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 <asm/uaccess.h>
  31#include "internal.h"
  32
  33struct bdev_inode {
  34        struct block_device bdev;
  35        struct inode vfs_inode;
  36};
  37
  38static const struct address_space_operations def_blk_aops;
  39
  40static inline struct bdev_inode *BDEV_I(struct inode *inode)
  41{
  42        return container_of(inode, struct bdev_inode, vfs_inode);
  43}
  44
  45inline struct block_device *I_BDEV(struct inode *inode)
  46{
  47        return &BDEV_I(inode)->bdev;
  48}
  49EXPORT_SYMBOL(I_BDEV);
  50
  51/*
  52 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
  53 * need to move it onto the dirty list of @dst so that the inode is always on
  54 * the right list.
  55 */
  56static void bdev_inode_switch_bdi(struct inode *inode,
  57                        struct backing_dev_info *dst)
  58{
  59        struct backing_dev_info *old = inode->i_data.backing_dev_info;
  60
  61        if (unlikely(dst == old))               /* deadlock avoidance */
  62                return;
  63        bdi_lock_two(&old->wb, &dst->wb);
  64        spin_lock(&inode->i_lock);
  65        inode->i_data.backing_dev_info = dst;
  66        if (inode->i_state & I_DIRTY)
  67                list_move(&inode->i_wb_list, &dst->wb.b_dirty);
  68        spin_unlock(&inode->i_lock);
  69        spin_unlock(&old->wb.list_lock);
  70        spin_unlock(&dst->wb.list_lock);
  71}
  72
  73/* Kill _all_ buffers and pagecache , dirty or not.. */
  74void kill_bdev(struct block_device *bdev)
  75{
  76        struct address_space *mapping = bdev->bd_inode->i_mapping;
  77
  78        if (mapping->nrpages == 0)
  79                return;
  80
  81        invalidate_bh_lrus();
  82        truncate_inode_pages(mapping, 0);
  83}       
  84EXPORT_SYMBOL(kill_bdev);
  85
  86/* Invalidate clean unused buffers and pagecache. */
  87void invalidate_bdev(struct block_device *bdev)
  88{
  89        struct address_space *mapping = bdev->bd_inode->i_mapping;
  90
  91        if (mapping->nrpages == 0)
  92                return;
  93
  94        invalidate_bh_lrus();
  95        lru_add_drain_all();    /* make sure all lru add caches are flushed */
  96        invalidate_mapping_pages(mapping, 0, -1);
  97        /* 99% of the time, we don't need to flush the cleancache on the bdev.
  98         * But, for the strange corners, lets be cautious
  99         */
 100        cleancache_invalidate_inode(mapping);
 101}
 102EXPORT_SYMBOL(invalidate_bdev);
 103
 104int set_blocksize(struct block_device *bdev, int size)
 105{
 106        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 107        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 108                return -EINVAL;
 109
 110        /* Size cannot be smaller than the size supported by the device */
 111        if (size < bdev_logical_block_size(bdev))
 112                return -EINVAL;
 113
 114        /* Don't change the size if it is same as current */
 115        if (bdev->bd_block_size != size) {
 116                sync_blockdev(bdev);
 117                bdev->bd_block_size = size;
 118                bdev->bd_inode->i_blkbits = blksize_bits(size);
 119                kill_bdev(bdev);
 120        }
 121        return 0;
 122}
 123
 124EXPORT_SYMBOL(set_blocksize);
 125
 126int sb_set_blocksize(struct super_block *sb, int size)
 127{
 128        if (set_blocksize(sb->s_bdev, size))
 129                return 0;
 130        /* If we get here, we know size is power of two
 131         * and it's value is between 512 and PAGE_SIZE */
 132        sb->s_blocksize = size;
 133        sb->s_blocksize_bits = blksize_bits(size);
 134        return sb->s_blocksize;
 135}
 136
 137EXPORT_SYMBOL(sb_set_blocksize);
 138
 139int sb_min_blocksize(struct super_block *sb, int size)
 140{
 141        int minsize = bdev_logical_block_size(sb->s_bdev);
 142        if (size < minsize)
 143                size = minsize;
 144        return sb_set_blocksize(sb, size);
 145}
 146
 147EXPORT_SYMBOL(sb_min_blocksize);
 148
 149static int
 150blkdev_get_block(struct inode *inode, sector_t iblock,
 151                struct buffer_head *bh, int create)
 152{
 153        bh->b_bdev = I_BDEV(inode);
 154        bh->b_blocknr = iblock;
 155        set_buffer_mapped(bh);
 156        return 0;
 157}
 158
 159static ssize_t
 160blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 161                        loff_t offset, unsigned long nr_segs)
 162{
 163        struct file *file = iocb->ki_filp;
 164        struct inode *inode = file->f_mapping->host;
 165
 166        return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 167                                    nr_segs, blkdev_get_block, NULL, NULL, 0);
 168}
 169
 170int __sync_blockdev(struct block_device *bdev, int wait)
 171{
 172        if (!bdev)
 173                return 0;
 174        if (!wait)
 175                return filemap_flush(bdev->bd_inode->i_mapping);
 176        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 177}
 178
 179/*
 180 * Write out and wait upon all the dirty data associated with a block
 181 * device via its mapping.  Does not take the superblock lock.
 182 */
 183int sync_blockdev(struct block_device *bdev)
 184{
 185        return __sync_blockdev(bdev, 1);
 186}
 187EXPORT_SYMBOL(sync_blockdev);
 188
 189/*
 190 * Write out and wait upon all dirty data associated with this
 191 * device.   Filesystem data as well as the underlying block
 192 * device.  Takes the superblock lock.
 193 */
 194int fsync_bdev(struct block_device *bdev)
 195{
 196        struct super_block *sb = get_super(bdev);
 197        if (sb) {
 198                int res = sync_filesystem(sb);
 199                drop_super(sb);
 200                return res;
 201        }
 202        return sync_blockdev(bdev);
 203}
 204EXPORT_SYMBOL(fsync_bdev);
 205
 206/**
 207 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 208 * @bdev:       blockdevice to lock
 209 *
 210 * If a superblock is found on this device, we take the s_umount semaphore
 211 * on it to make sure nobody unmounts until the snapshot creation is done.
 212 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 213 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 214 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 215 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 216 * actually.
 217 */
 218struct super_block *freeze_bdev(struct block_device *bdev)
 219{
 220        struct super_block *sb;
 221        int error = 0;
 222
 223        mutex_lock(&bdev->bd_fsfreeze_mutex);
 224        if (++bdev->bd_fsfreeze_count > 1) {
 225                /*
 226                 * We don't even need to grab a reference - the first call
 227                 * to freeze_bdev grab an active reference and only the last
 228                 * thaw_bdev drops it.
 229                 */
 230                sb = get_super(bdev);
 231                drop_super(sb);
 232                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 233                return sb;
 234        }
 235
 236        sb = get_active_super(bdev);
 237        if (!sb)
 238                goto out;
 239        error = freeze_super(sb);
 240        if (error) {
 241                deactivate_super(sb);
 242                bdev->bd_fsfreeze_count--;
 243                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 244                return ERR_PTR(error);
 245        }
 246        deactivate_super(sb);
 247 out:
 248        sync_blockdev(bdev);
 249        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 250        return sb;      /* thaw_bdev releases s->s_umount */
 251}
 252EXPORT_SYMBOL(freeze_bdev);
 253
 254/**
 255 * thaw_bdev  -- unlock filesystem
 256 * @bdev:       blockdevice to unlock
 257 * @sb:         associated superblock
 258 *
 259 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 260 */
 261int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 262{
 263        int error = -EINVAL;
 264
 265        mutex_lock(&bdev->bd_fsfreeze_mutex);
 266        if (!bdev->bd_fsfreeze_count)
 267                goto out;
 268
 269        error = 0;
 270        if (--bdev->bd_fsfreeze_count > 0)
 271                goto out;
 272
 273        if (!sb)
 274                goto out;
 275
 276        error = thaw_super(sb);
 277        if (error) {
 278                bdev->bd_fsfreeze_count++;
 279                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 280                return error;
 281        }
 282out:
 283        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 284        return 0;
 285}
 286EXPORT_SYMBOL(thaw_bdev);
 287
 288static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 289{
 290        return block_write_full_page(page, blkdev_get_block, wbc);
 291}
 292
 293static int blkdev_readpage(struct file * file, struct page * page)
 294{
 295        return block_read_full_page(page, blkdev_get_block);
 296}
 297
 298static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 299                        loff_t pos, unsigned len, unsigned flags,
 300                        struct page **pagep, void **fsdata)
 301{
 302        return block_write_begin(mapping, pos, len, flags, pagep,
 303                                 blkdev_get_block);
 304}
 305
 306static int blkdev_write_end(struct file *file, struct address_space *mapping,
 307                        loff_t pos, unsigned len, unsigned copied,
 308                        struct page *page, void *fsdata)
 309{
 310        int ret;
 311        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 312
 313        unlock_page(page);
 314        page_cache_release(page);
 315
 316        return ret;
 317}
 318
 319/*
 320 * private llseek:
 321 * for a block special file file_inode(file)->i_size is zero
 322 * so we compute the size by hand (just as in block_read/write above)
 323 */
 324static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 325{
 326        struct inode *bd_inode = file->f_mapping->host;
 327        loff_t size;
 328        loff_t retval;
 329
 330        mutex_lock(&bd_inode->i_mutex);
 331        size = i_size_read(bd_inode);
 332
 333        retval = -EINVAL;
 334        switch (whence) {
 335                case SEEK_END:
 336                        offset += size;
 337                        break;
 338                case SEEK_CUR:
 339                        offset += file->f_pos;
 340                case SEEK_SET:
 341                        break;
 342                default:
 343                        goto out;
 344        }
 345        if (offset >= 0 && offset <= size) {
 346                if (offset != file->f_pos) {
 347                        file->f_pos = offset;
 348                }
 349                retval = offset;
 350        }
 351out:
 352        mutex_unlock(&bd_inode->i_mutex);
 353        return retval;
 354}
 355        
 356int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 357{
 358        struct inode *bd_inode = filp->f_mapping->host;
 359        struct block_device *bdev = I_BDEV(bd_inode);
 360        int error;
 361        
 362        error = filemap_write_and_wait_range(filp->f_mapping, start, end);
 363        if (error)
 364                return error;
 365
 366        /*
 367         * There is no need to serialise calls to blkdev_issue_flush with
 368         * i_mutex and doing so causes performance issues with concurrent
 369         * O_SYNC writers to a block device.
 370         */
 371        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 372        if (error == -EOPNOTSUPP)
 373                error = 0;
 374
 375        return error;
 376}
 377EXPORT_SYMBOL(blkdev_fsync);
 378
 379/*
 380 * pseudo-fs
 381 */
 382
 383static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 384static struct kmem_cache * bdev_cachep __read_mostly;
 385
 386static struct inode *bdev_alloc_inode(struct super_block *sb)
 387{
 388        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 389        if (!ei)
 390                return NULL;
 391        return &ei->vfs_inode;
 392}
 393
 394static void bdev_i_callback(struct rcu_head *head)
 395{
 396        struct inode *inode = container_of(head, struct inode, i_rcu);
 397        struct bdev_inode *bdi = BDEV_I(inode);
 398
 399        kmem_cache_free(bdev_cachep, bdi);
 400}
 401
 402static void bdev_destroy_inode(struct inode *inode)
 403{
 404        call_rcu(&inode->i_rcu, bdev_i_callback);
 405}
 406
 407static void init_once(void *foo)
 408{
 409        struct bdev_inode *ei = (struct bdev_inode *) foo;
 410        struct block_device *bdev = &ei->bdev;
 411
 412        memset(bdev, 0, sizeof(*bdev));
 413        mutex_init(&bdev->bd_mutex);
 414        INIT_LIST_HEAD(&bdev->bd_inodes);
 415        INIT_LIST_HEAD(&bdev->bd_list);
 416#ifdef CONFIG_SYSFS
 417        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 418#endif
 419        inode_init_once(&ei->vfs_inode);
 420        /* Initialize mutex for freeze. */
 421        mutex_init(&bdev->bd_fsfreeze_mutex);
 422}
 423
 424static inline void __bd_forget(struct inode *inode)
 425{
 426        list_del_init(&inode->i_devices);
 427        inode->i_bdev = NULL;
 428        inode->i_mapping = &inode->i_data;
 429}
 430
 431static void bdev_evict_inode(struct inode *inode)
 432{
 433        struct block_device *bdev = &BDEV_I(inode)->bdev;
 434        struct list_head *p;
 435        truncate_inode_pages(&inode->i_data, 0);
 436        invalidate_inode_buffers(inode); /* is it needed here? */
 437        clear_inode(inode);
 438        spin_lock(&bdev_lock);
 439        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 440                __bd_forget(list_entry(p, struct inode, i_devices));
 441        }
 442        list_del_init(&bdev->bd_list);
 443        spin_unlock(&bdev_lock);
 444}
 445
 446static const struct super_operations bdev_sops = {
 447        .statfs = simple_statfs,
 448        .alloc_inode = bdev_alloc_inode,
 449        .destroy_inode = bdev_destroy_inode,
 450        .drop_inode = generic_delete_inode,
 451        .evict_inode = bdev_evict_inode,
 452};
 453
 454static struct dentry *bd_mount(struct file_system_type *fs_type,
 455        int flags, const char *dev_name, void *data)
 456{
 457        return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
 458}
 459
 460static struct file_system_type bd_type = {
 461        .name           = "bdev",
 462        .mount          = bd_mount,
 463        .kill_sb        = kill_anon_super,
 464};
 465
 466static struct super_block *blockdev_superblock __read_mostly;
 467
 468void __init bdev_cache_init(void)
 469{
 470        int err;
 471        static struct vfsmount *bd_mnt;
 472
 473        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 474                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 475                                SLAB_MEM_SPREAD|SLAB_PANIC),
 476                        init_once);
 477        err = register_filesystem(&bd_type);
 478        if (err)
 479                panic("Cannot register bdev pseudo-fs");
 480        bd_mnt = kern_mount(&bd_type);
 481        if (IS_ERR(bd_mnt))
 482                panic("Cannot create bdev pseudo-fs");
 483        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 484}
 485
 486/*
 487 * Most likely _very_ bad one - but then it's hardly critical for small
 488 * /dev and can be fixed when somebody will need really large one.
 489 * Keep in mind that it will be fed through icache hash function too.
 490 */
 491static inline unsigned long hash(dev_t dev)
 492{
 493        return MAJOR(dev)+MINOR(dev);
 494}
 495
 496static int bdev_test(struct inode *inode, void *data)
 497{
 498        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 499}
 500
 501static int bdev_set(struct inode *inode, void *data)
 502{
 503        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 504        return 0;
 505}
 506
 507static LIST_HEAD(all_bdevs);
 508
 509struct block_device *bdget(dev_t dev)
 510{
 511        struct block_device *bdev;
 512        struct inode *inode;
 513
 514        inode = iget5_locked(blockdev_superblock, hash(dev),
 515                        bdev_test, bdev_set, &dev);
 516
 517        if (!inode)
 518                return NULL;
 519
 520        bdev = &BDEV_I(inode)->bdev;
 521
 522        if (inode->i_state & I_NEW) {
 523                bdev->bd_contains = NULL;
 524                bdev->bd_super = NULL;
 525                bdev->bd_inode = inode;
 526                bdev->bd_block_size = (1 << inode->i_blkbits);
 527                bdev->bd_part_count = 0;
 528                bdev->bd_invalidated = 0;
 529                inode->i_mode = S_IFBLK;
 530                inode->i_rdev = dev;
 531                inode->i_bdev = bdev;
 532                inode->i_data.a_ops = &def_blk_aops;
 533                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 534                inode->i_data.backing_dev_info = &default_backing_dev_info;
 535                spin_lock(&bdev_lock);
 536                list_add(&bdev->bd_list, &all_bdevs);
 537                spin_unlock(&bdev_lock);
 538                unlock_new_inode(inode);
 539        }
 540        return bdev;
 541}
 542
 543EXPORT_SYMBOL(bdget);
 544
 545/**
 546 * bdgrab -- Grab a reference to an already referenced block device
 547 * @bdev:       Block device to grab a reference to.
 548 */
 549struct block_device *bdgrab(struct block_device *bdev)
 550{
 551        ihold(bdev->bd_inode);
 552        return bdev;
 553}
 554EXPORT_SYMBOL(bdgrab);
 555
 556long nr_blockdev_pages(void)
 557{
 558        struct block_device *bdev;
 559        long ret = 0;
 560        spin_lock(&bdev_lock);
 561        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 562                ret += bdev->bd_inode->i_mapping->nrpages;
 563        }
 564        spin_unlock(&bdev_lock);
 565        return ret;
 566}
 567
 568void bdput(struct block_device *bdev)
 569{
 570        iput(bdev->bd_inode);
 571}
 572
 573EXPORT_SYMBOL(bdput);
 574 
 575static struct block_device *bd_acquire(struct inode *inode)
 576{
 577        struct block_device *bdev;
 578
 579        spin_lock(&bdev_lock);
 580        bdev = inode->i_bdev;
 581        if (bdev) {
 582                ihold(bdev->bd_inode);
 583                spin_unlock(&bdev_lock);
 584                return bdev;
 585        }
 586        spin_unlock(&bdev_lock);
 587
 588        bdev = bdget(inode->i_rdev);
 589        if (bdev) {
 590                spin_lock(&bdev_lock);
 591                if (!inode->i_bdev) {
 592                        /*
 593                         * We take an additional reference to bd_inode,
 594                         * and it's released in clear_inode() of inode.
 595                         * So, we can access it via ->i_mapping always
 596                         * without igrab().
 597                         */
 598                        ihold(bdev->bd_inode);
 599                        inode->i_bdev = bdev;
 600                        inode->i_mapping = bdev->bd_inode->i_mapping;
 601                        list_add(&inode->i_devices, &bdev->bd_inodes);
 602                }
 603                spin_unlock(&bdev_lock);
 604        }
 605        return bdev;
 606}
 607
 608static inline int sb_is_blkdev_sb(struct super_block *sb)
 609{
 610        return sb == blockdev_superblock;
 611}
 612
 613/* Call when you free inode */
 614
 615void bd_forget(struct inode *inode)
 616{
 617        struct block_device *bdev = NULL;
 618
 619        spin_lock(&bdev_lock);
 620        if (inode->i_bdev) {
 621                if (!sb_is_blkdev_sb(inode->i_sb))
 622                        bdev = inode->i_bdev;
 623                __bd_forget(inode);
 624        }
 625        spin_unlock(&bdev_lock);
 626
 627        if (bdev)
 628                iput(bdev->bd_inode);
 629}
 630
 631/**
 632 * bd_may_claim - test whether a block device can be claimed
 633 * @bdev: block device of interest
 634 * @whole: whole block device containing @bdev, may equal @bdev
 635 * @holder: holder trying to claim @bdev
 636 *
 637 * Test whether @bdev can be claimed by @holder.
 638 *
 639 * CONTEXT:
 640 * spin_lock(&bdev_lock).
 641 *
 642 * RETURNS:
 643 * %true if @bdev can be claimed, %false otherwise.
 644 */
 645static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 646                         void *holder)
 647{
 648        if (bdev->bd_holder == holder)
 649                return true;     /* already a holder */
 650        else if (bdev->bd_holder != NULL)
 651                return false;    /* held by someone else */
 652        else if (bdev->bd_contains == bdev)
 653                return true;     /* is a whole device which isn't held */
 654
 655        else if (whole->bd_holder == bd_may_claim)
 656                return true;     /* is a partition of a device that is being partitioned */
 657        else if (whole->bd_holder != NULL)
 658                return false;    /* is a partition of a held device */
 659        else
 660                return true;     /* is a partition of an un-held device */
 661}
 662
 663/**
 664 * bd_prepare_to_claim - prepare to claim a block device
 665 * @bdev: block device of interest
 666 * @whole: the whole device containing @bdev, may equal @bdev
 667 * @holder: holder trying to claim @bdev
 668 *
 669 * Prepare to claim @bdev.  This function fails if @bdev is already
 670 * claimed by another holder and waits if another claiming is in
 671 * progress.  This function doesn't actually claim.  On successful
 672 * return, the caller has ownership of bd_claiming and bd_holder[s].
 673 *
 674 * CONTEXT:
 675 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 676 * it multiple times.
 677 *
 678 * RETURNS:
 679 * 0 if @bdev can be claimed, -EBUSY otherwise.
 680 */
 681static int bd_prepare_to_claim(struct block_device *bdev,
 682                               struct block_device *whole, void *holder)
 683{
 684retry:
 685        /* if someone else claimed, fail */
 686        if (!bd_may_claim(bdev, whole, holder))
 687                return -EBUSY;
 688
 689        /* if claiming is already in progress, wait for it to finish */
 690        if (whole->bd_claiming) {
 691                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 692                DEFINE_WAIT(wait);
 693
 694                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 695                spin_unlock(&bdev_lock);
 696                schedule();
 697                finish_wait(wq, &wait);
 698                spin_lock(&bdev_lock);
 699                goto retry;
 700        }
 701
 702        /* yay, all mine */
 703        return 0;
 704}
 705
 706/**
 707 * bd_start_claiming - start claiming a block device
 708 * @bdev: block device of interest
 709 * @holder: holder trying to claim @bdev
 710 *
 711 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 712 * exclusively and mark that an exclusive open is in progress.  Each
 713 * successful call to this function must be matched with a call to
 714 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 715 * fail).
 716 *
 717 * This function is used to gain exclusive access to the block device
 718 * without actually causing other exclusive open attempts to fail. It
 719 * should be used when the open sequence itself requires exclusive
 720 * access but may subsequently fail.
 721 *
 722 * CONTEXT:
 723 * Might sleep.
 724 *
 725 * RETURNS:
 726 * Pointer to the block device containing @bdev on success, ERR_PTR()
 727 * value on failure.
 728 */
 729static struct block_device *bd_start_claiming(struct block_device *bdev,
 730                                              void *holder)
 731{
 732        struct gendisk *disk;
 733        struct block_device *whole;
 734        int partno, err;
 735
 736        might_sleep();
 737
 738        /*
 739         * @bdev might not have been initialized properly yet, look up
 740         * and grab the outer block device the hard way.
 741         */
 742        disk = get_gendisk(bdev->bd_dev, &partno);
 743        if (!disk)
 744                return ERR_PTR(-ENXIO);
 745
 746        /*
 747         * Normally, @bdev should equal what's returned from bdget_disk()
 748         * if partno is 0; however, some drivers (floppy) use multiple
 749         * bdev's for the same physical device and @bdev may be one of the
 750         * aliases.  Keep @bdev if partno is 0.  This means claimer
 751         * tracking is broken for those devices but it has always been that
 752         * way.
 753         */
 754        if (partno)
 755                whole = bdget_disk(disk, 0);
 756        else
 757                whole = bdgrab(bdev);
 758
 759        module_put(disk->fops->owner);
 760        put_disk(disk);
 761        if (!whole)
 762                return ERR_PTR(-ENOMEM);
 763
 764        /* prepare to claim, if successful, mark claiming in progress */
 765        spin_lock(&bdev_lock);
 766
 767        err = bd_prepare_to_claim(bdev, whole, holder);
 768        if (err == 0) {
 769                whole->bd_claiming = holder;
 770                spin_unlock(&bdev_lock);
 771                return whole;
 772        } else {
 773                spin_unlock(&bdev_lock);
 774                bdput(whole);
 775                return ERR_PTR(err);
 776        }
 777}
 778
 779#ifdef CONFIG_SYSFS
 780struct bd_holder_disk {
 781        struct list_head        list;
 782        struct gendisk          *disk;
 783        int                     refcnt;
 784};
 785
 786static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 787                                                  struct gendisk *disk)
 788{
 789        struct bd_holder_disk *holder;
 790
 791        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 792                if (holder->disk == disk)
 793                        return holder;
 794        return NULL;
 795}
 796
 797static int add_symlink(struct kobject *from, struct kobject *to)
 798{
 799        return sysfs_create_link(from, to, kobject_name(to));
 800}
 801
 802static void del_symlink(struct kobject *from, struct kobject *to)
 803{
 804        sysfs_remove_link(from, kobject_name(to));
 805}
 806
 807/**
 808 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 809 * @bdev: the claimed slave bdev
 810 * @disk: the holding disk
 811 *
 812 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 813 *
 814 * This functions creates the following sysfs symlinks.
 815 *
 816 * - from "slaves" directory of the holder @disk to the claimed @bdev
 817 * - from "holders" directory of the @bdev to the holder @disk
 818 *
 819 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
 820 * passed to bd_link_disk_holder(), then:
 821 *
 822 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
 823 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 824 *
 825 * The caller must have claimed @bdev before calling this function and
 826 * ensure that both @bdev and @disk are valid during the creation and
 827 * lifetime of these symlinks.
 828 *
 829 * CONTEXT:
 830 * Might sleep.
 831 *
 832 * RETURNS:
 833 * 0 on success, -errno on failure.
 834 */
 835int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 836{
 837        struct bd_holder_disk *holder;
 838        int ret = 0;
 839
 840        mutex_lock(&bdev->bd_mutex);
 841
 842        WARN_ON_ONCE(!bdev->bd_holder);
 843
 844        /* FIXME: remove the following once add_disk() handles errors */
 845        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
 846                goto out_unlock;
 847
 848        holder = bd_find_holder_disk(bdev, disk);
 849        if (holder) {
 850                holder->refcnt++;
 851                goto out_unlock;
 852        }
 853
 854        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 855        if (!holder) {
 856                ret = -ENOMEM;
 857                goto out_unlock;
 858        }
 859
 860        INIT_LIST_HEAD(&holder->list);
 861        holder->disk = disk;
 862        holder->refcnt = 1;
 863
 864        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 865        if (ret)
 866                goto out_free;
 867
 868        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
 869        if (ret)
 870                goto out_del;
 871        /*
 872         * bdev could be deleted beneath us which would implicitly destroy
 873         * the holder directory.  Hold on to it.
 874         */
 875        kobject_get(bdev->bd_part->holder_dir);
 876
 877        list_add(&holder->list, &bdev->bd_holder_disks);
 878        goto out_unlock;
 879
 880out_del:
 881        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 882out_free:
 883        kfree(holder);
 884out_unlock:
 885        mutex_unlock(&bdev->bd_mutex);
 886        return ret;
 887}
 888EXPORT_SYMBOL_GPL(bd_link_disk_holder);
 889
 890/**
 891 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
 892 * @bdev: the calimed slave bdev
 893 * @disk: the holding disk
 894 *
 895 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 896 *
 897 * CONTEXT:
 898 * Might sleep.
 899 */
 900void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 901{
 902        struct bd_holder_disk *holder;
 903
 904        mutex_lock(&bdev->bd_mutex);
 905
 906        holder = bd_find_holder_disk(bdev, disk);
 907
 908        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 909                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 910                del_symlink(bdev->bd_part->holder_dir,
 911                            &disk_to_dev(disk)->kobj);
 912                kobject_put(bdev->bd_part->holder_dir);
 913                list_del_init(&holder->list);
 914                kfree(holder);
 915        }
 916
 917        mutex_unlock(&bdev->bd_mutex);
 918}
 919EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
 920#endif
 921
 922/**
 923 * flush_disk - invalidates all buffer-cache entries on a disk
 924 *
 925 * @bdev:      struct block device to be flushed
 926 * @kill_dirty: flag to guide handling of dirty inodes
 927 *
 928 * Invalidates all buffer-cache entries on a disk. It should be called
 929 * when a disk has been changed -- either by a media change or online
 930 * resize.
 931 */
 932static void flush_disk(struct block_device *bdev, bool kill_dirty)
 933{
 934        if (__invalidate_device(bdev, kill_dirty)) {
 935                char name[BDEVNAME_SIZE] = "";
 936
 937                if (bdev->bd_disk)
 938                        disk_name(bdev->bd_disk, 0, name);
 939                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 940                       "resized disk %s\n", name);
 941        }
 942
 943        if (!bdev->bd_disk)
 944                return;
 945        if (disk_part_scan_enabled(bdev->bd_disk))
 946                bdev->bd_invalidated = 1;
 947}
 948
 949/**
 950 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 951 * @disk: struct gendisk to check
 952 * @bdev: struct bdev to adjust.
 953 *
 954 * This routine checks to see if the bdev size does not match the disk size
 955 * and adjusts it if it differs.
 956 */
 957void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 958{
 959        loff_t disk_size, bdev_size;
 960
 961        disk_size = (loff_t)get_capacity(disk) << 9;
 962        bdev_size = i_size_read(bdev->bd_inode);
 963        if (disk_size != bdev_size) {
 964                char name[BDEVNAME_SIZE];
 965
 966                disk_name(disk, 0, name);
 967                printk(KERN_INFO
 968                       "%s: detected capacity change from %lld to %lld\n",
 969                       name, bdev_size, disk_size);
 970                i_size_write(bdev->bd_inode, disk_size);
 971                flush_disk(bdev, false);
 972        }
 973}
 974EXPORT_SYMBOL(check_disk_size_change);
 975
 976/**
 977 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 978 * @disk: struct gendisk to be revalidated
 979 *
 980 * This routine is a wrapper for lower-level driver's revalidate_disk
 981 * call-backs.  It is used to do common pre and post operations needed
 982 * for all revalidate_disk operations.
 983 */
 984int revalidate_disk(struct gendisk *disk)
 985{
 986        struct block_device *bdev;
 987        int ret = 0;
 988
 989        if (disk->fops->revalidate_disk)
 990                ret = disk->fops->revalidate_disk(disk);
 991
 992        bdev = bdget_disk(disk, 0);
 993        if (!bdev)
 994                return ret;
 995
 996        mutex_lock(&bdev->bd_mutex);
 997        check_disk_size_change(disk, bdev);
 998        bdev->bd_invalidated = 0;
 999        mutex_unlock(&bdev->bd_mutex);
1000        bdput(bdev);
1001        return ret;
1002}
1003EXPORT_SYMBOL(revalidate_disk);
1004
1005/*
1006 * This routine checks whether a removable media has been changed,
1007 * and invalidates all buffer-cache-entries in that case. This
1008 * is a relatively slow routine, so we have to try to minimize using
1009 * it. Thus it is called only upon a 'mount' or 'open'. This
1010 * is the best way of combining speed and utility, I think.
1011 * People changing diskettes in the middle of an operation deserve
1012 * to lose :-)
1013 */
1014int check_disk_change(struct block_device *bdev)
1015{
1016        struct gendisk *disk = bdev->bd_disk;
1017        const struct block_device_operations *bdops = disk->fops;
1018        unsigned int events;
1019
1020        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1021                                   DISK_EVENT_EJECT_REQUEST);
1022        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1023                return 0;
1024
1025        flush_disk(bdev, true);
1026        if (bdops->revalidate_disk)
1027                bdops->revalidate_disk(bdev->bd_disk);
1028        return 1;
1029}
1030
1031EXPORT_SYMBOL(check_disk_change);
1032
1033void bd_set_size(struct block_device *bdev, loff_t size)
1034{
1035        unsigned bsize = bdev_logical_block_size(bdev);
1036
1037        mutex_lock(&bdev->bd_inode->i_mutex);
1038        i_size_write(bdev->bd_inode, size);
1039        mutex_unlock(&bdev->bd_inode->i_mutex);
1040        while (bsize < PAGE_CACHE_SIZE) {
1041                if (size & bsize)
1042                        break;
1043                bsize <<= 1;
1044        }
1045        bdev->bd_block_size = bsize;
1046        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1047}
1048EXPORT_SYMBOL(bd_set_size);
1049
1050static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1051
1052/*
1053 * bd_mutex locking:
1054 *
1055 *  mutex_lock(part->bd_mutex)
1056 *    mutex_lock_nested(whole->bd_mutex, 1)
1057 */
1058
1059static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1060{
1061        struct gendisk *disk;
1062        struct module *owner;
1063        int ret;
1064        int partno;
1065        int perm = 0;
1066
1067        if (mode & FMODE_READ)
1068                perm |= MAY_READ;
1069        if (mode & FMODE_WRITE)
1070                perm |= MAY_WRITE;
1071        /*
1072         * hooks: /n/, see "layering violations".
1073         */
1074        if (!for_part) {
1075                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1076                if (ret != 0) {
1077                        bdput(bdev);
1078                        return ret;
1079                }
1080        }
1081
1082 restart:
1083
1084        ret = -ENXIO;
1085        disk = get_gendisk(bdev->bd_dev, &partno);
1086        if (!disk)
1087                goto out;
1088        owner = disk->fops->owner;
1089
1090        disk_block_events(disk);
1091        mutex_lock_nested(&bdev->bd_mutex, for_part);
1092        if (!bdev->bd_openers) {
1093                bdev->bd_disk = disk;
1094                bdev->bd_queue = disk->queue;
1095                bdev->bd_contains = bdev;
1096                if (!partno) {
1097                        struct backing_dev_info *bdi;
1098
1099                        ret = -ENXIO;
1100                        bdev->bd_part = disk_get_part(disk, partno);
1101                        if (!bdev->bd_part)
1102                                goto out_clear;
1103
1104                        ret = 0;
1105                        if (disk->fops->open) {
1106                                ret = disk->fops->open(bdev, mode);
1107                                if (ret == -ERESTARTSYS) {
1108                                        /* Lost a race with 'disk' being
1109                                         * deleted, try again.
1110                                         * See md.c
1111                                         */
1112                                        disk_put_part(bdev->bd_part);
1113                                        bdev->bd_part = NULL;
1114                                        bdev->bd_disk = NULL;
1115                                        bdev->bd_queue = NULL;
1116                                        mutex_unlock(&bdev->bd_mutex);
1117                                        disk_unblock_events(disk);
1118                                        put_disk(disk);
1119                                        module_put(owner);
1120                                        goto restart;
1121                                }
1122                        }
1123
1124                        if (!ret) {
1125                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1126                                bdi = blk_get_backing_dev_info(bdev);
1127                                if (bdi == NULL)
1128                                        bdi = &default_backing_dev_info;
1129                                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1130                        }
1131
1132                        /*
1133                         * If the device is invalidated, rescan partition
1134                         * if open succeeded or failed with -ENOMEDIUM.
1135                         * The latter is necessary to prevent ghost
1136                         * partitions on a removed medium.
1137                         */
1138                        if (bdev->bd_invalidated) {
1139                                if (!ret)
1140                                        rescan_partitions(disk, bdev);
1141                                else if (ret == -ENOMEDIUM)
1142                                        invalidate_partitions(disk, bdev);
1143                        }
1144                        if (ret)
1145                                goto out_clear;
1146                } else {
1147                        struct block_device *whole;
1148                        whole = bdget_disk(disk, 0);
1149                        ret = -ENOMEM;
1150                        if (!whole)
1151                                goto out_clear;
1152                        BUG_ON(for_part);
1153                        ret = __blkdev_get(whole, mode, 1);
1154                        if (ret)
1155                                goto out_clear;
1156                        bdev->bd_contains = whole;
1157                        bdev_inode_switch_bdi(bdev->bd_inode,
1158                                whole->bd_inode->i_data.backing_dev_info);
1159                        bdev->bd_part = disk_get_part(disk, partno);
1160                        if (!(disk->flags & GENHD_FL_UP) ||
1161                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1162                                ret = -ENXIO;
1163                                goto out_clear;
1164                        }
1165                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1166                }
1167        } else {
1168                if (bdev->bd_contains == bdev) {
1169                        ret = 0;
1170                        if (bdev->bd_disk->fops->open)
1171                                ret = bdev->bd_disk->fops->open(bdev, mode);
1172                        /* the same as first opener case, read comment there */
1173                        if (bdev->bd_invalidated) {
1174                                if (!ret)
1175                                        rescan_partitions(bdev->bd_disk, bdev);
1176                                else if (ret == -ENOMEDIUM)
1177                                        invalidate_partitions(bdev->bd_disk, bdev);
1178                        }
1179                        if (ret)
1180                                goto out_unlock_bdev;
1181                }
1182                /* only one opener holds refs to the module and disk */
1183                put_disk(disk);
1184                module_put(owner);
1185        }
1186        bdev->bd_openers++;
1187        if (for_part)
1188                bdev->bd_part_count++;
1189        mutex_unlock(&bdev->bd_mutex);
1190        disk_unblock_events(disk);
1191        return 0;
1192
1193 out_clear:
1194        disk_put_part(bdev->bd_part);
1195        bdev->bd_disk = NULL;
1196        bdev->bd_part = NULL;
1197        bdev->bd_queue = NULL;
1198        bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1199        if (bdev != bdev->bd_contains)
1200                __blkdev_put(bdev->bd_contains, mode, 1);
1201        bdev->bd_contains = NULL;
1202 out_unlock_bdev:
1203        mutex_unlock(&bdev->bd_mutex);
1204        disk_unblock_events(disk);
1205        put_disk(disk);
1206        module_put(owner);
1207 out:
1208        bdput(bdev);
1209
1210        return ret;
1211}
1212
1213/**
1214 * blkdev_get - open a block device
1215 * @bdev: block_device to open
1216 * @mode: FMODE_* mask
1217 * @holder: exclusive holder identifier
1218 *
1219 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1220 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1221 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1222 *
1223 * On success, the reference count of @bdev is unchanged.  On failure,
1224 * @bdev is put.
1225 *
1226 * CONTEXT:
1227 * Might sleep.
1228 *
1229 * RETURNS:
1230 * 0 on success, -errno on failure.
1231 */
1232int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1233{
1234        struct block_device *whole = NULL;
1235        int res;
1236
1237        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1238
1239        if ((mode & FMODE_EXCL) && holder) {
1240                whole = bd_start_claiming(bdev, holder);
1241                if (IS_ERR(whole)) {
1242                        bdput(bdev);
1243                        return PTR_ERR(whole);
1244                }
1245        }
1246
1247        res = __blkdev_get(bdev, mode, 0);
1248
1249        if (whole) {
1250                struct gendisk *disk = whole->bd_disk;
1251
1252                /* finish claiming */
1253                mutex_lock(&bdev->bd_mutex);
1254                spin_lock(&bdev_lock);
1255
1256                if (!res) {
1257                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1258                        /*
1259                         * Note that for a whole device bd_holders
1260                         * will be incremented twice, and bd_holder
1261                         * will be set to bd_may_claim before being
1262                         * set to holder
1263                         */
1264                        whole->bd_holders++;
1265                        whole->bd_holder = bd_may_claim;
1266                        bdev->bd_holders++;
1267                        bdev->bd_holder = holder;
1268                }
1269
1270                /* tell others that we're done */
1271                BUG_ON(whole->bd_claiming != holder);
1272                whole->bd_claiming = NULL;
1273                wake_up_bit(&whole->bd_claiming, 0);
1274
1275                spin_unlock(&bdev_lock);
1276
1277                /*
1278                 * Block event polling for write claims if requested.  Any
1279                 * write holder makes the write_holder state stick until
1280                 * all are released.  This is good enough and tracking
1281                 * individual writeable reference is too fragile given the
1282                 * way @mode is used in blkdev_get/put().
1283                 */
1284                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1285                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1286                        bdev->bd_write_holder = true;
1287                        disk_block_events(disk);
1288                }
1289
1290                mutex_unlock(&bdev->bd_mutex);
1291                bdput(whole);
1292        }
1293
1294        return res;
1295}
1296EXPORT_SYMBOL(blkdev_get);
1297
1298/**
1299 * blkdev_get_by_path - open a block device by name
1300 * @path: path to the block device to open
1301 * @mode: FMODE_* mask
1302 * @holder: exclusive holder identifier
1303 *
1304 * Open the blockdevice described by the device file at @path.  @mode
1305 * and @holder are identical to blkdev_get().
1306 *
1307 * On success, the returned block_device has reference count of one.
1308 *
1309 * CONTEXT:
1310 * Might sleep.
1311 *
1312 * RETURNS:
1313 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1314 */
1315struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1316                                        void *holder)
1317{
1318        struct block_device *bdev;
1319        int err;
1320
1321        bdev = lookup_bdev(path);
1322        if (IS_ERR(bdev))
1323                return bdev;
1324
1325        err = blkdev_get(bdev, mode, holder);
1326        if (err)
1327                return ERR_PTR(err);
1328
1329        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1330                blkdev_put(bdev, mode);
1331                return ERR_PTR(-EACCES);
1332        }
1333
1334        return bdev;
1335}
1336EXPORT_SYMBOL(blkdev_get_by_path);
1337
1338/**
1339 * blkdev_get_by_dev - open a block device by device number
1340 * @dev: device number of block device to open
1341 * @mode: FMODE_* mask
1342 * @holder: exclusive holder identifier
1343 *
1344 * Open the blockdevice described by device number @dev.  @mode and
1345 * @holder are identical to blkdev_get().
1346 *
1347 * Use it ONLY if you really do not have anything better - i.e. when
1348 * you are behind a truly sucky interface and all you are given is a
1349 * device number.  _Never_ to be used for internal purposes.  If you
1350 * ever need it - reconsider your API.
1351 *
1352 * On success, the returned block_device has reference count of one.
1353 *
1354 * CONTEXT:
1355 * Might sleep.
1356 *
1357 * RETURNS:
1358 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1359 */
1360struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1361{
1362        struct block_device *bdev;
1363        int err;
1364
1365        bdev = bdget(dev);
1366        if (!bdev)
1367                return ERR_PTR(-ENOMEM);
1368
1369        err = blkdev_get(bdev, mode, holder);
1370        if (err)
1371                return ERR_PTR(err);
1372
1373        return bdev;
1374}
1375EXPORT_SYMBOL(blkdev_get_by_dev);
1376
1377static int blkdev_open(struct inode * inode, struct file * filp)
1378{
1379        struct block_device *bdev;
1380
1381        /*
1382         * Preserve backwards compatibility and allow large file access
1383         * even if userspace doesn't ask for it explicitly. Some mkfs
1384         * binary needs it. We might want to drop this workaround
1385         * during an unstable branch.
1386         */
1387        filp->f_flags |= O_LARGEFILE;
1388
1389        if (filp->f_flags & O_NDELAY)
1390                filp->f_mode |= FMODE_NDELAY;
1391        if (filp->f_flags & O_EXCL)
1392                filp->f_mode |= FMODE_EXCL;
1393        if ((filp->f_flags & O_ACCMODE) == 3)
1394                filp->f_mode |= FMODE_WRITE_IOCTL;
1395
1396        bdev = bd_acquire(inode);
1397        if (bdev == NULL)
1398                return -ENOMEM;
1399
1400        filp->f_mapping = bdev->bd_inode->i_mapping;
1401
1402        return blkdev_get(bdev, filp->f_mode, filp);
1403}
1404
1405static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1406{
1407        int ret = 0;
1408        struct gendisk *disk = bdev->bd_disk;
1409        struct block_device *victim = NULL;
1410
1411        mutex_lock_nested(&bdev->bd_mutex, for_part);
1412        if (for_part)
1413                bdev->bd_part_count--;
1414
1415        if (!--bdev->bd_openers) {
1416                WARN_ON_ONCE(bdev->bd_holders);
1417                sync_blockdev(bdev);
1418                kill_bdev(bdev);
1419                /* ->release can cause the old bdi to disappear,
1420                 * so must switch it out first
1421                 */
1422                bdev_inode_switch_bdi(bdev->bd_inode,
1423                                        &default_backing_dev_info);
1424        }
1425        if (bdev->bd_contains == bdev) {
1426                if (disk->fops->release)
1427                        ret = disk->fops->release(disk, mode);
1428        }
1429        if (!bdev->bd_openers) {
1430                struct module *owner = disk->fops->owner;
1431
1432                disk_put_part(bdev->bd_part);
1433                bdev->bd_part = NULL;
1434                bdev->bd_disk = NULL;
1435                if (bdev != bdev->bd_contains)
1436                        victim = bdev->bd_contains;
1437                bdev->bd_contains = NULL;
1438
1439                put_disk(disk);
1440                module_put(owner);
1441        }
1442        mutex_unlock(&bdev->bd_mutex);
1443        bdput(bdev);
1444        if (victim)
1445                __blkdev_put(victim, mode, 1);
1446        return ret;
1447}
1448
1449int blkdev_put(struct block_device *bdev, fmode_t mode)
1450{
1451        mutex_lock(&bdev->bd_mutex);
1452
1453        if (mode & FMODE_EXCL) {
1454                bool bdev_free;
1455
1456                /*
1457                 * Release a claim on the device.  The holder fields
1458                 * are protected with bdev_lock.  bd_mutex is to
1459                 * synchronize disk_holder unlinking.
1460                 */
1461                spin_lock(&bdev_lock);
1462
1463                WARN_ON_ONCE(--bdev->bd_holders < 0);
1464                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1465
1466                /* bd_contains might point to self, check in a separate step */
1467                if ((bdev_free = !bdev->bd_holders))
1468                        bdev->bd_holder = NULL;
1469                if (!bdev->bd_contains->bd_holders)
1470                        bdev->bd_contains->bd_holder = NULL;
1471
1472                spin_unlock(&bdev_lock);
1473
1474                /*
1475                 * If this was the last claim, remove holder link and
1476                 * unblock evpoll if it was a write holder.
1477                 */
1478                if (bdev_free && bdev->bd_write_holder) {
1479                        disk_unblock_events(bdev->bd_disk);
1480                        bdev->bd_write_holder = false;
1481                }
1482        }
1483
1484        /*
1485         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1486         * event.  This is to ensure detection of media removal commanded
1487         * from userland - e.g. eject(1).
1488         */
1489        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1490
1491        mutex_unlock(&bdev->bd_mutex);
1492
1493        return __blkdev_put(bdev, mode, 0);
1494}
1495EXPORT_SYMBOL(blkdev_put);
1496
1497static int blkdev_close(struct inode * inode, struct file * filp)
1498{
1499        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1500
1501        return blkdev_put(bdev, filp->f_mode);
1502}
1503
1504static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1505{
1506        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1507        fmode_t mode = file->f_mode;
1508
1509        /*
1510         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1511         * to updated it before every ioctl.
1512         */
1513        if (file->f_flags & O_NDELAY)
1514                mode |= FMODE_NDELAY;
1515        else
1516                mode &= ~FMODE_NDELAY;
1517
1518        return blkdev_ioctl(bdev, mode, cmd, arg);
1519}
1520
1521/*
1522 * Write data to the block device.  Only intended for the block device itself
1523 * and the raw driver which basically is a fake block device.
1524 *
1525 * Does not take i_mutex for the write and thus is not for general purpose
1526 * use.
1527 */
1528ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1529                         unsigned long nr_segs, loff_t pos)
1530{
1531        struct file *file = iocb->ki_filp;
1532        struct blk_plug plug;
1533        ssize_t ret;
1534
1535        BUG_ON(iocb->ki_pos != pos);
1536
1537        blk_start_plug(&plug);
1538        ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1539        if (ret > 0 || ret == -EIOCBQUEUED) {
1540                ssize_t err;
1541
1542                err = generic_write_sync(file, pos, ret);
1543                if (err < 0 && ret > 0)
1544                        ret = err;
1545        }
1546        blk_finish_plug(&plug);
1547        return ret;
1548}
1549EXPORT_SYMBOL_GPL(blkdev_aio_write);
1550
1551static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1552                         unsigned long nr_segs, loff_t pos)
1553{
1554        struct file *file = iocb->ki_filp;
1555        struct inode *bd_inode = file->f_mapping->host;
1556        loff_t size = i_size_read(bd_inode);
1557
1558        if (pos >= size)
1559                return 0;
1560
1561        size -= pos;
1562        if (size < INT_MAX)
1563                nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1564        return generic_file_aio_read(iocb, iov, nr_segs, pos);
1565}
1566
1567/*
1568 * Try to release a page associated with block device when the system
1569 * is under memory pressure.
1570 */
1571static int blkdev_releasepage(struct page *page, gfp_t wait)
1572{
1573        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1574
1575        if (super && super->s_op->bdev_try_to_free_page)
1576                return super->s_op->bdev_try_to_free_page(super, page, wait);
1577
1578        return try_to_free_buffers(page);
1579}
1580
1581static const struct address_space_operations def_blk_aops = {
1582        .readpage       = blkdev_readpage,
1583        .writepage      = blkdev_writepage,
1584        .write_begin    = blkdev_write_begin,
1585        .write_end      = blkdev_write_end,
1586        .writepages     = generic_writepages,
1587        .releasepage    = blkdev_releasepage,
1588        .direct_IO      = blkdev_direct_IO,
1589};
1590
1591const struct file_operations def_blk_fops = {
1592        .open           = blkdev_open,
1593        .release        = blkdev_close,
1594        .llseek         = block_llseek,
1595        .read           = do_sync_read,
1596        .write          = do_sync_write,
1597        .aio_read       = blkdev_aio_read,
1598        .aio_write      = blkdev_aio_write,
1599        .mmap           = generic_file_mmap,
1600        .fsync          = blkdev_fsync,
1601        .unlocked_ioctl = block_ioctl,
1602#ifdef CONFIG_COMPAT
1603        .compat_ioctl   = compat_blkdev_ioctl,
1604#endif
1605        .splice_read    = generic_file_splice_read,
1606        .splice_write   = generic_file_splice_write,
1607};
1608
1609int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1610{
1611        int res;
1612        mm_segment_t old_fs = get_fs();
1613        set_fs(KERNEL_DS);
1614        res = blkdev_ioctl(bdev, 0, cmd, arg);
1615        set_fs(old_fs);
1616        return res;
1617}
1618
1619EXPORT_SYMBOL(ioctl_by_bdev);
1620
1621/**
1622 * lookup_bdev  - lookup a struct block_device by name
1623 * @pathname:   special file representing the block device
1624 *
1625 * Get a reference to the blockdevice at @pathname in the current
1626 * namespace if possible and return it.  Return ERR_PTR(error)
1627 * otherwise.
1628 */
1629struct block_device *lookup_bdev(const char *pathname)
1630{
1631        struct block_device *bdev;
1632        struct inode *inode;
1633        struct path path;
1634        int error;
1635
1636        if (!pathname || !*pathname)
1637                return ERR_PTR(-EINVAL);
1638
1639        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1640        if (error)
1641                return ERR_PTR(error);
1642
1643        inode = path.dentry->d_inode;
1644        error = -ENOTBLK;
1645        if (!S_ISBLK(inode->i_mode))
1646                goto fail;
1647        error = -EACCES;
1648        if (path.mnt->mnt_flags & MNT_NODEV)
1649                goto fail;
1650        error = -ENOMEM;
1651        bdev = bd_acquire(inode);
1652        if (!bdev)
1653                goto fail;
1654out:
1655        path_put(&path);
1656        return bdev;
1657fail:
1658        bdev = ERR_PTR(error);
1659        goto out;
1660}
1661EXPORT_SYMBOL(lookup_bdev);
1662
1663int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1664{
1665        struct super_block *sb = get_super(bdev);
1666        int res = 0;
1667
1668        if (sb) {
1669                /*
1670                 * no need to lock the super, get_super holds the
1671                 * read mutex so the filesystem cannot go away
1672                 * under us (->put_super runs with the write lock
1673                 * hold).
1674                 */
1675                shrink_dcache_sb(sb);
1676                res = invalidate_inodes(sb, kill_dirty);
1677                drop_super(sb);
1678        }
1679        invalidate_bdev(bdev);
1680        return res;
1681}
1682EXPORT_SYMBOL(__invalidate_device);
1683
1684void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1685{
1686        struct inode *inode, *old_inode = NULL;
1687
1688        spin_lock(&inode_sb_list_lock);
1689        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1690                struct address_space *mapping = inode->i_mapping;
1691
1692                spin_lock(&inode->i_lock);
1693                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1694                    mapping->nrpages == 0) {
1695                        spin_unlock(&inode->i_lock);
1696                        continue;
1697                }
1698                __iget(inode);
1699                spin_unlock(&inode->i_lock);
1700                spin_unlock(&inode_sb_list_lock);
1701                /*
1702                 * We hold a reference to 'inode' so it couldn't have been
1703                 * removed from s_inodes list while we dropped the
1704                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1705                 * be holding the last reference and we cannot iput it under
1706                 * inode_sb_list_lock. So we keep the reference and iput it
1707                 * later.
1708                 */
1709                iput(old_inode);
1710                old_inode = inode;
1711
1712                func(I_BDEV(inode), arg);
1713
1714                spin_lock(&inode_sb_list_lock);
1715        }
1716        spin_unlock(&inode_sb_list_lock);
1717        iput(old_inode);
1718}
1719
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.