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/backing-dev.h>
  18#include <linux/module.h>
  19#include <linux/blkpg.h>
  20#include <linux/magic.h>
  21#include <linux/buffer_head.h>
  22#include <linux/swap.h>
  23#include <linux/pagevec.h>
  24#include <linux/writeback.h>
  25#include <linux/mpage.h>
  26#include <linux/mount.h>
  27#include <linux/uio.h>
  28#include <linux/namei.h>
  29#include <linux/log2.h>
  30#include <linux/cleancache.h>
  31#include <linux/dax.h>
  32#include <linux/badblocks.h>
  33#include <linux/falloc.h>
  34#include <asm/uaccess.h>
  35#include "internal.h"
  36
  37struct bdev_inode {
  38        struct block_device bdev;
  39        struct inode vfs_inode;
  40};
  41
  42static const struct address_space_operations def_blk_aops;
  43
  44static inline struct bdev_inode *BDEV_I(struct inode *inode)
  45{
  46        return container_of(inode, struct bdev_inode, vfs_inode);
  47}
  48
  49struct block_device *I_BDEV(struct inode *inode)
  50{
  51        return &BDEV_I(inode)->bdev;
  52}
  53EXPORT_SYMBOL(I_BDEV);
  54
  55void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
  56{
  57        struct va_format vaf;
  58        va_list args;
  59
  60        va_start(args, fmt);
  61        vaf.fmt = fmt;
  62        vaf.va = &args;
  63        printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
  64        va_end(args);
  65}
  66
  67static void bdev_write_inode(struct block_device *bdev)
  68{
  69        struct inode *inode = bdev->bd_inode;
  70        int ret;
  71
  72        spin_lock(&inode->i_lock);
  73        while (inode->i_state & I_DIRTY) {
  74                spin_unlock(&inode->i_lock);
  75                ret = write_inode_now(inode, true);
  76                if (ret) {
  77                        char name[BDEVNAME_SIZE];
  78                        pr_warn_ratelimited("VFS: Dirty inode writeback failed "
  79                                            "for block device %s (err=%d).\n",
  80                                            bdevname(bdev, name), ret);
  81                }
  82                spin_lock(&inode->i_lock);
  83        }
  84        spin_unlock(&inode->i_lock);
  85}
  86
  87/* Kill _all_ buffers and pagecache , dirty or not.. */
  88void kill_bdev(struct block_device *bdev)
  89{
  90        struct address_space *mapping = bdev->bd_inode->i_mapping;
  91
  92        if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
  93                return;
  94
  95        invalidate_bh_lrus();
  96        truncate_inode_pages(mapping, 0);
  97}       
  98EXPORT_SYMBOL(kill_bdev);
  99
 100/* Invalidate clean unused buffers and pagecache. */
 101void invalidate_bdev(struct block_device *bdev)
 102{
 103        struct address_space *mapping = bdev->bd_inode->i_mapping;
 104
 105        if (mapping->nrpages == 0)
 106                return;
 107
 108        invalidate_bh_lrus();
 109        lru_add_drain_all();    /* make sure all lru add caches are flushed */
 110        invalidate_mapping_pages(mapping, 0, -1);
 111        /* 99% of the time, we don't need to flush the cleancache on the bdev.
 112         * But, for the strange corners, lets be cautious
 113         */
 114        cleancache_invalidate_inode(mapping);
 115}
 116EXPORT_SYMBOL(invalidate_bdev);
 117
 118int set_blocksize(struct block_device *bdev, int size)
 119{
 120        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 121        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 122                return -EINVAL;
 123
 124        /* Size cannot be smaller than the size supported by the device */
 125        if (size < bdev_logical_block_size(bdev))
 126                return -EINVAL;
 127
 128        /* Don't change the size if it is same as current */
 129        if (bdev->bd_block_size != size) {
 130                sync_blockdev(bdev);
 131                bdev->bd_block_size = size;
 132                bdev->bd_inode->i_blkbits = blksize_bits(size);
 133                kill_bdev(bdev);
 134        }
 135        return 0;
 136}
 137
 138EXPORT_SYMBOL(set_blocksize);
 139
 140int sb_set_blocksize(struct super_block *sb, int size)
 141{
 142        if (set_blocksize(sb->s_bdev, size))
 143                return 0;
 144        /* If we get here, we know size is power of two
 145         * and it's value is between 512 and PAGE_SIZE */
 146        sb->s_blocksize = size;
 147        sb->s_blocksize_bits = blksize_bits(size);
 148        return sb->s_blocksize;
 149}
 150
 151EXPORT_SYMBOL(sb_set_blocksize);
 152
 153int sb_min_blocksize(struct super_block *sb, int size)
 154{
 155        int minsize = bdev_logical_block_size(sb->s_bdev);
 156        if (size < minsize)
 157                size = minsize;
 158        return sb_set_blocksize(sb, size);
 159}
 160
 161EXPORT_SYMBOL(sb_min_blocksize);
 162
 163static int
 164blkdev_get_block(struct inode *inode, sector_t iblock,
 165                struct buffer_head *bh, int create)
 166{
 167        bh->b_bdev = I_BDEV(inode);
 168        bh->b_blocknr = iblock;
 169        set_buffer_mapped(bh);
 170        return 0;
 171}
 172
 173static struct inode *bdev_file_inode(struct file *file)
 174{
 175        return file->f_mapping->host;
 176}
 177
 178static ssize_t
 179blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 180{
 181        struct file *file = iocb->ki_filp;
 182        struct inode *inode = bdev_file_inode(file);
 183
 184        return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter,
 185                                    blkdev_get_block, NULL, NULL,
 186                                    DIO_SKIP_DIO_COUNT);
 187}
 188
 189int __sync_blockdev(struct block_device *bdev, int wait)
 190{
 191        if (!bdev)
 192                return 0;
 193        if (!wait)
 194                return filemap_flush(bdev->bd_inode->i_mapping);
 195        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 196}
 197
 198/*
 199 * Write out and wait upon all the dirty data associated with a block
 200 * device via its mapping.  Does not take the superblock lock.
 201 */
 202int sync_blockdev(struct block_device *bdev)
 203{
 204        return __sync_blockdev(bdev, 1);
 205}
 206EXPORT_SYMBOL(sync_blockdev);
 207
 208/*
 209 * Write out and wait upon all dirty data associated with this
 210 * device.   Filesystem data as well as the underlying block
 211 * device.  Takes the superblock lock.
 212 */
 213int fsync_bdev(struct block_device *bdev)
 214{
 215        struct super_block *sb = get_super(bdev);
 216        if (sb) {
 217                int res = sync_filesystem(sb);
 218                drop_super(sb);
 219                return res;
 220        }
 221        return sync_blockdev(bdev);
 222}
 223EXPORT_SYMBOL(fsync_bdev);
 224
 225/**
 226 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 227 * @bdev:       blockdevice to lock
 228 *
 229 * If a superblock is found on this device, we take the s_umount semaphore
 230 * on it to make sure nobody unmounts until the snapshot creation is done.
 231 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 232 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 233 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 234 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 235 * actually.
 236 */
 237struct super_block *freeze_bdev(struct block_device *bdev)
 238{
 239        struct super_block *sb;
 240        int error = 0;
 241
 242        mutex_lock(&bdev->bd_fsfreeze_mutex);
 243        if (++bdev->bd_fsfreeze_count > 1) {
 244                /*
 245                 * We don't even need to grab a reference - the first call
 246                 * to freeze_bdev grab an active reference and only the last
 247                 * thaw_bdev drops it.
 248                 */
 249                sb = get_super(bdev);
 250                if (sb)
 251                        drop_super(sb);
 252                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 253                return sb;
 254        }
 255
 256        sb = get_active_super(bdev);
 257        if (!sb)
 258                goto out;
 259        if (sb->s_op->freeze_super)
 260                error = sb->s_op->freeze_super(sb);
 261        else
 262                error = freeze_super(sb);
 263        if (error) {
 264                deactivate_super(sb);
 265                bdev->bd_fsfreeze_count--;
 266                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 267                return ERR_PTR(error);
 268        }
 269        deactivate_super(sb);
 270 out:
 271        sync_blockdev(bdev);
 272        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 273        return sb;      /* thaw_bdev releases s->s_umount */
 274}
 275EXPORT_SYMBOL(freeze_bdev);
 276
 277/**
 278 * thaw_bdev  -- unlock filesystem
 279 * @bdev:       blockdevice to unlock
 280 * @sb:         associated superblock
 281 *
 282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 283 */
 284int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 285{
 286        int error = -EINVAL;
 287
 288        mutex_lock(&bdev->bd_fsfreeze_mutex);
 289        if (!bdev->bd_fsfreeze_count)
 290                goto out;
 291
 292        error = 0;
 293        if (--bdev->bd_fsfreeze_count > 0)
 294                goto out;
 295
 296        if (!sb)
 297                goto out;
 298
 299        if (sb->s_op->thaw_super)
 300                error = sb->s_op->thaw_super(sb);
 301        else
 302                error = thaw_super(sb);
 303        if (error)
 304                bdev->bd_fsfreeze_count++;
 305out:
 306        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 307        return error;
 308}
 309EXPORT_SYMBOL(thaw_bdev);
 310
 311static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 312{
 313        return block_write_full_page(page, blkdev_get_block, wbc);
 314}
 315
 316static int blkdev_readpage(struct file * file, struct page * page)
 317{
 318        return block_read_full_page(page, blkdev_get_block);
 319}
 320
 321static int blkdev_readpages(struct file *file, struct address_space *mapping,
 322                        struct list_head *pages, unsigned nr_pages)
 323{
 324        return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
 325}
 326
 327static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 328                        loff_t pos, unsigned len, unsigned flags,
 329                        struct page **pagep, void **fsdata)
 330{
 331        return block_write_begin(mapping, pos, len, flags, pagep,
 332                                 blkdev_get_block);
 333}
 334
 335static int blkdev_write_end(struct file *file, struct address_space *mapping,
 336                        loff_t pos, unsigned len, unsigned copied,
 337                        struct page *page, void *fsdata)
 338{
 339        int ret;
 340        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 341
 342        unlock_page(page);
 343        put_page(page);
 344
 345        return ret;
 346}
 347
 348/*
 349 * private llseek:
 350 * for a block special file file_inode(file)->i_size is zero
 351 * so we compute the size by hand (just as in block_read/write above)
 352 */
 353static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 354{
 355        struct inode *bd_inode = bdev_file_inode(file);
 356        loff_t retval;
 357
 358        inode_lock(bd_inode);
 359        retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
 360        inode_unlock(bd_inode);
 361        return retval;
 362}
 363        
 364int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 365{
 366        struct inode *bd_inode = bdev_file_inode(filp);
 367        struct block_device *bdev = I_BDEV(bd_inode);
 368        int error;
 369        
 370        error = filemap_write_and_wait_range(filp->f_mapping, start, end);
 371        if (error)
 372                return error;
 373
 374        /*
 375         * There is no need to serialise calls to blkdev_issue_flush with
 376         * i_mutex and doing so causes performance issues with concurrent
 377         * O_SYNC writers to a block device.
 378         */
 379        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 380        if (error == -EOPNOTSUPP)
 381                error = 0;
 382
 383        return error;
 384}
 385EXPORT_SYMBOL(blkdev_fsync);
 386
 387/**
 388 * bdev_read_page() - Start reading a page from a block device
 389 * @bdev: The device to read the page from
 390 * @sector: The offset on the device to read the page to (need not be aligned)
 391 * @page: The page to read
 392 *
 393 * On entry, the page should be locked.  It will be unlocked when the page
 394 * has been read.  If the block driver implements rw_page synchronously,
 395 * that will be true on exit from this function, but it need not be.
 396 *
 397 * Errors returned by this function are usually "soft", eg out of memory, or
 398 * queue full; callers should try a different route to read this page rather
 399 * than propagate an error back up the stack.
 400 *
 401 * Return: negative errno if an error occurs, 0 if submission was successful.
 402 */
 403int bdev_read_page(struct block_device *bdev, sector_t sector,
 404                        struct page *page)
 405{
 406        const struct block_device_operations *ops = bdev->bd_disk->fops;
 407        int result = -EOPNOTSUPP;
 408
 409        if (!ops->rw_page || bdev_get_integrity(bdev))
 410                return result;
 411
 412        result = blk_queue_enter(bdev->bd_queue, false);
 413        if (result)
 414                return result;
 415        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
 416        blk_queue_exit(bdev->bd_queue);
 417        return result;
 418}
 419EXPORT_SYMBOL_GPL(bdev_read_page);
 420
 421/**
 422 * bdev_write_page() - Start writing a page to a block device
 423 * @bdev: The device to write the page to
 424 * @sector: The offset on the device to write the page to (need not be aligned)
 425 * @page: The page to write
 426 * @wbc: The writeback_control for the write
 427 *
 428 * On entry, the page should be locked and not currently under writeback.
 429 * On exit, if the write started successfully, the page will be unlocked and
 430 * under writeback.  If the write failed already (eg the driver failed to
 431 * queue the page to the device), the page will still be locked.  If the
 432 * caller is a ->writepage implementation, it will need to unlock the page.
 433 *
 434 * Errors returned by this function are usually "soft", eg out of memory, or
 435 * queue full; callers should try a different route to write this page rather
 436 * than propagate an error back up the stack.
 437 *
 438 * Return: negative errno if an error occurs, 0 if submission was successful.
 439 */
 440int bdev_write_page(struct block_device *bdev, sector_t sector,
 441                        struct page *page, struct writeback_control *wbc)
 442{
 443        int result;
 444        const struct block_device_operations *ops = bdev->bd_disk->fops;
 445
 446        if (!ops->rw_page || bdev_get_integrity(bdev))
 447                return -EOPNOTSUPP;
 448        result = blk_queue_enter(bdev->bd_queue, false);
 449        if (result)
 450                return result;
 451
 452        set_page_writeback(page);
 453        result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
 454        if (result)
 455                end_page_writeback(page);
 456        else
 457                unlock_page(page);
 458        blk_queue_exit(bdev->bd_queue);
 459        return result;
 460}
 461EXPORT_SYMBOL_GPL(bdev_write_page);
 462
 463/**
 464 * bdev_direct_access() - Get the address for directly-accessibly memory
 465 * @bdev: The device containing the memory
 466 * @dax: control and output parameters for ->direct_access
 467 *
 468 * If a block device is made up of directly addressable memory, this function
 469 * will tell the caller the PFN and the address of the memory.  The address
 470 * may be directly dereferenced within the kernel without the need to call
 471 * ioremap(), kmap() or similar.  The PFN is suitable for inserting into
 472 * page tables.
 473 *
 474 * Return: negative errno if an error occurs, otherwise the number of bytes
 475 * accessible at this address.
 476 */
 477long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
 478{
 479        sector_t sector = dax->sector;
 480        long avail, size = dax->size;
 481        const struct block_device_operations *ops = bdev->bd_disk->fops;
 482
 483        /*
 484         * The device driver is allowed to sleep, in order to make the
 485         * memory directly accessible.
 486         */
 487        might_sleep();
 488
 489        if (size < 0)
 490                return size;
 491        if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
 492                return -EOPNOTSUPP;
 493        if ((sector + DIV_ROUND_UP(size, 512)) >
 494                                        part_nr_sects_read(bdev->bd_part))
 495                return -ERANGE;
 496        sector += get_start_sect(bdev);
 497        if (sector % (PAGE_SIZE / 512))
 498                return -EINVAL;
 499        avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
 500        if (!avail)
 501                return -ERANGE;
 502        if (avail > 0 && avail & ~PAGE_MASK)
 503                return -ENXIO;
 504        return min(avail, size);
 505}
 506EXPORT_SYMBOL_GPL(bdev_direct_access);
 507
 508/**
 509 * bdev_dax_supported() - Check if the device supports dax for filesystem
 510 * @sb: The superblock of the device
 511 * @blocksize: The block size of the device
 512 *
 513 * This is a library function for filesystems to check if the block device
 514 * can be mounted with dax option.
 515 *
 516 * Return: negative errno if unsupported, 0 if supported.
 517 */
 518int bdev_dax_supported(struct super_block *sb, int blocksize)
 519{
 520        struct blk_dax_ctl dax = {
 521                .sector = 0,
 522                .size = PAGE_SIZE,
 523        };
 524        int err;
 525
 526        if (blocksize != PAGE_SIZE) {
 527                vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
 528                return -EINVAL;
 529        }
 530
 531        err = bdev_direct_access(sb->s_bdev, &dax);
 532        if (err < 0) {
 533                switch (err) {
 534                case -EOPNOTSUPP:
 535                        vfs_msg(sb, KERN_ERR,
 536                                "error: device does not support dax");
 537                        break;
 538                case -EINVAL:
 539                        vfs_msg(sb, KERN_ERR,
 540                                "error: unaligned partition for dax");
 541                        break;
 542                default:
 543                        vfs_msg(sb, KERN_ERR,
 544                                "error: dax access failed (%d)", err);
 545                }
 546                return err;
 547        }
 548
 549        return 0;
 550}
 551EXPORT_SYMBOL_GPL(bdev_dax_supported);
 552
 553/**
 554 * bdev_dax_capable() - Return if the raw device is capable for dax
 555 * @bdev: The device for raw block device access
 556 */
 557bool bdev_dax_capable(struct block_device *bdev)
 558{
 559        struct blk_dax_ctl dax = {
 560                .size = PAGE_SIZE,
 561        };
 562
 563        if (!IS_ENABLED(CONFIG_FS_DAX))
 564                return false;
 565
 566        dax.sector = 0;
 567        if (bdev_direct_access(bdev, &dax) < 0)
 568                return false;
 569
 570        dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
 571        if (bdev_direct_access(bdev, &dax) < 0)
 572                return false;
 573
 574        return true;
 575}
 576
 577/*
 578 * pseudo-fs
 579 */
 580
 581static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 582static struct kmem_cache * bdev_cachep __read_mostly;
 583
 584static struct inode *bdev_alloc_inode(struct super_block *sb)
 585{
 586        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 587        if (!ei)
 588                return NULL;
 589        return &ei->vfs_inode;
 590}
 591
 592static void bdev_i_callback(struct rcu_head *head)
 593{
 594        struct inode *inode = container_of(head, struct inode, i_rcu);
 595        struct bdev_inode *bdi = BDEV_I(inode);
 596
 597        kmem_cache_free(bdev_cachep, bdi);
 598}
 599
 600static void bdev_destroy_inode(struct inode *inode)
 601{
 602        call_rcu(&inode->i_rcu, bdev_i_callback);
 603}
 604
 605static void init_once(void *foo)
 606{
 607        struct bdev_inode *ei = (struct bdev_inode *) foo;
 608        struct block_device *bdev = &ei->bdev;
 609
 610        memset(bdev, 0, sizeof(*bdev));
 611        mutex_init(&bdev->bd_mutex);
 612        INIT_LIST_HEAD(&bdev->bd_list);
 613#ifdef CONFIG_SYSFS
 614        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 615#endif
 616        inode_init_once(&ei->vfs_inode);
 617        /* Initialize mutex for freeze. */
 618        mutex_init(&bdev->bd_fsfreeze_mutex);
 619}
 620
 621static void bdev_evict_inode(struct inode *inode)
 622{
 623        struct block_device *bdev = &BDEV_I(inode)->bdev;
 624        truncate_inode_pages_final(&inode->i_data);
 625        invalidate_inode_buffers(inode); /* is it needed here? */
 626        clear_inode(inode);
 627        spin_lock(&bdev_lock);
 628        list_del_init(&bdev->bd_list);
 629        spin_unlock(&bdev_lock);
 630}
 631
 632static const struct super_operations bdev_sops = {
 633        .statfs = simple_statfs,
 634        .alloc_inode = bdev_alloc_inode,
 635        .destroy_inode = bdev_destroy_inode,
 636        .drop_inode = generic_delete_inode,
 637        .evict_inode = bdev_evict_inode,
 638};
 639
 640static struct dentry *bd_mount(struct file_system_type *fs_type,
 641        int flags, const char *dev_name, void *data)
 642{
 643        struct dentry *dent;
 644        dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
 645        if (!IS_ERR(dent))
 646                dent->d_sb->s_iflags |= SB_I_CGROUPWB;
 647        return dent;
 648}
 649
 650static struct file_system_type bd_type = {
 651        .name           = "bdev",
 652        .mount          = bd_mount,
 653        .kill_sb        = kill_anon_super,
 654};
 655
 656struct super_block *blockdev_superblock __read_mostly;
 657EXPORT_SYMBOL_GPL(blockdev_superblock);
 658
 659void __init bdev_cache_init(void)
 660{
 661        int err;
 662        static struct vfsmount *bd_mnt;
 663
 664        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 665                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 666                                SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
 667                        init_once);
 668        err = register_filesystem(&bd_type);
 669        if (err)
 670                panic("Cannot register bdev pseudo-fs");
 671        bd_mnt = kern_mount(&bd_type);
 672        if (IS_ERR(bd_mnt))
 673                panic("Cannot create bdev pseudo-fs");
 674        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 675}
 676
 677/*
 678 * Most likely _very_ bad one - but then it's hardly critical for small
 679 * /dev and can be fixed when somebody will need really large one.
 680 * Keep in mind that it will be fed through icache hash function too.
 681 */
 682static inline unsigned long hash(dev_t dev)
 683{
 684        return MAJOR(dev)+MINOR(dev);
 685}
 686
 687static int bdev_test(struct inode *inode, void *data)
 688{
 689        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 690}
 691
 692static int bdev_set(struct inode *inode, void *data)
 693{
 694        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 695        return 0;
 696}
 697
 698static LIST_HEAD(all_bdevs);
 699
 700struct block_device *bdget(dev_t dev)
 701{
 702        struct block_device *bdev;
 703        struct inode *inode;
 704
 705        inode = iget5_locked(blockdev_superblock, hash(dev),
 706                        bdev_test, bdev_set, &dev);
 707
 708        if (!inode)
 709                return NULL;
 710
 711        bdev = &BDEV_I(inode)->bdev;
 712
 713        if (inode->i_state & I_NEW) {
 714                bdev->bd_contains = NULL;
 715                bdev->bd_super = NULL;
 716                bdev->bd_inode = inode;
 717                bdev->bd_block_size = (1 << inode->i_blkbits);
 718                bdev->bd_part_count = 0;
 719                bdev->bd_invalidated = 0;
 720                inode->i_mode = S_IFBLK;
 721                inode->i_rdev = dev;
 722                inode->i_bdev = bdev;
 723                inode->i_data.a_ops = &def_blk_aops;
 724                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 725                spin_lock(&bdev_lock);
 726                list_add(&bdev->bd_list, &all_bdevs);
 727                spin_unlock(&bdev_lock);
 728                unlock_new_inode(inode);
 729        }
 730        return bdev;
 731}
 732
 733EXPORT_SYMBOL(bdget);
 734
 735/**
 736 * bdgrab -- Grab a reference to an already referenced block device
 737 * @bdev:       Block device to grab a reference to.
 738 */
 739struct block_device *bdgrab(struct block_device *bdev)
 740{
 741        ihold(bdev->bd_inode);
 742        return bdev;
 743}
 744EXPORT_SYMBOL(bdgrab);
 745
 746long nr_blockdev_pages(void)
 747{
 748        struct block_device *bdev;
 749        long ret = 0;
 750        spin_lock(&bdev_lock);
 751        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 752                ret += bdev->bd_inode->i_mapping->nrpages;
 753        }
 754        spin_unlock(&bdev_lock);
 755        return ret;
 756}
 757
 758void bdput(struct block_device *bdev)
 759{
 760        iput(bdev->bd_inode);
 761}
 762
 763EXPORT_SYMBOL(bdput);
 764 
 765static struct block_device *bd_acquire(struct inode *inode)
 766{
 767        struct block_device *bdev;
 768
 769        spin_lock(&bdev_lock);
 770        bdev = inode->i_bdev;
 771        if (bdev) {
 772                bdgrab(bdev);
 773                spin_unlock(&bdev_lock);
 774                return bdev;
 775        }
 776        spin_unlock(&bdev_lock);
 777
 778        bdev = bdget(inode->i_rdev);
 779        if (bdev) {
 780                spin_lock(&bdev_lock);
 781                if (!inode->i_bdev) {
 782                        /*
 783                         * We take an additional reference to bd_inode,
 784                         * and it's released in clear_inode() of inode.
 785                         * So, we can access it via ->i_mapping always
 786                         * without igrab().
 787                         */
 788                        bdgrab(bdev);
 789                        inode->i_bdev = bdev;
 790                        inode->i_mapping = bdev->bd_inode->i_mapping;
 791                }
 792                spin_unlock(&bdev_lock);
 793        }
 794        return bdev;
 795}
 796
 797/* Call when you free inode */
 798
 799void bd_forget(struct inode *inode)
 800{
 801        struct block_device *bdev = NULL;
 802
 803        spin_lock(&bdev_lock);
 804        if (!sb_is_blkdev_sb(inode->i_sb))
 805                bdev = inode->i_bdev;
 806        inode->i_bdev = NULL;
 807        inode->i_mapping = &inode->i_data;
 808        spin_unlock(&bdev_lock);
 809
 810        if (bdev)
 811                bdput(bdev);
 812}
 813
 814/**
 815 * bd_may_claim - test whether a block device can be claimed
 816 * @bdev: block device of interest
 817 * @whole: whole block device containing @bdev, may equal @bdev
 818 * @holder: holder trying to claim @bdev
 819 *
 820 * Test whether @bdev can be claimed by @holder.
 821 *
 822 * CONTEXT:
 823 * spin_lock(&bdev_lock).
 824 *
 825 * RETURNS:
 826 * %true if @bdev can be claimed, %false otherwise.
 827 */
 828static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 829                         void *holder)
 830{
 831        if (bdev->bd_holder == holder)
 832                return true;     /* already a holder */
 833        else if (bdev->bd_holder != NULL)
 834                return false;    /* held by someone else */
 835        else if (bdev->bd_contains == bdev)
 836                return true;     /* is a whole device which isn't held */
 837
 838        else if (whole->bd_holder == bd_may_claim)
 839                return true;     /* is a partition of a device that is being partitioned */
 840        else if (whole->bd_holder != NULL)
 841                return false;    /* is a partition of a held device */
 842        else
 843                return true;     /* is a partition of an un-held device */
 844}
 845
 846/**
 847 * bd_prepare_to_claim - prepare to claim a block device
 848 * @bdev: block device of interest
 849 * @whole: the whole device containing @bdev, may equal @bdev
 850 * @holder: holder trying to claim @bdev
 851 *
 852 * Prepare to claim @bdev.  This function fails if @bdev is already
 853 * claimed by another holder and waits if another claiming is in
 854 * progress.  This function doesn't actually claim.  On successful
 855 * return, the caller has ownership of bd_claiming and bd_holder[s].
 856 *
 857 * CONTEXT:
 858 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 859 * it multiple times.
 860 *
 861 * RETURNS:
 862 * 0 if @bdev can be claimed, -EBUSY otherwise.
 863 */
 864static int bd_prepare_to_claim(struct block_device *bdev,
 865                               struct block_device *whole, void *holder)
 866{
 867retry:
 868        /* if someone else claimed, fail */
 869        if (!bd_may_claim(bdev, whole, holder))
 870                return -EBUSY;
 871
 872        /* if claiming is already in progress, wait for it to finish */
 873        if (whole->bd_claiming) {
 874                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 875                DEFINE_WAIT(wait);
 876
 877                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 878                spin_unlock(&bdev_lock);
 879                schedule();
 880                finish_wait(wq, &wait);
 881                spin_lock(&bdev_lock);
 882                goto retry;
 883        }
 884
 885        /* yay, all mine */
 886        return 0;
 887}
 888
 889/**
 890 * bd_start_claiming - start claiming a block device
 891 * @bdev: block device of interest
 892 * @holder: holder trying to claim @bdev
 893 *
 894 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 895 * exclusively and mark that an exclusive open is in progress.  Each
 896 * successful call to this function must be matched with a call to
 897 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 898 * fail).
 899 *
 900 * This function is used to gain exclusive access to the block device
 901 * without actually causing other exclusive open attempts to fail. It
 902 * should be used when the open sequence itself requires exclusive
 903 * access but may subsequently fail.
 904 *
 905 * CONTEXT:
 906 * Might sleep.
 907 *
 908 * RETURNS:
 909 * Pointer to the block device containing @bdev on success, ERR_PTR()
 910 * value on failure.
 911 */
 912static struct block_device *bd_start_claiming(struct block_device *bdev,
 913                                              void *holder)
 914{
 915        struct gendisk *disk;
 916        struct block_device *whole;
 917        int partno, err;
 918
 919        might_sleep();
 920
 921        /*
 922         * @bdev might not have been initialized properly yet, look up
 923         * and grab the outer block device the hard way.
 924         */
 925        disk = get_gendisk(bdev->bd_dev, &partno);
 926        if (!disk)
 927                return ERR_PTR(-ENXIO);
 928
 929        /*
 930         * Normally, @bdev should equal what's returned from bdget_disk()
 931         * if partno is 0; however, some drivers (floppy) use multiple
 932         * bdev's for the same physical device and @bdev may be one of the
 933         * aliases.  Keep @bdev if partno is 0.  This means claimer
 934         * tracking is broken for those devices but it has always been that
 935         * way.
 936         */
 937        if (partno)
 938                whole = bdget_disk(disk, 0);
 939        else
 940                whole = bdgrab(bdev);
 941
 942        module_put(disk->fops->owner);
 943        put_disk(disk);
 944        if (!whole)
 945                return ERR_PTR(-ENOMEM);
 946
 947        /* prepare to claim, if successful, mark claiming in progress */
 948        spin_lock(&bdev_lock);
 949
 950        err = bd_prepare_to_claim(bdev, whole, holder);
 951        if (err == 0) {
 952                whole->bd_claiming = holder;
 953                spin_unlock(&bdev_lock);
 954                return whole;
 955        } else {
 956                spin_unlock(&bdev_lock);
 957                bdput(whole);
 958                return ERR_PTR(err);
 959        }
 960}
 961
 962#ifdef CONFIG_SYSFS
 963struct bd_holder_disk {
 964        struct list_head        list;
 965        struct gendisk          *disk;
 966        int                     refcnt;
 967};
 968
 969static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 970                                                  struct gendisk *disk)
 971{
 972        struct bd_holder_disk *holder;
 973
 974        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 975                if (holder->disk == disk)
 976                        return holder;
 977        return NULL;
 978}
 979
 980static int add_symlink(struct kobject *from, struct kobject *to)
 981{
 982        return sysfs_create_link(from, to, kobject_name(to));
 983}
 984
 985static void del_symlink(struct kobject *from, struct kobject *to)
 986{
 987        sysfs_remove_link(from, kobject_name(to));
 988}
 989
 990/**
 991 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 992 * @bdev: the claimed slave bdev
 993 * @disk: the holding disk
 994 *
 995 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 996 *
 997 * This functions creates the following sysfs symlinks.
 998 *
 999 * - from "slaves" directory of the holder @disk to the claimed @bdev
1000 * - from "holders" directory of the @bdev to the holder @disk
1001 *
1002 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1003 * passed to bd_link_disk_holder(), then:
1004 *
1005 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1006 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1007 *
1008 * The caller must have claimed @bdev before calling this function and
1009 * ensure that both @bdev and @disk are valid during the creation and
1010 * lifetime of these symlinks.
1011 *
1012 * CONTEXT:
1013 * Might sleep.
1014 *
1015 * RETURNS:
1016 * 0 on success, -errno on failure.
1017 */
1018int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1019{
1020        struct bd_holder_disk *holder;
1021        int ret = 0;
1022
1023        mutex_lock(&bdev->bd_mutex);
1024
1025        WARN_ON_ONCE(!bdev->bd_holder);
1026
1027        /* FIXME: remove the following once add_disk() handles errors */
1028        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1029                goto out_unlock;
1030
1031        holder = bd_find_holder_disk(bdev, disk);
1032        if (holder) {
1033                holder->refcnt++;
1034                goto out_unlock;
1035        }
1036
1037        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1038        if (!holder) {
1039                ret = -ENOMEM;
1040                goto out_unlock;
1041        }
1042
1043        INIT_LIST_HEAD(&holder->list);
1044        holder->disk = disk;
1045        holder->refcnt = 1;
1046
1047        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1048        if (ret)
1049                goto out_free;
1050
1051        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1052        if (ret)
1053                goto out_del;
1054        /*
1055         * bdev could be deleted beneath us which would implicitly destroy
1056         * the holder directory.  Hold on to it.
1057         */
1058        kobject_get(bdev->bd_part->holder_dir);
1059
1060        list_add(&holder->list, &bdev->bd_holder_disks);
1061        goto out_unlock;
1062
1063out_del:
1064        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1065out_free:
1066        kfree(holder);
1067out_unlock:
1068        mutex_unlock(&bdev->bd_mutex);
1069        return ret;
1070}
1071EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1072
1073/**
1074 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1075 * @bdev: the calimed slave bdev
1076 * @disk: the holding disk
1077 *
1078 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1079 *
1080 * CONTEXT:
1081 * Might sleep.
1082 */
1083void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1084{
1085        struct bd_holder_disk *holder;
1086
1087        mutex_lock(&bdev->bd_mutex);
1088
1089        holder = bd_find_holder_disk(bdev, disk);
1090
1091        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1092                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1093                del_symlink(bdev->bd_part->holder_dir,
1094                            &disk_to_dev(disk)->kobj);
1095                kobject_put(bdev->bd_part->holder_dir);
1096                list_del_init(&holder->list);
1097                kfree(holder);
1098        }
1099
1100        mutex_unlock(&bdev->bd_mutex);
1101}
1102EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1103#endif
1104
1105/**
1106 * flush_disk - invalidates all buffer-cache entries on a disk
1107 *
1108 * @bdev:      struct block device to be flushed
1109 * @kill_dirty: flag to guide handling of dirty inodes
1110 *
1111 * Invalidates all buffer-cache entries on a disk. It should be called
1112 * when a disk has been changed -- either by a media change or online
1113 * resize.
1114 */
1115static void flush_disk(struct block_device *bdev, bool kill_dirty)
1116{
1117        if (__invalidate_device(bdev, kill_dirty)) {
1118                printk(KERN_WARNING "VFS: busy inodes on changed media or "
1119                       "resized disk %s\n",
1120                       bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1121        }
1122
1123        if (!bdev->bd_disk)
1124                return;
1125        if (disk_part_scan_enabled(bdev->bd_disk))
1126                bdev->bd_invalidated = 1;
1127}
1128
1129/**
1130 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1131 * @disk: struct gendisk to check
1132 * @bdev: struct bdev to adjust.
1133 *
1134 * This routine checks to see if the bdev size does not match the disk size
1135 * and adjusts it if it differs.
1136 */
1137void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1138{
1139        loff_t disk_size, bdev_size;
1140
1141        disk_size = (loff_t)get_capacity(disk) << 9;
1142        bdev_size = i_size_read(bdev->bd_inode);
1143        if (disk_size != bdev_size) {
1144                printk(KERN_INFO
1145                       "%s: detected capacity change from %lld to %lld\n",
1146                       disk->disk_name, bdev_size, disk_size);
1147                i_size_write(bdev->bd_inode, disk_size);
1148                flush_disk(bdev, false);
1149        }
1150}
1151EXPORT_SYMBOL(check_disk_size_change);
1152
1153/**
1154 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1155 * @disk: struct gendisk to be revalidated
1156 *
1157 * This routine is a wrapper for lower-level driver's revalidate_disk
1158 * call-backs.  It is used to do common pre and post operations needed
1159 * for all revalidate_disk operations.
1160 */
1161int revalidate_disk(struct gendisk *disk)
1162{
1163        struct block_device *bdev;
1164        int ret = 0;
1165
1166        if (disk->fops->revalidate_disk)
1167                ret = disk->fops->revalidate_disk(disk);
1168        blk_integrity_revalidate(disk);
1169        bdev = bdget_disk(disk, 0);
1170        if (!bdev)
1171                return ret;
1172
1173        mutex_lock(&bdev->bd_mutex);
1174        check_disk_size_change(disk, bdev);
1175        bdev->bd_invalidated = 0;
1176        mutex_unlock(&bdev->bd_mutex);
1177        bdput(bdev);
1178        return ret;
1179}
1180EXPORT_SYMBOL(revalidate_disk);
1181
1182/*
1183 * This routine checks whether a removable media has been changed,
1184 * and invalidates all buffer-cache-entries in that case. This
1185 * is a relatively slow routine, so we have to try to minimize using
1186 * it. Thus it is called only upon a 'mount' or 'open'. This
1187 * is the best way of combining speed and utility, I think.
1188 * People changing diskettes in the middle of an operation deserve
1189 * to lose :-)
1190 */
1191int check_disk_change(struct block_device *bdev)
1192{
1193        struct gendisk *disk = bdev->bd_disk;
1194        const struct block_device_operations *bdops = disk->fops;
1195        unsigned int events;
1196
1197        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1198                                   DISK_EVENT_EJECT_REQUEST);
1199        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1200                return 0;
1201
1202        flush_disk(bdev, true);
1203        if (bdops->revalidate_disk)
1204                bdops->revalidate_disk(bdev->bd_disk);
1205        return 1;
1206}
1207
1208EXPORT_SYMBOL(check_disk_change);
1209
1210void bd_set_size(struct block_device *bdev, loff_t size)
1211{
1212        unsigned bsize = bdev_logical_block_size(bdev);
1213
1214        inode_lock(bdev->bd_inode);
1215        i_size_write(bdev->bd_inode, size);
1216        inode_unlock(bdev->bd_inode);
1217        while (bsize < PAGE_SIZE) {
1218                if (size & bsize)
1219                        break;
1220                bsize <<= 1;
1221        }
1222        bdev->bd_block_size = bsize;
1223        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1224}
1225EXPORT_SYMBOL(bd_set_size);
1226
1227static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1228
1229/*
1230 * bd_mutex locking:
1231 *
1232 *  mutex_lock(part->bd_mutex)
1233 *    mutex_lock_nested(whole->bd_mutex, 1)
1234 */
1235
1236static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1237{
1238        struct gendisk *disk;
1239        struct module *owner;
1240        int ret;
1241        int partno;
1242        int perm = 0;
1243
1244        if (mode & FMODE_READ)
1245                perm |= MAY_READ;
1246        if (mode & FMODE_WRITE)
1247                perm |= MAY_WRITE;
1248        /*
1249         * hooks: /n/, see "layering violations".
1250         */
1251        if (!for_part) {
1252                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1253                if (ret != 0) {
1254                        bdput(bdev);
1255                        return ret;
1256                }
1257        }
1258
1259 restart:
1260
1261        ret = -ENXIO;
1262        disk = get_gendisk(bdev->bd_dev, &partno);
1263        if (!disk)
1264                goto out;
1265        owner = disk->fops->owner;
1266
1267        disk_block_events(disk);
1268        mutex_lock_nested(&bdev->bd_mutex, for_part);
1269        if (!bdev->bd_openers) {
1270                bdev->bd_disk = disk;
1271                bdev->bd_queue = disk->queue;
1272                bdev->bd_contains = bdev;
1273
1274                if (!partno) {
1275                        ret = -ENXIO;
1276                        bdev->bd_part = disk_get_part(disk, partno);
1277                        if (!bdev->bd_part)
1278                                goto out_clear;
1279
1280                        ret = 0;
1281                        if (disk->fops->open) {
1282                                ret = disk->fops->open(bdev, mode);
1283                                if (ret == -ERESTARTSYS) {
1284                                        /* Lost a race with 'disk' being
1285                                         * deleted, try again.
1286                                         * See md.c
1287                                         */
1288                                        disk_put_part(bdev->bd_part);
1289                                        bdev->bd_part = NULL;
1290                                        bdev->bd_disk = NULL;
1291                                        bdev->bd_queue = NULL;
1292                                        mutex_unlock(&bdev->bd_mutex);
1293                                        disk_unblock_events(disk);
1294                                        put_disk(disk);
1295                                        module_put(owner);
1296                                        goto restart;
1297                                }
1298                        }
1299
1300                        if (!ret)
1301                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1302
1303                        /*
1304                         * If the device is invalidated, rescan partition
1305                         * if open succeeded or failed with -ENOMEDIUM.
1306                         * The latter is necessary to prevent ghost
1307                         * partitions on a removed medium.
1308                         */
1309                        if (bdev->bd_invalidated) {
1310                                if (!ret)
1311                                        rescan_partitions(disk, bdev);
1312                                else if (ret == -ENOMEDIUM)
1313                                        invalidate_partitions(disk, bdev);
1314                        }
1315
1316                        if (ret)
1317                                goto out_clear;
1318                } else {
1319                        struct block_device *whole;
1320                        whole = bdget_disk(disk, 0);
1321                        ret = -ENOMEM;
1322                        if (!whole)
1323                                goto out_clear;
1324                        BUG_ON(for_part);
1325                        ret = __blkdev_get(whole, mode, 1);
1326                        if (ret)
1327                                goto out_clear;
1328                        bdev->bd_contains = whole;
1329                        bdev->bd_part = disk_get_part(disk, partno);
1330                        if (!(disk->flags & GENHD_FL_UP) ||
1331                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1332                                ret = -ENXIO;
1333                                goto out_clear;
1334                        }
1335                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1336                }
1337        } else {
1338                if (bdev->bd_contains == bdev) {
1339                        ret = 0;
1340                        if (bdev->bd_disk->fops->open)
1341                                ret = bdev->bd_disk->fops->open(bdev, mode);
1342                        /* the same as first opener case, read comment there */
1343                        if (bdev->bd_invalidated) {
1344                                if (!ret)
1345                                        rescan_partitions(bdev->bd_disk, bdev);
1346                                else if (ret == -ENOMEDIUM)
1347                                        invalidate_partitions(bdev->bd_disk, bdev);
1348                        }
1349                        if (ret)
1350                                goto out_unlock_bdev;
1351                }
1352                /* only one opener holds refs to the module and disk */
1353                put_disk(disk);
1354                module_put(owner);
1355        }
1356        bdev->bd_openers++;
1357        if (for_part)
1358                bdev->bd_part_count++;
1359        mutex_unlock(&bdev->bd_mutex);
1360        disk_unblock_events(disk);
1361        return 0;
1362
1363 out_clear:
1364        disk_put_part(bdev->bd_part);
1365        bdev->bd_disk = NULL;
1366        bdev->bd_part = NULL;
1367        bdev->bd_queue = NULL;
1368        if (bdev != bdev->bd_contains)
1369                __blkdev_put(bdev->bd_contains, mode, 1);
1370        bdev->bd_contains = NULL;
1371 out_unlock_bdev:
1372        mutex_unlock(&bdev->bd_mutex);
1373        disk_unblock_events(disk);
1374        put_disk(disk);
1375        module_put(owner);
1376 out:
1377        bdput(bdev);
1378
1379        return ret;
1380}
1381
1382/**
1383 * blkdev_get - open a block device
1384 * @bdev: block_device to open
1385 * @mode: FMODE_* mask
1386 * @holder: exclusive holder identifier
1387 *
1388 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1389 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1390 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1391 *
1392 * On success, the reference count of @bdev is unchanged.  On failure,
1393 * @bdev is put.
1394 *
1395 * CONTEXT:
1396 * Might sleep.
1397 *
1398 * RETURNS:
1399 * 0 on success, -errno on failure.
1400 */
1401int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1402{
1403        struct block_device *whole = NULL;
1404        int res;
1405
1406        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1407
1408        if ((mode & FMODE_EXCL) && holder) {
1409                whole = bd_start_claiming(bdev, holder);
1410                if (IS_ERR(whole)) {
1411                        bdput(bdev);
1412                        return PTR_ERR(whole);
1413                }
1414        }
1415
1416        res = __blkdev_get(bdev, mode, 0);
1417
1418        if (whole) {
1419                struct gendisk *disk = whole->bd_disk;
1420
1421                /* finish claiming */
1422                mutex_lock(&bdev->bd_mutex);
1423                spin_lock(&bdev_lock);
1424
1425                if (!res) {
1426                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1427                        /*
1428                         * Note that for a whole device bd_holders
1429                         * will be incremented twice, and bd_holder
1430                         * will be set to bd_may_claim before being
1431                         * set to holder
1432                         */
1433                        whole->bd_holders++;
1434                        whole->bd_holder = bd_may_claim;
1435                        bdev->bd_holders++;
1436                        bdev->bd_holder = holder;
1437                }
1438
1439                /* tell others that we're done */
1440                BUG_ON(whole->bd_claiming != holder);
1441                whole->bd_claiming = NULL;
1442                wake_up_bit(&whole->bd_claiming, 0);
1443
1444                spin_unlock(&bdev_lock);
1445
1446                /*
1447                 * Block event polling for write claims if requested.  Any
1448                 * write holder makes the write_holder state stick until
1449                 * all are released.  This is good enough and tracking
1450                 * individual writeable reference is too fragile given the
1451                 * way @mode is used in blkdev_get/put().
1452                 */
1453                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1454                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1455                        bdev->bd_write_holder = true;
1456                        disk_block_events(disk);
1457                }
1458
1459                mutex_unlock(&bdev->bd_mutex);
1460                bdput(whole);
1461        }
1462
1463        return res;
1464}
1465EXPORT_SYMBOL(blkdev_get);
1466
1467/**
1468 * blkdev_get_by_path - open a block device by name
1469 * @path: path to the block device to open
1470 * @mode: FMODE_* mask
1471 * @holder: exclusive holder identifier
1472 *
1473 * Open the blockdevice described by the device file at @path.  @mode
1474 * and @holder are identical to blkdev_get().
1475 *
1476 * On success, the returned block_device has reference count of one.
1477 *
1478 * CONTEXT:
1479 * Might sleep.
1480 *
1481 * RETURNS:
1482 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1483 */
1484struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1485                                        void *holder)
1486{
1487        struct block_device *bdev;
1488        int err;
1489
1490        bdev = lookup_bdev(path);
1491        if (IS_ERR(bdev))
1492                return bdev;
1493
1494        err = blkdev_get(bdev, mode, holder);
1495        if (err)
1496                return ERR_PTR(err);
1497
1498        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1499                blkdev_put(bdev, mode);
1500                return ERR_PTR(-EACCES);
1501        }
1502
1503        return bdev;
1504}
1505EXPORT_SYMBOL(blkdev_get_by_path);
1506
1507/**
1508 * blkdev_get_by_dev - open a block device by device number
1509 * @dev: device number of block device to open
1510 * @mode: FMODE_* mask
1511 * @holder: exclusive holder identifier
1512 *
1513 * Open the blockdevice described by device number @dev.  @mode and
1514 * @holder are identical to blkdev_get().
1515 *
1516 * Use it ONLY if you really do not have anything better - i.e. when
1517 * you are behind a truly sucky interface and all you are given is a
1518 * device number.  _Never_ to be used for internal purposes.  If you
1519 * ever need it - reconsider your API.
1520 *
1521 * On success, the returned block_device has reference count of one.
1522 *
1523 * CONTEXT:
1524 * Might sleep.
1525 *
1526 * RETURNS:
1527 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1528 */
1529struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1530{
1531        struct block_device *bdev;
1532        int err;
1533
1534        bdev = bdget(dev);
1535        if (!bdev)
1536                return ERR_PTR(-ENOMEM);
1537
1538        err = blkdev_get(bdev, mode, holder);
1539        if (err)
1540                return ERR_PTR(err);
1541
1542        return bdev;
1543}
1544EXPORT_SYMBOL(blkdev_get_by_dev);
1545
1546static int blkdev_open(struct inode * inode, struct file * filp)
1547{
1548        struct block_device *bdev;
1549
1550        /*
1551         * Preserve backwards compatibility and allow large file access
1552         * even if userspace doesn't ask for it explicitly. Some mkfs
1553         * binary needs it. We might want to drop this workaround
1554         * during an unstable branch.
1555         */
1556        filp->f_flags |= O_LARGEFILE;
1557
1558        if (filp->f_flags & O_NDELAY)
1559                filp->f_mode |= FMODE_NDELAY;
1560        if (filp->f_flags & O_EXCL)
1561                filp->f_mode |= FMODE_EXCL;
1562        if ((filp->f_flags & O_ACCMODE) == 3)
1563                filp->f_mode |= FMODE_WRITE_IOCTL;
1564
1565        bdev = bd_acquire(inode);
1566        if (bdev == NULL)
1567                return -ENOMEM;
1568
1569        filp->f_mapping = bdev->bd_inode->i_mapping;
1570
1571        return blkdev_get(bdev, filp->f_mode, filp);
1572}
1573
1574static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1575{
1576        struct gendisk *disk = bdev->bd_disk;
1577        struct block_device *victim = NULL;
1578
1579        mutex_lock_nested(&bdev->bd_mutex, for_part);
1580        if (for_part)
1581                bdev->bd_part_count--;
1582
1583        if (!--bdev->bd_openers) {
1584                WARN_ON_ONCE(bdev->bd_holders);
1585                sync_blockdev(bdev);
1586                kill_bdev(bdev);
1587
1588                bdev_write_inode(bdev);
1589                /*
1590                 * Detaching bdev inode from its wb in __destroy_inode()
1591                 * is too late: the queue which embeds its bdi (along with
1592                 * root wb) can be gone as soon as we put_disk() below.
1593                 */
1594                inode_detach_wb(bdev->bd_inode);
1595        }
1596        if (bdev->bd_contains == bdev) {
1597                if (disk->fops->release)
1598                        disk->fops->release(disk, mode);
1599        }
1600        if (!bdev->bd_openers) {
1601                struct module *owner = disk->fops->owner;
1602
1603                disk_put_part(bdev->bd_part);
1604                bdev->bd_part = NULL;
1605                bdev->bd_disk = NULL;
1606                if (bdev != bdev->bd_contains)
1607                        victim = bdev->bd_contains;
1608                bdev->bd_contains = NULL;
1609
1610                put_disk(disk);
1611                module_put(owner);
1612        }
1613        mutex_unlock(&bdev->bd_mutex);
1614        bdput(bdev);
1615        if (victim)
1616                __blkdev_put(victim, mode, 1);
1617}
1618
1619void blkdev_put(struct block_device *bdev, fmode_t mode)
1620{
1621        mutex_lock(&bdev->bd_mutex);
1622
1623        if (mode & FMODE_EXCL) {
1624                bool bdev_free;
1625
1626                /*
1627                 * Release a claim on the device.  The holder fields
1628                 * are protected with bdev_lock.  bd_mutex is to
1629                 * synchronize disk_holder unlinking.
1630                 */
1631                spin_lock(&bdev_lock);
1632
1633                WARN_ON_ONCE(--bdev->bd_holders < 0);
1634                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1635
1636                /* bd_contains might point to self, check in a separate step */
1637                if ((bdev_free = !bdev->bd_holders))
1638                        bdev->bd_holder = NULL;
1639                if (!bdev->bd_contains->bd_holders)
1640                        bdev->bd_contains->bd_holder = NULL;
1641
1642                spin_unlock(&bdev_lock);
1643
1644                /*
1645                 * If this was the last claim, remove holder link and
1646                 * unblock evpoll if it was a write holder.
1647                 */
1648                if (bdev_free && bdev->bd_write_holder) {
1649                        disk_unblock_events(bdev->bd_disk);
1650                        bdev->bd_write_holder = false;
1651                }
1652        }
1653
1654        /*
1655         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1656         * event.  This is to ensure detection of media removal commanded
1657         * from userland - e.g. eject(1).
1658         */
1659        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1660
1661        mutex_unlock(&bdev->bd_mutex);
1662
1663        __blkdev_put(bdev, mode, 0);
1664}
1665EXPORT_SYMBOL(blkdev_put);
1666
1667static int blkdev_close(struct inode * inode, struct file * filp)
1668{
1669        struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1670        blkdev_put(bdev, filp->f_mode);
1671        return 0;
1672}
1673
1674static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1675{
1676        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1677        fmode_t mode = file->f_mode;
1678
1679        /*
1680         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1681         * to updated it before every ioctl.
1682         */
1683        if (file->f_flags & O_NDELAY)
1684                mode |= FMODE_NDELAY;
1685        else
1686                mode &= ~FMODE_NDELAY;
1687
1688        return blkdev_ioctl(bdev, mode, cmd, arg);
1689}
1690
1691/*
1692 * Write data to the block device.  Only intended for the block device itself
1693 * and the raw driver which basically is a fake block device.
1694 *
1695 * Does not take i_mutex for the write and thus is not for general purpose
1696 * use.
1697 */
1698ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1699{
1700        struct file *file = iocb->ki_filp;
1701        struct inode *bd_inode = bdev_file_inode(file);
1702        loff_t size = i_size_read(bd_inode);
1703        struct blk_plug plug;
1704        ssize_t ret;
1705
1706        if (bdev_read_only(I_BDEV(bd_inode)))
1707                return -EPERM;
1708
1709        if (!iov_iter_count(from))
1710                return 0;
1711
1712        if (iocb->ki_pos >= size)
1713                return -ENOSPC;
1714
1715        iov_iter_truncate(from, size - iocb->ki_pos);
1716
1717        blk_start_plug(&plug);
1718        ret = __generic_file_write_iter(iocb, from);
1719        if (ret > 0)
1720                ret = generic_write_sync(iocb, ret);
1721        blk_finish_plug(&plug);
1722        return ret;
1723}
1724EXPORT_SYMBOL_GPL(blkdev_write_iter);
1725
1726ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1727{
1728        struct file *file = iocb->ki_filp;
1729        struct inode *bd_inode = bdev_file_inode(file);
1730        loff_t size = i_size_read(bd_inode);
1731        loff_t pos = iocb->ki_pos;
1732
1733        if (pos >= size)
1734                return 0;
1735
1736        size -= pos;
1737        iov_iter_truncate(to, size);
1738        return generic_file_read_iter(iocb, to);
1739}
1740EXPORT_SYMBOL_GPL(blkdev_read_iter);
1741
1742/*
1743 * Try to release a page associated with block device when the system
1744 * is under memory pressure.
1745 */
1746static int blkdev_releasepage(struct page *page, gfp_t wait)
1747{
1748        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1749
1750        if (super && super->s_op->bdev_try_to_free_page)
1751                return super->s_op->bdev_try_to_free_page(super, page, wait);
1752
1753        return try_to_free_buffers(page);
1754}
1755
1756static int blkdev_writepages(struct address_space *mapping,
1757                             struct writeback_control *wbc)
1758{
1759        if (dax_mapping(mapping)) {
1760                struct block_device *bdev = I_BDEV(mapping->host);
1761
1762                return dax_writeback_mapping_range(mapping, bdev, wbc);
1763        }
1764        return generic_writepages(mapping, wbc);
1765}
1766
1767static const struct address_space_operations def_blk_aops = {
1768        .readpage       = blkdev_readpage,
1769        .readpages      = blkdev_readpages,
1770        .writepage      = blkdev_writepage,
1771        .write_begin    = blkdev_write_begin,
1772        .write_end      = blkdev_write_end,
1773        .writepages     = blkdev_writepages,
1774        .releasepage    = blkdev_releasepage,
1775        .direct_IO      = blkdev_direct_IO,
1776        .is_dirty_writeback = buffer_check_dirty_writeback,
1777};
1778
1779#define BLKDEV_FALLOC_FL_SUPPORTED                                      \
1780                (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |           \
1781                 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1782
1783static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1784                             loff_t len)
1785{
1786        struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1787        struct request_queue *q = bdev_get_queue(bdev);
1788        struct address_space *mapping;
1789        loff_t end = start + len - 1;
1790        loff_t isize;
1791        int error;
1792
1793        /* Fail if we don't recognize the flags. */
1794        if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1795                return -EOPNOTSUPP;
1796
1797        /* Don't go off the end of the device. */
1798        isize = i_size_read(bdev->bd_inode);
1799        if (start >= isize)
1800                return -EINVAL;
1801        if (end >= isize) {
1802                if (mode & FALLOC_FL_KEEP_SIZE) {
1803                        len = isize - start;
1804                        end = start + len - 1;
1805                } else
1806                        return -EINVAL;
1807        }
1808
1809        /*
1810         * Don't allow IO that isn't aligned to logical block size.
1811         */
1812        if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1813                return -EINVAL;
1814
1815        /* Invalidate the page cache, including dirty pages. */
1816        mapping = bdev->bd_inode->i_mapping;
1817        truncate_inode_pages_range(mapping, start, end);
1818
1819        switch (mode) {
1820        case FALLOC_FL_ZERO_RANGE:
1821        case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1822                error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1823                                            GFP_KERNEL, false);
1824                break;
1825        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1826                /* Only punch if the device can do zeroing discard. */
1827                if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
1828                        return -EOPNOTSUPP;
1829                error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1830                                             GFP_KERNEL, 0);
1831                break;
1832        case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1833                if (!blk_queue_discard(q))
1834                        return -EOPNOTSUPP;
1835                error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1836                                             GFP_KERNEL, 0);
1837                break;
1838        default:
1839                return -EOPNOTSUPP;
1840        }
1841        if (error)
1842                return error;
1843
1844        /*
1845         * Invalidate again; if someone wandered in and dirtied a page,
1846         * the caller will be given -EBUSY.  The third argument is
1847         * inclusive, so the rounding here is safe.
1848         */
1849        return invalidate_inode_pages2_range(mapping,
1850                                             start >> PAGE_SHIFT,
1851                                             end >> PAGE_SHIFT);
1852}
1853
1854const struct file_operations def_blk_fops = {
1855        .open           = blkdev_open,
1856        .release        = blkdev_close,
1857        .llseek         = block_llseek,
1858        .read_iter      = blkdev_read_iter,
1859        .write_iter     = blkdev_write_iter,
1860        .mmap           = generic_file_mmap,
1861        .fsync          = blkdev_fsync,
1862        .unlocked_ioctl = block_ioctl,
1863#ifdef CONFIG_COMPAT
1864        .compat_ioctl   = compat_blkdev_ioctl,
1865#endif
1866        .splice_read    = generic_file_splice_read,
1867        .splice_write   = iter_file_splice_write,
1868        .fallocate      = blkdev_fallocate,
1869};
1870
1871int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1872{
1873        int res;
1874        mm_segment_t old_fs = get_fs();
1875        set_fs(KERNEL_DS);
1876        res = blkdev_ioctl(bdev, 0, cmd, arg);
1877        set_fs(old_fs);
1878        return res;
1879}
1880
1881EXPORT_SYMBOL(ioctl_by_bdev);
1882
1883/**
1884 * lookup_bdev  - lookup a struct block_device by name
1885 * @pathname:   special file representing the block device
1886 *
1887 * Get a reference to the blockdevice at @pathname in the current
1888 * namespace if possible and return it.  Return ERR_PTR(error)
1889 * otherwise.
1890 */
1891struct block_device *lookup_bdev(const char *pathname)
1892{
1893        struct block_device *bdev;
1894        struct inode *inode;
1895        struct path path;
1896        int error;
1897
1898        if (!pathname || !*pathname)
1899                return ERR_PTR(-EINVAL);
1900
1901        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1902        if (error)
1903                return ERR_PTR(error);
1904
1905        inode = d_backing_inode(path.dentry);
1906        error = -ENOTBLK;
1907        if (!S_ISBLK(inode->i_mode))
1908                goto fail;
1909        error = -EACCES;
1910        if (!may_open_dev(&path))
1911                goto fail;
1912        error = -ENOMEM;
1913        bdev = bd_acquire(inode);
1914        if (!bdev)
1915                goto fail;
1916out:
1917        path_put(&path);
1918        return bdev;
1919fail:
1920        bdev = ERR_PTR(error);
1921        goto out;
1922}
1923EXPORT_SYMBOL(lookup_bdev);
1924
1925int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1926{
1927        struct super_block *sb = get_super(bdev);
1928        int res = 0;
1929
1930        if (sb) {
1931                /*
1932                 * no need to lock the super, get_super holds the
1933                 * read mutex so the filesystem cannot go away
1934                 * under us (->put_super runs with the write lock
1935                 * hold).
1936                 */
1937                shrink_dcache_sb(sb);
1938                res = invalidate_inodes(sb, kill_dirty);
1939                drop_super(sb);
1940        }
1941        invalidate_bdev(bdev);
1942        return res;
1943}
1944EXPORT_SYMBOL(__invalidate_device);
1945
1946void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1947{
1948        struct inode *inode, *old_inode = NULL;
1949
1950        spin_lock(&blockdev_superblock->s_inode_list_lock);
1951        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1952                struct address_space *mapping = inode->i_mapping;
1953
1954                spin_lock(&inode->i_lock);
1955                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1956                    mapping->nrpages == 0) {
1957                        spin_unlock(&inode->i_lock);
1958                        continue;
1959                }
1960                __iget(inode);
1961                spin_unlock(&inode->i_lock);
1962                spin_unlock(&blockdev_superblock->s_inode_list_lock);
1963                /*
1964                 * We hold a reference to 'inode' so it couldn't have been
1965                 * removed from s_inodes list while we dropped the
1966                 * s_inode_list_lock  We cannot iput the inode now as we can
1967                 * be holding the last reference and we cannot iput it under
1968                 * s_inode_list_lock. So we keep the reference and iput it
1969                 * later.
1970                 */
1971                iput(old_inode);
1972                old_inode = inode;
1973
1974                func(I_BDEV(inode), arg);
1975
1976                spin_lock(&blockdev_superblock->s_inode_list_lock);
1977        }
1978        spin_unlock(&blockdev_superblock->s_inode_list_lock);
1979        iput(old_inode);
1980}
1981