4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
52 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
53 * we need to move it onto the dirty list of @dst so that the inode is always
56 static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
59 spin_lock(&inode_lock);
60 inode->i_data.backing_dev_info = dst;
61 if (inode->i_state & I_DIRTY)
62 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
63 spin_unlock(&inode_lock);
66 static sector_t max_block(struct block_device *bdev)
68 sector_t retval = ~((sector_t)0);
69 loff_t sz = i_size_read(bdev->bd_inode);
72 unsigned int size = block_size(bdev);
73 unsigned int sizebits = blksize_bits(size);
74 retval = (sz >> sizebits);
79 /* Kill _all_ buffers and pagecache , dirty or not.. */
80 static void kill_bdev(struct block_device *bdev)
82 if (bdev->bd_inode->i_mapping->nrpages == 0)
85 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
88 int set_blocksize(struct block_device *bdev, int size)
90 /* Size must be a power of two, and between 512 and PAGE_SIZE */
91 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
94 /* Size cannot be smaller than the size supported by the device */
95 if (size < bdev_logical_block_size(bdev))
98 /* Don't change the size if it is same as current */
99 if (bdev->bd_block_size != size) {
101 bdev->bd_block_size = size;
102 bdev->bd_inode->i_blkbits = blksize_bits(size);
108 EXPORT_SYMBOL(set_blocksize);
110 int sb_set_blocksize(struct super_block *sb, int size)
112 if (set_blocksize(sb->s_bdev, size))
114 /* If we get here, we know size is power of two
115 * and it's value is between 512 and PAGE_SIZE */
116 sb->s_blocksize = size;
117 sb->s_blocksize_bits = blksize_bits(size);
118 return sb->s_blocksize;
121 EXPORT_SYMBOL(sb_set_blocksize);
123 int sb_min_blocksize(struct super_block *sb, int size)
125 int minsize = bdev_logical_block_size(sb->s_bdev);
128 return sb_set_blocksize(sb, size);
131 EXPORT_SYMBOL(sb_min_blocksize);
134 blkdev_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh, int create)
137 if (iblock >= max_block(I_BDEV(inode))) {
142 * for reads, we're just trying to fill a partial page.
143 * return a hole, they will have to call get_block again
144 * before they can fill it, and they will get -EIO at that
149 bh->b_bdev = I_BDEV(inode);
150 bh->b_blocknr = iblock;
151 set_buffer_mapped(bh);
156 blkdev_get_blocks(struct inode *inode, sector_t iblock,
157 struct buffer_head *bh, int create)
159 sector_t end_block = max_block(I_BDEV(inode));
160 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
162 if ((iblock + max_blocks) > end_block) {
163 max_blocks = end_block - iblock;
164 if ((long)max_blocks <= 0) {
166 return -EIO; /* write fully beyond EOF */
168 * It is a read which is fully beyond EOF. We return
169 * a !buffer_mapped buffer
175 bh->b_bdev = I_BDEV(inode);
176 bh->b_blocknr = iblock;
177 bh->b_size = max_blocks << inode->i_blkbits;
179 set_buffer_mapped(bh);
184 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
185 loff_t offset, unsigned long nr_segs)
187 struct file *file = iocb->ki_filp;
188 struct inode *inode = file->f_mapping->host;
190 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
191 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
194 int __sync_blockdev(struct block_device *bdev, int wait)
199 return filemap_flush(bdev->bd_inode->i_mapping);
200 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
204 * Write out and wait upon all the dirty data associated with a block
205 * device via its mapping. Does not take the superblock lock.
207 int sync_blockdev(struct block_device *bdev)
209 return __sync_blockdev(bdev, 1);
211 EXPORT_SYMBOL(sync_blockdev);
214 * Write out and wait upon all dirty data associated with this
215 * device. Filesystem data as well as the underlying block
216 * device. Takes the superblock lock.
218 int fsync_bdev(struct block_device *bdev)
220 struct super_block *sb = get_super(bdev);
222 int res = sync_filesystem(sb);
226 return sync_blockdev(bdev);
228 EXPORT_SYMBOL(fsync_bdev);
231 * freeze_bdev -- lock a filesystem and force it into a consistent state
232 * @bdev: blockdevice to lock
234 * If a superblock is found on this device, we take the s_umount semaphore
235 * on it to make sure nobody unmounts until the snapshot creation is done.
236 * The reference counter (bd_fsfreeze_count) guarantees that only the last
237 * unfreeze process can unfreeze the frozen filesystem actually when multiple
238 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
239 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
242 struct super_block *freeze_bdev(struct block_device *bdev)
244 struct super_block *sb;
247 mutex_lock(&bdev->bd_fsfreeze_mutex);
248 if (++bdev->bd_fsfreeze_count > 1) {
250 * We don't even need to grab a reference - the first call
251 * to freeze_bdev grab an active reference and only the last
252 * thaw_bdev drops it.
254 sb = get_super(bdev);
256 mutex_unlock(&bdev->bd_fsfreeze_mutex);
260 sb = get_active_super(bdev);
263 error = freeze_super(sb);
265 deactivate_super(sb);
266 bdev->bd_fsfreeze_count--;
267 mutex_unlock(&bdev->bd_fsfreeze_mutex);
268 return ERR_PTR(error);
270 deactivate_super(sb);
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return sb; /* thaw_bdev releases s->s_umount */
276 EXPORT_SYMBOL(freeze_bdev);
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
285 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
290 if (!bdev->bd_fsfreeze_count)
294 if (--bdev->bd_fsfreeze_count > 0)
300 error = thaw_super(sb);
302 bdev->bd_fsfreeze_count++;
303 mutex_unlock(&bdev->bd_fsfreeze_mutex);
307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
310 EXPORT_SYMBOL(thaw_bdev);
312 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
314 return block_write_full_page(page, blkdev_get_block, wbc);
317 static int blkdev_readpage(struct file * file, struct page * page)
319 return block_read_full_page(page, blkdev_get_block);
322 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned flags,
324 struct page **pagep, void **fsdata)
326 return block_write_begin(mapping, pos, len, flags, pagep,
330 static int blkdev_write_end(struct file *file, struct address_space *mapping,
331 loff_t pos, unsigned len, unsigned copied,
332 struct page *page, void *fsdata)
335 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
338 page_cache_release(page);
345 * for a block special file file->f_path.dentry->d_inode->i_size is zero
346 * so we compute the size by hand (just as in block_read/write above)
348 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
350 struct inode *bd_inode = file->f_mapping->host;
354 mutex_lock(&bd_inode->i_mutex);
355 size = i_size_read(bd_inode);
362 offset += file->f_pos;
365 if (offset >= 0 && offset <= size) {
366 if (offset != file->f_pos) {
367 file->f_pos = offset;
371 mutex_unlock(&bd_inode->i_mutex);
375 int blkdev_fsync(struct file *filp, int datasync)
377 struct inode *bd_inode = filp->f_mapping->host;
378 struct block_device *bdev = I_BDEV(bd_inode);
382 * There is no need to serialise calls to blkdev_issue_flush with
383 * i_mutex and doing so causes performance issues with concurrent
384 * O_SYNC writers to a block device.
386 mutex_unlock(&bd_inode->i_mutex);
388 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
389 if (error == -EOPNOTSUPP)
392 mutex_lock(&bd_inode->i_mutex);
396 EXPORT_SYMBOL(blkdev_fsync);
402 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
403 static struct kmem_cache * bdev_cachep __read_mostly;
405 static struct inode *bdev_alloc_inode(struct super_block *sb)
407 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
410 return &ei->vfs_inode;
413 static void bdev_destroy_inode(struct inode *inode)
415 struct bdev_inode *bdi = BDEV_I(inode);
417 kmem_cache_free(bdev_cachep, bdi);
420 static void init_once(void *foo)
422 struct bdev_inode *ei = (struct bdev_inode *) foo;
423 struct block_device *bdev = &ei->bdev;
425 memset(bdev, 0, sizeof(*bdev));
426 mutex_init(&bdev->bd_mutex);
427 INIT_LIST_HEAD(&bdev->bd_inodes);
428 INIT_LIST_HEAD(&bdev->bd_list);
429 inode_init_once(&ei->vfs_inode);
430 /* Initialize mutex for freeze. */
431 mutex_init(&bdev->bd_fsfreeze_mutex);
434 static inline void __bd_forget(struct inode *inode)
436 list_del_init(&inode->i_devices);
437 inode->i_bdev = NULL;
438 inode->i_mapping = &inode->i_data;
441 static void bdev_evict_inode(struct inode *inode)
443 struct block_device *bdev = &BDEV_I(inode)->bdev;
445 truncate_inode_pages(&inode->i_data, 0);
446 invalidate_inode_buffers(inode); /* is it needed here? */
447 end_writeback(inode);
448 spin_lock(&bdev_lock);
449 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
450 __bd_forget(list_entry(p, struct inode, i_devices));
452 list_del_init(&bdev->bd_list);
453 spin_unlock(&bdev_lock);
456 static const struct super_operations bdev_sops = {
457 .statfs = simple_statfs,
458 .alloc_inode = bdev_alloc_inode,
459 .destroy_inode = bdev_destroy_inode,
460 .drop_inode = generic_delete_inode,
461 .evict_inode = bdev_evict_inode,
464 static struct dentry *bd_mount(struct file_system_type *fs_type,
465 int flags, const char *dev_name, void *data)
467 return mount_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
470 static struct file_system_type bd_type = {
473 .kill_sb = kill_anon_super,
476 struct super_block *blockdev_superblock __read_mostly;
478 void __init bdev_cache_init(void)
481 struct vfsmount *bd_mnt;
483 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
484 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
485 SLAB_MEM_SPREAD|SLAB_PANIC),
487 err = register_filesystem(&bd_type);
489 panic("Cannot register bdev pseudo-fs");
490 bd_mnt = kern_mount(&bd_type);
492 panic("Cannot create bdev pseudo-fs");
494 * This vfsmount structure is only used to obtain the
495 * blockdev_superblock, so tell kmemleak not to report it.
497 kmemleak_not_leak(bd_mnt);
498 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
502 * Most likely _very_ bad one - but then it's hardly critical for small
503 * /dev and can be fixed when somebody will need really large one.
504 * Keep in mind that it will be fed through icache hash function too.
506 static inline unsigned long hash(dev_t dev)
508 return MAJOR(dev)+MINOR(dev);
511 static int bdev_test(struct inode *inode, void *data)
513 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
516 static int bdev_set(struct inode *inode, void *data)
518 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
522 static LIST_HEAD(all_bdevs);
524 struct block_device *bdget(dev_t dev)
526 struct block_device *bdev;
529 inode = iget5_locked(blockdev_superblock, hash(dev),
530 bdev_test, bdev_set, &dev);
535 bdev = &BDEV_I(inode)->bdev;
537 if (inode->i_state & I_NEW) {
538 bdev->bd_contains = NULL;
539 bdev->bd_inode = inode;
540 bdev->bd_block_size = (1 << inode->i_blkbits);
541 bdev->bd_part_count = 0;
542 bdev->bd_invalidated = 0;
543 inode->i_mode = S_IFBLK;
545 inode->i_bdev = bdev;
546 inode->i_data.a_ops = &def_blk_aops;
547 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
548 inode->i_data.backing_dev_info = &default_backing_dev_info;
549 spin_lock(&bdev_lock);
550 list_add(&bdev->bd_list, &all_bdevs);
551 spin_unlock(&bdev_lock);
552 unlock_new_inode(inode);
557 EXPORT_SYMBOL(bdget);
560 * bdgrab -- Grab a reference to an already referenced block device
561 * @bdev: Block device to grab a reference to.
563 struct block_device *bdgrab(struct block_device *bdev)
565 ihold(bdev->bd_inode);
569 long nr_blockdev_pages(void)
571 struct block_device *bdev;
573 spin_lock(&bdev_lock);
574 list_for_each_entry(bdev, &all_bdevs, bd_list) {
575 ret += bdev->bd_inode->i_mapping->nrpages;
577 spin_unlock(&bdev_lock);
581 void bdput(struct block_device *bdev)
583 iput(bdev->bd_inode);
586 EXPORT_SYMBOL(bdput);
588 static struct block_device *bd_acquire(struct inode *inode)
590 struct block_device *bdev;
592 spin_lock(&bdev_lock);
593 bdev = inode->i_bdev;
595 ihold(bdev->bd_inode);
596 spin_unlock(&bdev_lock);
599 spin_unlock(&bdev_lock);
601 bdev = bdget(inode->i_rdev);
603 spin_lock(&bdev_lock);
604 if (!inode->i_bdev) {
606 * We take an additional reference to bd_inode,
607 * and it's released in clear_inode() of inode.
608 * So, we can access it via ->i_mapping always
611 ihold(bdev->bd_inode);
612 inode->i_bdev = bdev;
613 inode->i_mapping = bdev->bd_inode->i_mapping;
614 list_add(&inode->i_devices, &bdev->bd_inodes);
616 spin_unlock(&bdev_lock);
621 /* Call when you free inode */
623 void bd_forget(struct inode *inode)
625 struct block_device *bdev = NULL;
627 spin_lock(&bdev_lock);
629 if (!sb_is_blkdev_sb(inode->i_sb))
630 bdev = inode->i_bdev;
633 spin_unlock(&bdev_lock);
636 iput(bdev->bd_inode);
640 * bd_may_claim - test whether a block device can be claimed
641 * @bdev: block device of interest
642 * @whole: whole block device containing @bdev, may equal @bdev
643 * @holder: holder trying to claim @bdev
645 * Test whther @bdev can be claimed by @holder.
648 * spin_lock(&bdev_lock).
651 * %true if @bdev can be claimed, %false otherwise.
653 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
656 if (bdev->bd_holder == holder)
657 return true; /* already a holder */
658 else if (bdev->bd_holder != NULL)
659 return false; /* held by someone else */
660 else if (bdev->bd_contains == bdev)
661 return true; /* is a whole device which isn't held */
663 else if (whole->bd_holder == bd_claim)
664 return true; /* is a partition of a device that is being partitioned */
665 else if (whole->bd_holder != NULL)
666 return false; /* is a partition of a held device */
668 return true; /* is a partition of an un-held device */
672 * bd_prepare_to_claim - prepare to claim a block device
673 * @bdev: block device of interest
674 * @whole: the whole device containing @bdev, may equal @bdev
675 * @holder: holder trying to claim @bdev
677 * Prepare to claim @bdev. This function fails if @bdev is already
678 * claimed by another holder and waits if another claiming is in
679 * progress. This function doesn't actually claim. On successful
680 * return, the caller has ownership of bd_claiming and bd_holder[s].
683 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
687 * 0 if @bdev can be claimed, -EBUSY otherwise.
689 static int bd_prepare_to_claim(struct block_device *bdev,
690 struct block_device *whole, void *holder)
693 /* if someone else claimed, fail */
694 if (!bd_may_claim(bdev, whole, holder))
697 /* if claiming is already in progress, wait for it to finish */
698 if (whole->bd_claiming) {
699 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
702 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
703 spin_unlock(&bdev_lock);
705 finish_wait(wq, &wait);
706 spin_lock(&bdev_lock);
715 * bd_start_claiming - start claiming a block device
716 * @bdev: block device of interest
717 * @holder: holder trying to claim @bdev
719 * @bdev is about to be opened exclusively. Check @bdev can be opened
720 * exclusively and mark that an exclusive open is in progress. Each
721 * successful call to this function must be matched with a call to
722 * either bd_finish_claiming() or bd_abort_claiming() (which do not
725 * This function is used to gain exclusive access to the block device
726 * without actually causing other exclusive open attempts to fail. It
727 * should be used when the open sequence itself requires exclusive
728 * access but may subsequently fail.
734 * Pointer to the block device containing @bdev on success, ERR_PTR()
737 static struct block_device *bd_start_claiming(struct block_device *bdev,
740 struct gendisk *disk;
741 struct block_device *whole;
747 * @bdev might not have been initialized properly yet, look up
748 * and grab the outer block device the hard way.
750 disk = get_gendisk(bdev->bd_dev, &partno);
752 return ERR_PTR(-ENXIO);
754 whole = bdget_disk(disk, 0);
755 module_put(disk->fops->owner);
758 return ERR_PTR(-ENOMEM);
760 /* prepare to claim, if successful, mark claiming in progress */
761 spin_lock(&bdev_lock);
763 err = bd_prepare_to_claim(bdev, whole, holder);
765 whole->bd_claiming = holder;
766 spin_unlock(&bdev_lock);
769 spin_unlock(&bdev_lock);
775 /* releases bdev_lock */
776 static void __bd_abort_claiming(struct block_device *whole, void *holder)
778 BUG_ON(whole->bd_claiming != holder);
779 whole->bd_claiming = NULL;
780 wake_up_bit(&whole->bd_claiming, 0);
782 spin_unlock(&bdev_lock);
787 * bd_abort_claiming - abort claiming a block device
788 * @whole: whole block device returned by bd_start_claiming()
789 * @holder: holder trying to claim @bdev
791 * Abort a claiming block started by bd_start_claiming(). Note that
792 * @whole is not the block device to be claimed but the whole device
793 * returned by bd_start_claiming().
796 * Grabs and releases bdev_lock.
798 static void bd_abort_claiming(struct block_device *whole, void *holder)
800 spin_lock(&bdev_lock);
801 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
804 /* increment holders when we have a legitimate claim. requires bdev_lock */
805 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
808 /* note that for a whole device bd_holders
809 * will be incremented twice, and bd_holder will
810 * be set to bd_claim before being set to holder
813 whole->bd_holder = bd_claim;
815 bdev->bd_holder = holder;
819 * bd_finish_claiming - finish claiming a block device
820 * @bdev: block device of interest (passed to bd_start_claiming())
821 * @whole: whole block device returned by bd_start_claiming()
822 * @holder: holder trying to claim @bdev
824 * Finish a claiming block started by bd_start_claiming().
827 * Grabs and releases bdev_lock.
829 static void bd_finish_claiming(struct block_device *bdev,
830 struct block_device *whole, void *holder)
832 spin_lock(&bdev_lock);
833 BUG_ON(!bd_may_claim(bdev, whole, holder));
834 __bd_claim(bdev, whole, holder);
835 __bd_abort_claiming(whole, holder); /* not actually an abort */
839 * bd_claim - claim a block device
840 * @bdev: block device to claim
841 * @holder: holder trying to claim @bdev
843 * Try to claim @bdev which must have been opened successfully.
849 * 0 if successful, -EBUSY if @bdev is already claimed.
851 int bd_claim(struct block_device *bdev, void *holder)
853 struct block_device *whole = bdev->bd_contains;
858 spin_lock(&bdev_lock);
859 res = bd_prepare_to_claim(bdev, whole, holder);
861 __bd_claim(bdev, whole, holder);
862 spin_unlock(&bdev_lock);
866 EXPORT_SYMBOL(bd_claim);
868 void bd_release(struct block_device *bdev)
870 spin_lock(&bdev_lock);
871 if (!--bdev->bd_contains->bd_holders)
872 bdev->bd_contains->bd_holder = NULL;
873 if (!--bdev->bd_holders)
874 bdev->bd_holder = NULL;
875 spin_unlock(&bdev_lock);
878 EXPORT_SYMBOL(bd_release);
881 static int add_symlink(struct kobject *from, struct kobject *to)
883 return sysfs_create_link(from, to, kobject_name(to));
886 static void del_symlink(struct kobject *from, struct kobject *to)
888 sysfs_remove_link(from, kobject_name(to));
892 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
893 * @bdev: the claimed slave bdev
894 * @disk: the holding disk
896 * This functions creates the following sysfs symlinks.
898 * - from "slaves" directory of the holder @disk to the claimed @bdev
899 * - from "holders" directory of the @bdev to the holder @disk
901 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
902 * passed to bd_link_disk_holder(), then:
904 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
905 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
907 * The caller must have claimed @bdev before calling this function and
908 * ensure that both @bdev and @disk are valid during the creation and
909 * lifetime of these symlinks.
915 * 0 on success, -errno on failure.
917 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
921 mutex_lock(&bdev->bd_mutex);
923 WARN_ON_ONCE(!bdev->bd_holder || bdev->bd_holder_disk);
925 /* FIXME: remove the following once add_disk() handles errors */
926 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
929 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
933 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
935 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
939 bdev->bd_holder_disk = disk;
941 mutex_unlock(&bdev->bd_mutex);
944 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
946 void bd_unlink_disk_holder(struct block_device *bdev)
948 struct gendisk *disk = bdev->bd_holder_disk;
950 bdev->bd_holder_disk = NULL;
954 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
955 del_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
957 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
961 * Tries to open block device by device number. Use it ONLY if you
962 * really do not have anything better - i.e. when you are behind a
963 * truly sucky interface and all you are given is a device number. _Never_
964 * to be used for internal purposes. If you ever need it - reconsider
967 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
969 struct block_device *bdev = bdget(dev);
972 err = blkdev_get(bdev, mode);
973 return err ? ERR_PTR(err) : bdev;
976 EXPORT_SYMBOL(open_by_devnum);
979 * flush_disk - invalidates all buffer-cache entries on a disk
981 * @bdev: struct block device to be flushed
983 * Invalidates all buffer-cache entries on a disk. It should be called
984 * when a disk has been changed -- either by a media change or online
987 static void flush_disk(struct block_device *bdev)
989 if (__invalidate_device(bdev)) {
990 char name[BDEVNAME_SIZE] = "";
993 disk_name(bdev->bd_disk, 0, name);
994 printk(KERN_WARNING "VFS: busy inodes on changed media or "
995 "resized disk %s\n", name);
1000 if (disk_partitionable(bdev->bd_disk))
1001 bdev->bd_invalidated = 1;
1005 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1006 * @disk: struct gendisk to check
1007 * @bdev: struct bdev to adjust.
1009 * This routine checks to see if the bdev size does not match the disk size
1010 * and adjusts it if it differs.
1012 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1014 loff_t disk_size, bdev_size;
1016 disk_size = (loff_t)get_capacity(disk) << 9;
1017 bdev_size = i_size_read(bdev->bd_inode);
1018 if (disk_size != bdev_size) {
1019 char name[BDEVNAME_SIZE];
1021 disk_name(disk, 0, name);
1023 "%s: detected capacity change from %lld to %lld\n",
1024 name, bdev_size, disk_size);
1025 i_size_write(bdev->bd_inode, disk_size);
1029 EXPORT_SYMBOL(check_disk_size_change);
1032 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1033 * @disk: struct gendisk to be revalidated
1035 * This routine is a wrapper for lower-level driver's revalidate_disk
1036 * call-backs. It is used to do common pre and post operations needed
1037 * for all revalidate_disk operations.
1039 int revalidate_disk(struct gendisk *disk)
1041 struct block_device *bdev;
1044 if (disk->fops->revalidate_disk)
1045 ret = disk->fops->revalidate_disk(disk);
1047 bdev = bdget_disk(disk, 0);
1051 mutex_lock(&bdev->bd_mutex);
1052 check_disk_size_change(disk, bdev);
1053 mutex_unlock(&bdev->bd_mutex);
1057 EXPORT_SYMBOL(revalidate_disk);
1060 * This routine checks whether a removable media has been changed,
1061 * and invalidates all buffer-cache-entries in that case. This
1062 * is a relatively slow routine, so we have to try to minimize using
1063 * it. Thus it is called only upon a 'mount' or 'open'. This
1064 * is the best way of combining speed and utility, I think.
1065 * People changing diskettes in the middle of an operation deserve
1068 int check_disk_change(struct block_device *bdev)
1070 struct gendisk *disk = bdev->bd_disk;
1071 const struct block_device_operations *bdops = disk->fops;
1073 if (!bdops->media_changed)
1075 if (!bdops->media_changed(bdev->bd_disk))
1079 if (bdops->revalidate_disk)
1080 bdops->revalidate_disk(bdev->bd_disk);
1084 EXPORT_SYMBOL(check_disk_change);
1086 void bd_set_size(struct block_device *bdev, loff_t size)
1088 unsigned bsize = bdev_logical_block_size(bdev);
1090 bdev->bd_inode->i_size = size;
1091 while (bsize < PAGE_CACHE_SIZE) {
1096 bdev->bd_block_size = bsize;
1097 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1099 EXPORT_SYMBOL(bd_set_size);
1101 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1106 * mutex_lock(part->bd_mutex)
1107 * mutex_lock_nested(whole->bd_mutex, 1)
1110 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1112 struct gendisk *disk;
1117 if (mode & FMODE_READ)
1119 if (mode & FMODE_WRITE)
1122 * hooks: /n/, see "layering violations".
1125 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1135 disk = get_gendisk(bdev->bd_dev, &partno);
1139 mutex_lock_nested(&bdev->bd_mutex, for_part);
1140 if (!bdev->bd_openers) {
1141 bdev->bd_disk = disk;
1142 bdev->bd_contains = bdev;
1144 struct backing_dev_info *bdi;
1147 bdev->bd_part = disk_get_part(disk, partno);
1151 if (disk->fops->open) {
1152 ret = disk->fops->open(bdev, mode);
1153 if (ret == -ERESTARTSYS) {
1154 /* Lost a race with 'disk' being
1155 * deleted, try again.
1158 disk_put_part(bdev->bd_part);
1159 bdev->bd_part = NULL;
1160 module_put(disk->fops->owner);
1162 bdev->bd_disk = NULL;
1163 mutex_unlock(&bdev->bd_mutex);
1169 if (!bdev->bd_openers) {
1170 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1171 bdi = blk_get_backing_dev_info(bdev);
1173 bdi = &default_backing_dev_info;
1174 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1176 if (bdev->bd_invalidated)
1177 rescan_partitions(disk, bdev);
1179 struct block_device *whole;
1180 whole = bdget_disk(disk, 0);
1185 ret = __blkdev_get(whole, mode, 1);
1188 bdev->bd_contains = whole;
1189 bdev_inode_switch_bdi(bdev->bd_inode,
1190 whole->bd_inode->i_data.backing_dev_info);
1191 bdev->bd_part = disk_get_part(disk, partno);
1192 if (!(disk->flags & GENHD_FL_UP) ||
1193 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1197 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1200 module_put(disk->fops->owner);
1203 if (bdev->bd_contains == bdev) {
1204 if (bdev->bd_disk->fops->open) {
1205 ret = bdev->bd_disk->fops->open(bdev, mode);
1207 goto out_unlock_bdev;
1209 if (bdev->bd_invalidated)
1210 rescan_partitions(bdev->bd_disk, bdev);
1215 bdev->bd_part_count++;
1216 mutex_unlock(&bdev->bd_mutex);
1220 disk_put_part(bdev->bd_part);
1221 bdev->bd_disk = NULL;
1222 bdev->bd_part = NULL;
1223 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1224 if (bdev != bdev->bd_contains)
1225 __blkdev_put(bdev->bd_contains, mode, 1);
1226 bdev->bd_contains = NULL;
1228 mutex_unlock(&bdev->bd_mutex);
1231 module_put(disk->fops->owner);
1238 int blkdev_get(struct block_device *bdev, fmode_t mode)
1240 return __blkdev_get(bdev, mode, 0);
1242 EXPORT_SYMBOL(blkdev_get);
1244 static int blkdev_open(struct inode * inode, struct file * filp)
1246 struct block_device *whole = NULL;
1247 struct block_device *bdev;
1251 * Preserve backwards compatibility and allow large file access
1252 * even if userspace doesn't ask for it explicitly. Some mkfs
1253 * binary needs it. We might want to drop this workaround
1254 * during an unstable branch.
1256 filp->f_flags |= O_LARGEFILE;
1258 if (filp->f_flags & O_NDELAY)
1259 filp->f_mode |= FMODE_NDELAY;
1260 if (filp->f_flags & O_EXCL)
1261 filp->f_mode |= FMODE_EXCL;
1262 if ((filp->f_flags & O_ACCMODE) == 3)
1263 filp->f_mode |= FMODE_WRITE_IOCTL;
1265 bdev = bd_acquire(inode);
1269 if (filp->f_mode & FMODE_EXCL) {
1270 whole = bd_start_claiming(bdev, filp);
1271 if (IS_ERR(whole)) {
1273 return PTR_ERR(whole);
1277 filp->f_mapping = bdev->bd_inode->i_mapping;
1279 res = blkdev_get(bdev, filp->f_mode);
1283 bd_finish_claiming(bdev, whole, filp);
1285 bd_abort_claiming(whole, filp);
1291 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1294 struct gendisk *disk = bdev->bd_disk;
1295 struct block_device *victim = NULL;
1297 mutex_lock_nested(&bdev->bd_mutex, for_part);
1299 bdev->bd_part_count--;
1301 if (!--bdev->bd_openers) {
1302 sync_blockdev(bdev);
1305 if (bdev->bd_contains == bdev) {
1306 if (disk->fops->release)
1307 ret = disk->fops->release(disk, mode);
1309 if (!bdev->bd_openers) {
1310 struct module *owner = disk->fops->owner;
1314 disk_put_part(bdev->bd_part);
1315 bdev->bd_part = NULL;
1316 bdev->bd_disk = NULL;
1317 bdev_inode_switch_bdi(bdev->bd_inode,
1318 &default_backing_dev_info);
1319 if (bdev != bdev->bd_contains)
1320 victim = bdev->bd_contains;
1321 bdev->bd_contains = NULL;
1323 mutex_unlock(&bdev->bd_mutex);
1326 __blkdev_put(victim, mode, 1);
1330 int blkdev_put(struct block_device *bdev, fmode_t mode)
1332 return __blkdev_put(bdev, mode, 0);
1334 EXPORT_SYMBOL(blkdev_put);
1336 static int blkdev_close(struct inode * inode, struct file * filp)
1338 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1339 if (bdev->bd_holder == filp)
1341 return blkdev_put(bdev, filp->f_mode);
1344 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1346 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1347 fmode_t mode = file->f_mode;
1350 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1351 * to updated it before every ioctl.
1353 if (file->f_flags & O_NDELAY)
1354 mode |= FMODE_NDELAY;
1356 mode &= ~FMODE_NDELAY;
1358 return blkdev_ioctl(bdev, mode, cmd, arg);
1362 * Write data to the block device. Only intended for the block device itself
1363 * and the raw driver which basically is a fake block device.
1365 * Does not take i_mutex for the write and thus is not for general purpose
1368 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1369 unsigned long nr_segs, loff_t pos)
1371 struct file *file = iocb->ki_filp;
1374 BUG_ON(iocb->ki_pos != pos);
1376 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1377 if (ret > 0 || ret == -EIOCBQUEUED) {
1380 err = generic_write_sync(file, pos, ret);
1381 if (err < 0 && ret > 0)
1386 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1389 * Try to release a page associated with block device when the system
1390 * is under memory pressure.
1392 static int blkdev_releasepage(struct page *page, gfp_t wait)
1394 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1396 if (super && super->s_op->bdev_try_to_free_page)
1397 return super->s_op->bdev_try_to_free_page(super, page, wait);
1399 return try_to_free_buffers(page);
1402 static const struct address_space_operations def_blk_aops = {
1403 .readpage = blkdev_readpage,
1404 .writepage = blkdev_writepage,
1405 .sync_page = block_sync_page,
1406 .write_begin = blkdev_write_begin,
1407 .write_end = blkdev_write_end,
1408 .writepages = generic_writepages,
1409 .releasepage = blkdev_releasepage,
1410 .direct_IO = blkdev_direct_IO,
1413 const struct file_operations def_blk_fops = {
1414 .open = blkdev_open,
1415 .release = blkdev_close,
1416 .llseek = block_llseek,
1417 .read = do_sync_read,
1418 .write = do_sync_write,
1419 .aio_read = generic_file_aio_read,
1420 .aio_write = blkdev_aio_write,
1421 .mmap = generic_file_mmap,
1422 .fsync = blkdev_fsync,
1423 .unlocked_ioctl = block_ioctl,
1424 #ifdef CONFIG_COMPAT
1425 .compat_ioctl = compat_blkdev_ioctl,
1427 .splice_read = generic_file_splice_read,
1428 .splice_write = generic_file_splice_write,
1431 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1434 mm_segment_t old_fs = get_fs();
1436 res = blkdev_ioctl(bdev, 0, cmd, arg);
1441 EXPORT_SYMBOL(ioctl_by_bdev);
1444 * lookup_bdev - lookup a struct block_device by name
1445 * @pathname: special file representing the block device
1447 * Get a reference to the blockdevice at @pathname in the current
1448 * namespace if possible and return it. Return ERR_PTR(error)
1451 struct block_device *lookup_bdev(const char *pathname)
1453 struct block_device *bdev;
1454 struct inode *inode;
1458 if (!pathname || !*pathname)
1459 return ERR_PTR(-EINVAL);
1461 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1463 return ERR_PTR(error);
1465 inode = path.dentry->d_inode;
1467 if (!S_ISBLK(inode->i_mode))
1470 if (path.mnt->mnt_flags & MNT_NODEV)
1473 bdev = bd_acquire(inode);
1480 bdev = ERR_PTR(error);
1483 EXPORT_SYMBOL(lookup_bdev);
1486 * open_bdev_exclusive - open a block device by name and set it up for use
1488 * @path: special file representing the block device
1489 * @mode: FMODE_... combination to pass be used
1490 * @holder: owner for exclusion
1492 * Open the blockdevice described by the special file at @path, claim it
1495 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1497 struct block_device *bdev, *whole;
1500 bdev = lookup_bdev(path);
1504 whole = bd_start_claiming(bdev, holder);
1505 if (IS_ERR(whole)) {
1510 error = blkdev_get(bdev, mode);
1512 goto out_abort_claiming;
1515 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1516 goto out_blkdev_put;
1518 bd_finish_claiming(bdev, whole, holder);
1522 blkdev_put(bdev, mode);
1524 bd_abort_claiming(whole, holder);
1525 return ERR_PTR(error);
1528 EXPORT_SYMBOL(open_bdev_exclusive);
1531 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1533 * @bdev: blockdevice to close
1534 * @mode: mode, must match that used to open.
1536 * This is the counterpart to open_bdev_exclusive().
1538 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1541 blkdev_put(bdev, mode);
1544 EXPORT_SYMBOL(close_bdev_exclusive);
1546 int __invalidate_device(struct block_device *bdev)
1548 struct super_block *sb = get_super(bdev);
1553 * no need to lock the super, get_super holds the
1554 * read mutex so the filesystem cannot go away
1555 * under us (->put_super runs with the write lock
1558 shrink_dcache_sb(sb);
1559 res = invalidate_inodes(sb);
1562 invalidate_bdev(bdev);
1565 EXPORT_SYMBOL(__invalidate_device);