem cannot go away
1656                 * under us (->put_super runs with the write lock
1657                 * hold).
1658                 */
1659                shrink_dcache_sb(sb);
1660                res = invalidate_inodes(sb, kill_dirty);
1661                drop_super(sb);
1662        }
1663        invalidate_bdev(bdev);
1664        return res;
1665}
1666EXPORT_SYMBOL(__invalidate_device);
1667
1668void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1669{
1670        struct inode *inode, *old_inode = NULL;
1671
1672        spin_lock(&inode_sb_list_lock);
1673        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1674                struct address_space *mapping = inode->i_mapping;
1675
1676                spin_lock(&inode->i_lock);
1677                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1678                    mapping->nrpages == 0) {
1679                        spin_unlock(&inode->i_lock);
1680                        continue;
1681                }
1682                __iget(inode);
1683                spin_unlock(&inode->i_lock);
1684                spin_unlock(&inode_sb_list_lock);
1685                /*
1686                 * We hold a reference to 'inode' so it couldn't have been
1687                 * removed from s_inodes list while we dropped the
1688                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1689                 * be holding the last reference and we cannot iput it under
1690                 * inode_sb_list_lock. So we keep the reference and iput it
1691                 * later.
1692                 */
1693                iput(old_inode);
1694                old_inode = inode;
1695
1696                func(I_BDEV(inode), arg);
1697
1698                spin_lock(&inode_sb_list_lock);
1699        }
1700        spin_unlock(&inode_sb_list_lock);
1701        iput(old_inode);
1702}
1703
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.