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/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <asm/uaccess.h>
34 struct block_device bdev;
35 struct inode vfs_inode;
38 static const struct address_space_operations def_blk_aops;
40 static inline struct bdev_inode *BDEV_I(struct inode *inode)
42 return container_of(inode, struct bdev_inode, vfs_inode);
45 inline struct block_device *I_BDEV(struct inode *inode)
47 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
52 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
53 * need to move it onto the dirty list of @dst so that the inode is always on
56 static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
59 struct backing_dev_info *old = inode->i_data.backing_dev_info;
61 if (unlikely(dst == old)) /* deadlock avoidance */
63 bdi_lock_two(&old->wb, &dst->wb);
64 spin_lock(&inode->i_lock);
65 inode->i_data.backing_dev_info = dst;
66 if (inode->i_state & I_DIRTY)
67 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
68 spin_unlock(&inode->i_lock);
69 spin_unlock(&old->wb.list_lock);
70 spin_unlock(&dst->wb.list_lock);
73 sector_t blkdev_max_block(struct block_device *bdev)
75 sector_t retval = ~((sector_t)0);
76 loff_t sz = i_size_read(bdev->bd_inode);
79 unsigned int size = block_size(bdev);
80 unsigned int sizebits = blksize_bits(size);
81 retval = (sz >> sizebits);
86 /* Kill _all_ buffers and pagecache , dirty or not.. */
87 void kill_bdev(struct block_device *bdev)
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
91 if (mapping->nrpages == 0)
95 truncate_inode_pages(mapping, 0);
97 EXPORT_SYMBOL(kill_bdev);
99 /* Invalidate clean unused buffers and pagecache. */
100 void invalidate_bdev(struct block_device *bdev)
102 struct address_space *mapping = bdev->bd_inode->i_mapping;
104 if (mapping->nrpages == 0)
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping, 0, -1);
110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
113 cleancache_invalidate_inode(mapping);
115 EXPORT_SYMBOL(invalidate_bdev);
117 int set_blocksize(struct block_device *bdev, int size)
119 struct address_space *mapping;
121 /* Size must be a power of two, and between 512 and PAGE_SIZE */
122 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
125 /* Size cannot be smaller than the size supported by the device */
126 if (size < bdev_logical_block_size(bdev))
129 /* Prevent starting I/O or mapping the device */
130 percpu_down_write(&bdev->bd_block_size_semaphore);
132 /* Check that the block device is not memory mapped */
133 mapping = bdev->bd_inode->i_mapping;
134 mutex_lock(&mapping->i_mmap_mutex);
135 if (!prio_tree_empty(&mapping->i_mmap) ||
136 !list_empty(&mapping->i_mmap_nonlinear)) {
137 mutex_unlock(&mapping->i_mmap_mutex);
138 percpu_up_write(&bdev->bd_block_size_semaphore);
141 mutex_unlock(&mapping->i_mmap_mutex);
143 /* Don't change the size if it is same as current */
144 if (bdev->bd_block_size != size) {
146 bdev->bd_block_size = size;
147 bdev->bd_inode->i_blkbits = blksize_bits(size);
151 percpu_up_write(&bdev->bd_block_size_semaphore);
156 EXPORT_SYMBOL(set_blocksize);
158 int sb_set_blocksize(struct super_block *sb, int size)
160 if (set_blocksize(sb->s_bdev, size))
162 /* If we get here, we know size is power of two
163 * and it's value is between 512 and PAGE_SIZE */
164 sb->s_blocksize = size;
165 sb->s_blocksize_bits = blksize_bits(size);
166 return sb->s_blocksize;
169 EXPORT_SYMBOL(sb_set_blocksize);
171 int sb_min_blocksize(struct super_block *sb, int size)
173 int minsize = bdev_logical_block_size(sb->s_bdev);
176 return sb_set_blocksize(sb, size);
179 EXPORT_SYMBOL(sb_min_blocksize);
182 blkdev_get_block(struct inode *inode, sector_t iblock,
183 struct buffer_head *bh, int create)
185 if (iblock >= blkdev_max_block(I_BDEV(inode))) {
190 * for reads, we're just trying to fill a partial page.
191 * return a hole, they will have to call get_block again
192 * before they can fill it, and they will get -EIO at that
197 bh->b_bdev = I_BDEV(inode);
198 bh->b_blocknr = iblock;
199 set_buffer_mapped(bh);
204 blkdev_get_blocks(struct inode *inode, sector_t iblock,
205 struct buffer_head *bh, int create)
207 sector_t end_block = blkdev_max_block(I_BDEV(inode));
208 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
210 if ((iblock + max_blocks) > end_block) {
211 max_blocks = end_block - iblock;
212 if ((long)max_blocks <= 0) {
214 return -EIO; /* write fully beyond EOF */
216 * It is a read which is fully beyond EOF. We return
217 * a !buffer_mapped buffer
223 bh->b_bdev = I_BDEV(inode);
224 bh->b_blocknr = iblock;
225 bh->b_size = max_blocks << inode->i_blkbits;
227 set_buffer_mapped(bh);
232 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
233 loff_t offset, unsigned long nr_segs)
235 struct file *file = iocb->ki_filp;
236 struct inode *inode = file->f_mapping->host;
238 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
239 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
242 int __sync_blockdev(struct block_device *bdev, int wait)
247 return filemap_flush(bdev->bd_inode->i_mapping);
248 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
252 * Write out and wait upon all the dirty data associated with a block
253 * device via its mapping. Does not take the superblock lock.
255 int sync_blockdev(struct block_device *bdev)
257 return __sync_blockdev(bdev, 1);
259 EXPORT_SYMBOL(sync_blockdev);
262 * Write out and wait upon all dirty data associated with this
263 * device. Filesystem data as well as the underlying block
264 * device. Takes the superblock lock.
266 int fsync_bdev(struct block_device *bdev)
268 struct super_block *sb = get_super(bdev);
270 int res = sync_filesystem(sb);
274 return sync_blockdev(bdev);
276 EXPORT_SYMBOL(fsync_bdev);
279 * freeze_bdev -- lock a filesystem and force it into a consistent state
280 * @bdev: blockdevice to lock
282 * If a superblock is found on this device, we take the s_umount semaphore
283 * on it to make sure nobody unmounts until the snapshot creation is done.
284 * The reference counter (bd_fsfreeze_count) guarantees that only the last
285 * unfreeze process can unfreeze the frozen filesystem actually when multiple
286 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
287 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
290 struct super_block *freeze_bdev(struct block_device *bdev)
292 struct super_block *sb;
295 mutex_lock(&bdev->bd_fsfreeze_mutex);
296 if (++bdev->bd_fsfreeze_count > 1) {
298 * We don't even need to grab a reference - the first call
299 * to freeze_bdev grab an active reference and only the last
300 * thaw_bdev drops it.
302 sb = get_super(bdev);
304 mutex_unlock(&bdev->bd_fsfreeze_mutex);
308 sb = get_active_super(bdev);
311 error = freeze_super(sb);
313 deactivate_super(sb);
314 bdev->bd_fsfreeze_count--;
315 mutex_unlock(&bdev->bd_fsfreeze_mutex);
316 return ERR_PTR(error);
318 deactivate_super(sb);
321 mutex_unlock(&bdev->bd_fsfreeze_mutex);
322 return sb; /* thaw_bdev releases s->s_umount */
324 EXPORT_SYMBOL(freeze_bdev);
327 * thaw_bdev -- unlock filesystem
328 * @bdev: blockdevice to unlock
329 * @sb: associated superblock
331 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
333 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
337 mutex_lock(&bdev->bd_fsfreeze_mutex);
338 if (!bdev->bd_fsfreeze_count)
342 if (--bdev->bd_fsfreeze_count > 0)
348 error = thaw_super(sb);
350 bdev->bd_fsfreeze_count++;
351 mutex_unlock(&bdev->bd_fsfreeze_mutex);
355 mutex_unlock(&bdev->bd_fsfreeze_mutex);
358 EXPORT_SYMBOL(thaw_bdev);
360 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
362 return block_write_full_page(page, blkdev_get_block, wbc);
365 static int blkdev_readpage(struct file * file, struct page * page)
367 return block_read_full_page(page, blkdev_get_block);
370 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
371 loff_t pos, unsigned len, unsigned flags,
372 struct page **pagep, void **fsdata)
374 return block_write_begin(mapping, pos, len, flags, pagep,
378 static int blkdev_write_end(struct file *file, struct address_space *mapping,
379 loff_t pos, unsigned len, unsigned copied,
380 struct page *page, void *fsdata)
383 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
386 page_cache_release(page);
393 * for a block special file file->f_path.dentry->d_inode->i_size is zero
394 * so we compute the size by hand (just as in block_read/write above)
396 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
398 struct inode *bd_inode = file->f_mapping->host;
402 mutex_lock(&bd_inode->i_mutex);
403 size = i_size_read(bd_inode);
411 offset += file->f_pos;
417 if (offset >= 0 && offset <= size) {
418 if (offset != file->f_pos) {
419 file->f_pos = offset;
424 mutex_unlock(&bd_inode->i_mutex);
428 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
430 struct inode *bd_inode = filp->f_mapping->host;
431 struct block_device *bdev = I_BDEV(bd_inode);
434 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
439 * There is no need to serialise calls to blkdev_issue_flush with
440 * i_mutex and doing so causes performance issues with concurrent
441 * O_SYNC writers to a block device.
443 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
444 if (error == -EOPNOTSUPP)
449 EXPORT_SYMBOL(blkdev_fsync);
455 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
456 static struct kmem_cache * bdev_cachep __read_mostly;
458 static struct inode *bdev_alloc_inode(struct super_block *sb)
460 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
464 if (unlikely(percpu_init_rwsem(&ei->bdev.bd_block_size_semaphore))) {
465 kmem_cache_free(bdev_cachep, ei);
469 return &ei->vfs_inode;
472 static void bdev_i_callback(struct rcu_head *head)
474 struct inode *inode = container_of(head, struct inode, i_rcu);
475 struct bdev_inode *bdi = BDEV_I(inode);
477 percpu_free_rwsem(&bdi->bdev.bd_block_size_semaphore);
479 kmem_cache_free(bdev_cachep, bdi);
482 static void bdev_destroy_inode(struct inode *inode)
484 call_rcu(&inode->i_rcu, bdev_i_callback);
487 static void init_once(void *foo)
489 struct bdev_inode *ei = (struct bdev_inode *) foo;
490 struct block_device *bdev = &ei->bdev;
492 memset(bdev, 0, sizeof(*bdev));
493 mutex_init(&bdev->bd_mutex);
494 INIT_LIST_HEAD(&bdev->bd_inodes);
495 INIT_LIST_HEAD(&bdev->bd_list);
497 INIT_LIST_HEAD(&bdev->bd_holder_disks);
499 inode_init_once(&ei->vfs_inode);
500 /* Initialize mutex for freeze. */
501 mutex_init(&bdev->bd_fsfreeze_mutex);
504 static inline void __bd_forget(struct inode *inode)
506 list_del_init(&inode->i_devices);
507 inode->i_bdev = NULL;
508 inode->i_mapping = &inode->i_data;
511 static void bdev_evict_inode(struct inode *inode)
513 struct block_device *bdev = &BDEV_I(inode)->bdev;
515 truncate_inode_pages(&inode->i_data, 0);
516 invalidate_inode_buffers(inode); /* is it needed here? */
518 spin_lock(&bdev_lock);
519 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
520 __bd_forget(list_entry(p, struct inode, i_devices));
522 list_del_init(&bdev->bd_list);
523 spin_unlock(&bdev_lock);
526 static const struct super_operations bdev_sops = {
527 .statfs = simple_statfs,
528 .alloc_inode = bdev_alloc_inode,
529 .destroy_inode = bdev_destroy_inode,
530 .drop_inode = generic_delete_inode,
531 .evict_inode = bdev_evict_inode,
534 static struct dentry *bd_mount(struct file_system_type *fs_type,
535 int flags, const char *dev_name, void *data)
537 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
540 static struct file_system_type bd_type = {
543 .kill_sb = kill_anon_super,
546 static struct super_block *blockdev_superblock __read_mostly;
548 void __init bdev_cache_init(void)
551 static struct vfsmount *bd_mnt;
553 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
554 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
555 SLAB_MEM_SPREAD|SLAB_PANIC),
557 err = register_filesystem(&bd_type);
559 panic("Cannot register bdev pseudo-fs");
560 bd_mnt = kern_mount(&bd_type);
562 panic("Cannot create bdev pseudo-fs");
563 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
567 * Most likely _very_ bad one - but then it's hardly critical for small
568 * /dev and can be fixed when somebody will need really large one.
569 * Keep in mind that it will be fed through icache hash function too.
571 static inline unsigned long hash(dev_t dev)
573 return MAJOR(dev)+MINOR(dev);
576 static int bdev_test(struct inode *inode, void *data)
578 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
581 static int bdev_set(struct inode *inode, void *data)
583 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
587 static LIST_HEAD(all_bdevs);
589 struct block_device *bdget(dev_t dev)
591 struct block_device *bdev;
594 inode = iget5_locked(blockdev_superblock, hash(dev),
595 bdev_test, bdev_set, &dev);
600 bdev = &BDEV_I(inode)->bdev;
602 if (inode->i_state & I_NEW) {
603 bdev->bd_contains = NULL;
604 bdev->bd_super = NULL;
605 bdev->bd_inode = inode;
606 bdev->bd_block_size = (1 << inode->i_blkbits);
607 bdev->bd_part_count = 0;
608 bdev->bd_invalidated = 0;
609 inode->i_mode = S_IFBLK;
611 inode->i_bdev = bdev;
612 inode->i_data.a_ops = &def_blk_aops;
613 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
614 inode->i_data.backing_dev_info = &default_backing_dev_info;
615 spin_lock(&bdev_lock);
616 list_add(&bdev->bd_list, &all_bdevs);
617 spin_unlock(&bdev_lock);
618 unlock_new_inode(inode);
623 EXPORT_SYMBOL(bdget);
626 * bdgrab -- Grab a reference to an already referenced block device
627 * @bdev: Block device to grab a reference to.
629 struct block_device *bdgrab(struct block_device *bdev)
631 ihold(bdev->bd_inode);
635 long nr_blockdev_pages(void)
637 struct block_device *bdev;
639 spin_lock(&bdev_lock);
640 list_for_each_entry(bdev, &all_bdevs, bd_list) {
641 ret += bdev->bd_inode->i_mapping->nrpages;
643 spin_unlock(&bdev_lock);
647 void bdput(struct block_device *bdev)
649 iput(bdev->bd_inode);
652 EXPORT_SYMBOL(bdput);
654 static struct block_device *bd_acquire(struct inode *inode)
656 struct block_device *bdev;
658 spin_lock(&bdev_lock);
659 bdev = inode->i_bdev;
661 ihold(bdev->bd_inode);
662 spin_unlock(&bdev_lock);
665 spin_unlock(&bdev_lock);
667 bdev = bdget(inode->i_rdev);
669 spin_lock(&bdev_lock);
670 if (!inode->i_bdev) {
672 * We take an additional reference to bd_inode,
673 * and it's released in clear_inode() of inode.
674 * So, we can access it via ->i_mapping always
677 ihold(bdev->bd_inode);
678 inode->i_bdev = bdev;
679 inode->i_mapping = bdev->bd_inode->i_mapping;
680 list_add(&inode->i_devices, &bdev->bd_inodes);
682 spin_unlock(&bdev_lock);
687 static inline int sb_is_blkdev_sb(struct super_block *sb)
689 return sb == blockdev_superblock;
692 /* Call when you free inode */
694 void bd_forget(struct inode *inode)
696 struct block_device *bdev = NULL;
698 spin_lock(&bdev_lock);
700 if (!sb_is_blkdev_sb(inode->i_sb))
701 bdev = inode->i_bdev;
704 spin_unlock(&bdev_lock);
707 iput(bdev->bd_inode);
711 * bd_may_claim - test whether a block device can be claimed
712 * @bdev: block device of interest
713 * @whole: whole block device containing @bdev, may equal @bdev
714 * @holder: holder trying to claim @bdev
716 * Test whether @bdev can be claimed by @holder.
719 * spin_lock(&bdev_lock).
722 * %true if @bdev can be claimed, %false otherwise.
724 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
727 if (bdev->bd_holder == holder)
728 return true; /* already a holder */
729 else if (bdev->bd_holder != NULL)
730 return false; /* held by someone else */
731 else if (bdev->bd_contains == bdev)
732 return true; /* is a whole device which isn't held */
734 else if (whole->bd_holder == bd_may_claim)
735 return true; /* is a partition of a device that is being partitioned */
736 else if (whole->bd_holder != NULL)
737 return false; /* is a partition of a held device */
739 return true; /* is a partition of an un-held device */
743 * bd_prepare_to_claim - prepare to claim a block device
744 * @bdev: block device of interest
745 * @whole: the whole device containing @bdev, may equal @bdev
746 * @holder: holder trying to claim @bdev
748 * Prepare to claim @bdev. This function fails if @bdev is already
749 * claimed by another holder and waits if another claiming is in
750 * progress. This function doesn't actually claim. On successful
751 * return, the caller has ownership of bd_claiming and bd_holder[s].
754 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
758 * 0 if @bdev can be claimed, -EBUSY otherwise.
760 static int bd_prepare_to_claim(struct block_device *bdev,
761 struct block_device *whole, void *holder)
764 /* if someone else claimed, fail */
765 if (!bd_may_claim(bdev, whole, holder))
768 /* if claiming is already in progress, wait for it to finish */
769 if (whole->bd_claiming) {
770 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
773 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
774 spin_unlock(&bdev_lock);
776 finish_wait(wq, &wait);
777 spin_lock(&bdev_lock);
786 * bd_start_claiming - start claiming a block device
787 * @bdev: block device of interest
788 * @holder: holder trying to claim @bdev
790 * @bdev is about to be opened exclusively. Check @bdev can be opened
791 * exclusively and mark that an exclusive open is in progress. Each
792 * successful call to this function must be matched with a call to
793 * either bd_finish_claiming() or bd_abort_claiming() (which do not
796 * This function is used to gain exclusive access to the block device
797 * without actually causing other exclusive open attempts to fail. It
798 * should be used when the open sequence itself requires exclusive
799 * access but may subsequently fail.
805 * Pointer to the block device containing @bdev on success, ERR_PTR()
808 static struct block_device *bd_start_claiming(struct block_device *bdev,
811 struct gendisk *disk;
812 struct block_device *whole;
818 * @bdev might not have been initialized properly yet, look up
819 * and grab the outer block device the hard way.
821 disk = get_gendisk(bdev->bd_dev, &partno);
823 return ERR_PTR(-ENXIO);
826 * Normally, @bdev should equal what's returned from bdget_disk()
827 * if partno is 0; however, some drivers (floppy) use multiple
828 * bdev's for the same physical device and @bdev may be one of the
829 * aliases. Keep @bdev if partno is 0. This means claimer
830 * tracking is broken for those devices but it has always been that
834 whole = bdget_disk(disk, 0);
836 whole = bdgrab(bdev);
838 module_put(disk->fops->owner);
841 return ERR_PTR(-ENOMEM);
843 /* prepare to claim, if successful, mark claiming in progress */
844 spin_lock(&bdev_lock);
846 err = bd_prepare_to_claim(bdev, whole, holder);
848 whole->bd_claiming = holder;
849 spin_unlock(&bdev_lock);
852 spin_unlock(&bdev_lock);
859 struct bd_holder_disk {
860 struct list_head list;
861 struct gendisk *disk;
865 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
866 struct gendisk *disk)
868 struct bd_holder_disk *holder;
870 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
871 if (holder->disk == disk)
876 static int add_symlink(struct kobject *from, struct kobject *to)
878 return sysfs_create_link(from, to, kobject_name(to));
881 static void del_symlink(struct kobject *from, struct kobject *to)
883 sysfs_remove_link(from, kobject_name(to));
887 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
888 * @bdev: the claimed slave bdev
889 * @disk: the holding disk
891 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
893 * This functions creates the following sysfs symlinks.
895 * - from "slaves" directory of the holder @disk to the claimed @bdev
896 * - from "holders" directory of the @bdev to the holder @disk
898 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
899 * passed to bd_link_disk_holder(), then:
901 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
902 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
904 * The caller must have claimed @bdev before calling this function and
905 * ensure that both @bdev and @disk are valid during the creation and
906 * lifetime of these symlinks.
912 * 0 on success, -errno on failure.
914 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
916 struct bd_holder_disk *holder;
919 mutex_lock(&bdev->bd_mutex);
921 WARN_ON_ONCE(!bdev->bd_holder);
923 /* FIXME: remove the following once add_disk() handles errors */
924 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
927 holder = bd_find_holder_disk(bdev, disk);
933 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
939 INIT_LIST_HEAD(&holder->list);
943 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
947 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
951 * bdev could be deleted beneath us which would implicitly destroy
952 * the holder directory. Hold on to it.
954 kobject_get(bdev->bd_part->holder_dir);
956 list_add(&holder->list, &bdev->bd_holder_disks);
960 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
964 mutex_unlock(&bdev->bd_mutex);
967 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
970 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
971 * @bdev: the calimed slave bdev
972 * @disk: the holding disk
974 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
979 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
981 struct bd_holder_disk *holder;
983 mutex_lock(&bdev->bd_mutex);
985 holder = bd_find_holder_disk(bdev, disk);
987 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
988 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
989 del_symlink(bdev->bd_part->holder_dir,
990 &disk_to_dev(disk)->kobj);
991 kobject_put(bdev->bd_part->holder_dir);
992 list_del_init(&holder->list);
996 mutex_unlock(&bdev->bd_mutex);
998 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1002 * flush_disk - invalidates all buffer-cache entries on a disk
1004 * @bdev: struct block device to be flushed
1005 * @kill_dirty: flag to guide handling of dirty inodes
1007 * Invalidates all buffer-cache entries on a disk. It should be called
1008 * when a disk has been changed -- either by a media change or online
1011 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1013 if (__invalidate_device(bdev, kill_dirty)) {
1014 char name[BDEVNAME_SIZE] = "";
1017 disk_name(bdev->bd_disk, 0, name);
1018 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1019 "resized disk %s\n", name);
1024 if (disk_part_scan_enabled(bdev->bd_disk))
1025 bdev->bd_invalidated = 1;
1029 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1030 * @disk: struct gendisk to check
1031 * @bdev: struct bdev to adjust.
1033 * This routine checks to see if the bdev size does not match the disk size
1034 * and adjusts it if it differs.
1036 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1038 loff_t disk_size, bdev_size;
1040 disk_size = (loff_t)get_capacity(disk) << 9;
1041 bdev_size = i_size_read(bdev->bd_inode);
1042 if (disk_size != bdev_size) {
1043 char name[BDEVNAME_SIZE];
1045 disk_name(disk, 0, name);
1047 "%s: detected capacity change from %lld to %lld\n",
1048 name, bdev_size, disk_size);
1049 i_size_write(bdev->bd_inode, disk_size);
1050 flush_disk(bdev, false);
1053 EXPORT_SYMBOL(check_disk_size_change);
1056 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1057 * @disk: struct gendisk to be revalidated
1059 * This routine is a wrapper for lower-level driver's revalidate_disk
1060 * call-backs. It is used to do common pre and post operations needed
1061 * for all revalidate_disk operations.
1063 int revalidate_disk(struct gendisk *disk)
1065 struct block_device *bdev;
1068 if (disk->fops->revalidate_disk)
1069 ret = disk->fops->revalidate_disk(disk);
1071 bdev = bdget_disk(disk, 0);
1075 mutex_lock(&bdev->bd_mutex);
1076 check_disk_size_change(disk, bdev);
1077 mutex_unlock(&bdev->bd_mutex);
1081 EXPORT_SYMBOL(revalidate_disk);
1084 * This routine checks whether a removable media has been changed,
1085 * and invalidates all buffer-cache-entries in that case. This
1086 * is a relatively slow routine, so we have to try to minimize using
1087 * it. Thus it is called only upon a 'mount' or 'open'. This
1088 * is the best way of combining speed and utility, I think.
1089 * People changing diskettes in the middle of an operation deserve
1092 int check_disk_change(struct block_device *bdev)
1094 struct gendisk *disk = bdev->bd_disk;
1095 const struct block_device_operations *bdops = disk->fops;
1096 unsigned int events;
1098 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1099 DISK_EVENT_EJECT_REQUEST);
1100 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1103 flush_disk(bdev, true);
1104 if (bdops->revalidate_disk)
1105 bdops->revalidate_disk(bdev->bd_disk);
1109 EXPORT_SYMBOL(check_disk_change);
1111 void bd_set_size(struct block_device *bdev, loff_t size)
1113 unsigned bsize = bdev_logical_block_size(bdev);
1115 bdev->bd_inode->i_size = size;
1116 while (bsize < PAGE_CACHE_SIZE) {
1121 bdev->bd_block_size = bsize;
1122 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1124 EXPORT_SYMBOL(bd_set_size);
1126 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1131 * mutex_lock(part->bd_mutex)
1132 * mutex_lock_nested(whole->bd_mutex, 1)
1135 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1137 struct gendisk *disk;
1138 struct module *owner;
1143 if (mode & FMODE_READ)
1145 if (mode & FMODE_WRITE)
1148 * hooks: /n/, see "layering violations".
1151 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1161 disk = get_gendisk(bdev->bd_dev, &partno);
1164 owner = disk->fops->owner;
1166 disk_block_events(disk);
1167 mutex_lock_nested(&bdev->bd_mutex, for_part);
1168 if (!bdev->bd_openers) {
1169 bdev->bd_disk = disk;
1170 bdev->bd_queue = disk->queue;
1171 bdev->bd_contains = bdev;
1173 struct backing_dev_info *bdi;
1176 bdev->bd_part = disk_get_part(disk, partno);
1181 if (disk->fops->open) {
1182 ret = disk->fops->open(bdev, mode);
1183 if (ret == -ERESTARTSYS) {
1184 /* Lost a race with 'disk' being
1185 * deleted, try again.
1188 disk_put_part(bdev->bd_part);
1189 bdev->bd_part = NULL;
1190 bdev->bd_disk = NULL;
1191 bdev->bd_queue = NULL;
1192 mutex_unlock(&bdev->bd_mutex);
1193 disk_unblock_events(disk);
1200 if (!ret && !bdev->bd_openers) {
1201 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1202 bdi = blk_get_backing_dev_info(bdev);
1204 bdi = &default_backing_dev_info;
1205 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1209 * If the device is invalidated, rescan partition
1210 * if open succeeded or failed with -ENOMEDIUM.
1211 * The latter is necessary to prevent ghost
1212 * partitions on a removed medium.
1214 if (bdev->bd_invalidated) {
1216 rescan_partitions(disk, bdev);
1217 else if (ret == -ENOMEDIUM)
1218 invalidate_partitions(disk, bdev);
1223 struct block_device *whole;
1224 whole = bdget_disk(disk, 0);
1229 ret = __blkdev_get(whole, mode, 1);
1232 bdev->bd_contains = whole;
1233 bdev_inode_switch_bdi(bdev->bd_inode,
1234 whole->bd_inode->i_data.backing_dev_info);
1235 bdev->bd_part = disk_get_part(disk, partno);
1236 if (!(disk->flags & GENHD_FL_UP) ||
1237 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1241 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1244 if (bdev->bd_contains == bdev) {
1246 if (bdev->bd_disk->fops->open)
1247 ret = bdev->bd_disk->fops->open(bdev, mode);
1248 /* the same as first opener case, read comment there */
1249 if (bdev->bd_invalidated) {
1251 rescan_partitions(bdev->bd_disk, bdev);
1252 else if (ret == -ENOMEDIUM)
1253 invalidate_partitions(bdev->bd_disk, bdev);
1256 goto out_unlock_bdev;
1258 /* only one opener holds refs to the module and disk */
1264 bdev->bd_part_count++;
1265 mutex_unlock(&bdev->bd_mutex);
1266 disk_unblock_events(disk);
1270 disk_put_part(bdev->bd_part);
1271 bdev->bd_disk = NULL;
1272 bdev->bd_part = NULL;
1273 bdev->bd_queue = NULL;
1274 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1275 if (bdev != bdev->bd_contains)
1276 __blkdev_put(bdev->bd_contains, mode, 1);
1277 bdev->bd_contains = NULL;
1279 mutex_unlock(&bdev->bd_mutex);
1280 disk_unblock_events(disk);
1290 * blkdev_get - open a block device
1291 * @bdev: block_device to open
1292 * @mode: FMODE_* mask
1293 * @holder: exclusive holder identifier
1295 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1296 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1297 * @holder is invalid. Exclusive opens may nest for the same @holder.
1299 * On success, the reference count of @bdev is unchanged. On failure,
1306 * 0 on success, -errno on failure.
1308 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1310 struct block_device *whole = NULL;
1313 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1315 if ((mode & FMODE_EXCL) && holder) {
1316 whole = bd_start_claiming(bdev, holder);
1317 if (IS_ERR(whole)) {
1319 return PTR_ERR(whole);
1323 res = __blkdev_get(bdev, mode, 0);
1326 struct gendisk *disk = whole->bd_disk;
1328 /* finish claiming */
1329 mutex_lock(&bdev->bd_mutex);
1330 spin_lock(&bdev_lock);
1333 BUG_ON(!bd_may_claim(bdev, whole, holder));
1335 * Note that for a whole device bd_holders
1336 * will be incremented twice, and bd_holder
1337 * will be set to bd_may_claim before being
1340 whole->bd_holders++;
1341 whole->bd_holder = bd_may_claim;
1343 bdev->bd_holder = holder;
1346 /* tell others that we're done */
1347 BUG_ON(whole->bd_claiming != holder);
1348 whole->bd_claiming = NULL;
1349 wake_up_bit(&whole->bd_claiming, 0);
1351 spin_unlock(&bdev_lock);
1354 * Block event polling for write claims if requested. Any
1355 * write holder makes the write_holder state stick until
1356 * all are released. This is good enough and tracking
1357 * individual writeable reference is too fragile given the
1358 * way @mode is used in blkdev_get/put().
1360 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1361 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1362 bdev->bd_write_holder = true;
1363 disk_block_events(disk);
1366 mutex_unlock(&bdev->bd_mutex);
1372 EXPORT_SYMBOL(blkdev_get);
1375 * blkdev_get_by_path - open a block device by name
1376 * @path: path to the block device to open
1377 * @mode: FMODE_* mask
1378 * @holder: exclusive holder identifier
1380 * Open the blockdevice described by the device file at @path. @mode
1381 * and @holder are identical to blkdev_get().
1383 * On success, the returned block_device has reference count of one.
1389 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1391 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1394 struct block_device *bdev;
1397 bdev = lookup_bdev(path);
1401 err = blkdev_get(bdev, mode, holder);
1403 return ERR_PTR(err);
1405 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1406 blkdev_put(bdev, mode);
1407 return ERR_PTR(-EACCES);
1412 EXPORT_SYMBOL(blkdev_get_by_path);
1415 * blkdev_get_by_dev - open a block device by device number
1416 * @dev: device number of block device to open
1417 * @mode: FMODE_* mask
1418 * @holder: exclusive holder identifier
1420 * Open the blockdevice described by device number @dev. @mode and
1421 * @holder are identical to blkdev_get().
1423 * Use it ONLY if you really do not have anything better - i.e. when
1424 * you are behind a truly sucky interface and all you are given is a
1425 * device number. _Never_ to be used for internal purposes. If you
1426 * ever need it - reconsider your API.
1428 * On success, the returned block_device has reference count of one.
1434 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1436 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1438 struct block_device *bdev;
1443 return ERR_PTR(-ENOMEM);
1445 err = blkdev_get(bdev, mode, holder);
1447 return ERR_PTR(err);
1451 EXPORT_SYMBOL(blkdev_get_by_dev);
1453 static int blkdev_open(struct inode * inode, struct file * filp)
1455 struct block_device *bdev;
1458 * Preserve backwards compatibility and allow large file access
1459 * even if userspace doesn't ask for it explicitly. Some mkfs
1460 * binary needs it. We might want to drop this workaround
1461 * during an unstable branch.
1463 filp->f_flags |= O_LARGEFILE;
1465 if (filp->f_flags & O_NDELAY)
1466 filp->f_mode |= FMODE_NDELAY;
1467 if (filp->f_flags & O_EXCL)
1468 filp->f_mode |= FMODE_EXCL;
1469 if ((filp->f_flags & O_ACCMODE) == 3)
1470 filp->f_mode |= FMODE_WRITE_IOCTL;
1472 bdev = bd_acquire(inode);
1476 filp->f_mapping = bdev->bd_inode->i_mapping;
1478 return blkdev_get(bdev, filp->f_mode, filp);
1481 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1484 struct gendisk *disk = bdev->bd_disk;
1485 struct block_device *victim = NULL;
1487 mutex_lock_nested(&bdev->bd_mutex, for_part);
1489 bdev->bd_part_count--;
1491 if (!--bdev->bd_openers) {
1492 WARN_ON_ONCE(bdev->bd_holders);
1493 sync_blockdev(bdev);
1495 /* ->release can cause the old bdi to disappear,
1496 * so must switch it out first
1498 bdev_inode_switch_bdi(bdev->bd_inode,
1499 &default_backing_dev_info);
1501 if (bdev->bd_contains == bdev) {
1502 if (disk->fops->release)
1503 ret = disk->fops->release(disk, mode);
1505 if (!bdev->bd_openers) {
1506 struct module *owner = disk->fops->owner;
1508 disk_put_part(bdev->bd_part);
1509 bdev->bd_part = NULL;
1510 bdev->bd_disk = NULL;
1511 if (bdev != bdev->bd_contains)
1512 victim = bdev->bd_contains;
1513 bdev->bd_contains = NULL;
1518 mutex_unlock(&bdev->bd_mutex);
1521 __blkdev_put(victim, mode, 1);
1525 int blkdev_put(struct block_device *bdev, fmode_t mode)
1527 mutex_lock(&bdev->bd_mutex);
1529 if (mode & FMODE_EXCL) {
1533 * Release a claim on the device. The holder fields
1534 * are protected with bdev_lock. bd_mutex is to
1535 * synchronize disk_holder unlinking.
1537 spin_lock(&bdev_lock);
1539 WARN_ON_ONCE(--bdev->bd_holders < 0);
1540 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1542 /* bd_contains might point to self, check in a separate step */
1543 if ((bdev_free = !bdev->bd_holders))
1544 bdev->bd_holder = NULL;
1545 if (!bdev->bd_contains->bd_holders)
1546 bdev->bd_contains->bd_holder = NULL;
1548 spin_unlock(&bdev_lock);
1551 * If this was the last claim, remove holder link and
1552 * unblock evpoll if it was a write holder.
1554 if (bdev_free && bdev->bd_write_holder) {
1555 disk_unblock_events(bdev->bd_disk);
1556 bdev->bd_write_holder = false;
1561 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1562 * event. This is to ensure detection of media removal commanded
1563 * from userland - e.g. eject(1).
1565 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1567 mutex_unlock(&bdev->bd_mutex);
1569 return __blkdev_put(bdev, mode, 0);
1571 EXPORT_SYMBOL(blkdev_put);
1573 static int blkdev_close(struct inode * inode, struct file * filp)
1575 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1577 return blkdev_put(bdev, filp->f_mode);
1580 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1582 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1583 fmode_t mode = file->f_mode;
1586 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1587 * to updated it before every ioctl.
1589 if (file->f_flags & O_NDELAY)
1590 mode |= FMODE_NDELAY;
1592 mode &= ~FMODE_NDELAY;
1594 return blkdev_ioctl(bdev, mode, cmd, arg);
1597 ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1598 unsigned long nr_segs, loff_t pos)
1601 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
1603 percpu_down_read(&bdev->bd_block_size_semaphore);
1605 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
1607 percpu_up_read(&bdev->bd_block_size_semaphore);
1611 EXPORT_SYMBOL_GPL(blkdev_aio_read);
1614 * Write data to the block device. Only intended for the block device itself
1615 * and the raw driver which basically is a fake block device.
1617 * Does not take i_mutex for the write and thus is not for general purpose
1620 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1621 unsigned long nr_segs, loff_t pos)
1623 struct file *file = iocb->ki_filp;
1624 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1625 struct blk_plug plug;
1628 BUG_ON(iocb->ki_pos != pos);
1630 blk_start_plug(&plug);
1632 percpu_down_read(&bdev->bd_block_size_semaphore);
1634 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1635 if (ret > 0 || ret == -EIOCBQUEUED) {
1638 err = generic_write_sync(file, pos, ret);
1639 if (err < 0 && ret > 0)
1643 percpu_up_read(&bdev->bd_block_size_semaphore);
1645 blk_finish_plug(&plug);
1649 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1651 static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1654 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1656 percpu_down_read(&bdev->bd_block_size_semaphore);
1658 ret = generic_file_mmap(file, vma);
1660 percpu_up_read(&bdev->bd_block_size_semaphore);
1666 * Try to release a page associated with block device when the system
1667 * is under memory pressure.
1669 static int blkdev_releasepage(struct page *page, gfp_t wait)
1671 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1673 if (super && super->s_op->bdev_try_to_free_page)
1674 return super->s_op->bdev_try_to_free_page(super, page, wait);
1676 return try_to_free_buffers(page);
1679 static const struct address_space_operations def_blk_aops = {
1680 .readpage = blkdev_readpage,
1681 .writepage = blkdev_writepage,
1682 .write_begin = blkdev_write_begin,
1683 .write_end = blkdev_write_end,
1684 .writepages = generic_writepages,
1685 .releasepage = blkdev_releasepage,
1686 .direct_IO = blkdev_direct_IO,
1689 const struct file_operations def_blk_fops = {
1690 .open = blkdev_open,
1691 .release = blkdev_close,
1692 .llseek = block_llseek,
1693 .read = do_sync_read,
1694 .write = do_sync_write,
1695 .aio_read = blkdev_aio_read,
1696 .aio_write = blkdev_aio_write,
1697 .mmap = blkdev_mmap,
1698 .fsync = blkdev_fsync,
1699 .unlocked_ioctl = block_ioctl,
1700 #ifdef CONFIG_COMPAT
1701 .compat_ioctl = compat_blkdev_ioctl,
1703 .splice_read = generic_file_splice_read,
1704 .splice_write = generic_file_splice_write,
1707 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1710 mm_segment_t old_fs = get_fs();
1712 res = blkdev_ioctl(bdev, 0, cmd, arg);
1717 EXPORT_SYMBOL(ioctl_by_bdev);
1720 * lookup_bdev - lookup a struct block_device by name
1721 * @pathname: special file representing the block device
1723 * Get a reference to the blockdevice at @pathname in the current
1724 * namespace if possible and return it. Return ERR_PTR(error)
1727 struct block_device *lookup_bdev(const char *pathname)
1729 struct block_device *bdev;
1730 struct inode *inode;
1734 if (!pathname || !*pathname)
1735 return ERR_PTR(-EINVAL);
1737 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1739 return ERR_PTR(error);
1741 inode = path.dentry->d_inode;
1743 if (!S_ISBLK(inode->i_mode))
1746 if (path.mnt->mnt_flags & MNT_NODEV)
1749 bdev = bd_acquire(inode);
1756 bdev = ERR_PTR(error);
1759 EXPORT_SYMBOL(lookup_bdev);
1761 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1763 struct super_block *sb = get_super(bdev);
1768 * no need to lock the super, get_super holds the
1769 * read mutex so the filesystem cannot go away
1770 * under us (->put_super runs with the write lock
1773 shrink_dcache_sb(sb);
1774 res = invalidate_inodes(sb, kill_dirty);
1777 invalidate_bdev(bdev);
1780 EXPORT_SYMBOL(__invalidate_device);
1782 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1784 struct inode *inode, *old_inode = NULL;
1786 spin_lock(&inode_sb_list_lock);
1787 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1788 struct address_space *mapping = inode->i_mapping;
1790 spin_lock(&inode->i_lock);
1791 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1792 mapping->nrpages == 0) {
1793 spin_unlock(&inode->i_lock);
1797 spin_unlock(&inode->i_lock);
1798 spin_unlock(&inode_sb_list_lock);
1800 * We hold a reference to 'inode' so it couldn't have been
1801 * removed from s_inodes list while we dropped the
1802 * inode_sb_list_lock. We cannot iput the inode now as we can
1803 * be holding the last reference and we cannot iput it under
1804 * inode_sb_list_lock. So we keep the reference and iput it
1810 func(I_BDEV(inode), arg);
1812 spin_lock(&inode_sb_list_lock);
1814 spin_unlock(&inode_sb_list_lock);