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);
699 if (!sb_is_blkdev_sb(inode->i_sb))
700 bdev = inode->i_bdev;
702 spin_unlock(&bdev_lock);
705 iput(bdev->bd_inode);
709 * bd_may_claim - test whether a block device can be claimed
710 * @bdev: block device of interest
711 * @whole: whole block device containing @bdev, may equal @bdev
712 * @holder: holder trying to claim @bdev
714 * Test whether @bdev can be claimed by @holder.
717 * spin_lock(&bdev_lock).
720 * %true if @bdev can be claimed, %false otherwise.
722 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
725 if (bdev->bd_holder == holder)
726 return true; /* already a holder */
727 else if (bdev->bd_holder != NULL)
728 return false; /* held by someone else */
729 else if (bdev->bd_contains == bdev)
730 return true; /* is a whole device which isn't held */
732 else if (whole->bd_holder == bd_may_claim)
733 return true; /* is a partition of a device that is being partitioned */
734 else if (whole->bd_holder != NULL)
735 return false; /* is a partition of a held device */
737 return true; /* is a partition of an un-held device */
741 * bd_prepare_to_claim - prepare to claim a block device
742 * @bdev: block device of interest
743 * @whole: the whole device containing @bdev, may equal @bdev
744 * @holder: holder trying to claim @bdev
746 * Prepare to claim @bdev. This function fails if @bdev is already
747 * claimed by another holder and waits if another claiming is in
748 * progress. This function doesn't actually claim. On successful
749 * return, the caller has ownership of bd_claiming and bd_holder[s].
752 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
756 * 0 if @bdev can be claimed, -EBUSY otherwise.
758 static int bd_prepare_to_claim(struct block_device *bdev,
759 struct block_device *whole, void *holder)
762 /* if someone else claimed, fail */
763 if (!bd_may_claim(bdev, whole, holder))
766 /* if claiming is already in progress, wait for it to finish */
767 if (whole->bd_claiming) {
768 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
771 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
772 spin_unlock(&bdev_lock);
774 finish_wait(wq, &wait);
775 spin_lock(&bdev_lock);
784 * bd_start_claiming - start claiming a block device
785 * @bdev: block device of interest
786 * @holder: holder trying to claim @bdev
788 * @bdev is about to be opened exclusively. Check @bdev can be opened
789 * exclusively and mark that an exclusive open is in progress. Each
790 * successful call to this function must be matched with a call to
791 * either bd_finish_claiming() or bd_abort_claiming() (which do not
794 * This function is used to gain exclusive access to the block device
795 * without actually causing other exclusive open attempts to fail. It
796 * should be used when the open sequence itself requires exclusive
797 * access but may subsequently fail.
803 * Pointer to the block device containing @bdev on success, ERR_PTR()
806 static struct block_device *bd_start_claiming(struct block_device *bdev,
809 struct gendisk *disk;
810 struct block_device *whole;
816 * @bdev might not have been initialized properly yet, look up
817 * and grab the outer block device the hard way.
819 disk = get_gendisk(bdev->bd_dev, &partno);
821 return ERR_PTR(-ENXIO);
824 * Normally, @bdev should equal what's returned from bdget_disk()
825 * if partno is 0; however, some drivers (floppy) use multiple
826 * bdev's for the same physical device and @bdev may be one of the
827 * aliases. Keep @bdev if partno is 0. This means claimer
828 * tracking is broken for those devices but it has always been that
832 whole = bdget_disk(disk, 0);
834 whole = bdgrab(bdev);
836 module_put(disk->fops->owner);
839 return ERR_PTR(-ENOMEM);
841 /* prepare to claim, if successful, mark claiming in progress */
842 spin_lock(&bdev_lock);
844 err = bd_prepare_to_claim(bdev, whole, holder);
846 whole->bd_claiming = holder;
847 spin_unlock(&bdev_lock);
850 spin_unlock(&bdev_lock);
857 struct bd_holder_disk {
858 struct list_head list;
859 struct gendisk *disk;
863 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
864 struct gendisk *disk)
866 struct bd_holder_disk *holder;
868 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
869 if (holder->disk == disk)
874 static int add_symlink(struct kobject *from, struct kobject *to)
876 return sysfs_create_link(from, to, kobject_name(to));
879 static void del_symlink(struct kobject *from, struct kobject *to)
881 sysfs_remove_link(from, kobject_name(to));
885 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
886 * @bdev: the claimed slave bdev
887 * @disk: the holding disk
889 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
891 * This functions creates the following sysfs symlinks.
893 * - from "slaves" directory of the holder @disk to the claimed @bdev
894 * - from "holders" directory of the @bdev to the holder @disk
896 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
897 * passed to bd_link_disk_holder(), then:
899 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
900 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
902 * The caller must have claimed @bdev before calling this function and
903 * ensure that both @bdev and @disk are valid during the creation and
904 * lifetime of these symlinks.
910 * 0 on success, -errno on failure.
912 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
914 struct bd_holder_disk *holder;
917 mutex_lock(&bdev->bd_mutex);
919 WARN_ON_ONCE(!bdev->bd_holder);
921 /* FIXME: remove the following once add_disk() handles errors */
922 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
925 holder = bd_find_holder_disk(bdev, disk);
931 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
937 INIT_LIST_HEAD(&holder->list);
941 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
945 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
949 * bdev could be deleted beneath us which would implicitly destroy
950 * the holder directory. Hold on to it.
952 kobject_get(bdev->bd_part->holder_dir);
954 list_add(&holder->list, &bdev->bd_holder_disks);
958 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
962 mutex_unlock(&bdev->bd_mutex);
965 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
968 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
969 * @bdev: the calimed slave bdev
970 * @disk: the holding disk
972 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
977 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
979 struct bd_holder_disk *holder;
981 mutex_lock(&bdev->bd_mutex);
983 holder = bd_find_holder_disk(bdev, disk);
985 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
986 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
987 del_symlink(bdev->bd_part->holder_dir,
988 &disk_to_dev(disk)->kobj);
989 kobject_put(bdev->bd_part->holder_dir);
990 list_del_init(&holder->list);
994 mutex_unlock(&bdev->bd_mutex);
996 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1000 * flush_disk - invalidates all buffer-cache entries on a disk
1002 * @bdev: struct block device to be flushed
1003 * @kill_dirty: flag to guide handling of dirty inodes
1005 * Invalidates all buffer-cache entries on a disk. It should be called
1006 * when a disk has been changed -- either by a media change or online
1009 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1011 if (__invalidate_device(bdev, kill_dirty)) {
1012 char name[BDEVNAME_SIZE] = "";
1015 disk_name(bdev->bd_disk, 0, name);
1016 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1017 "resized disk %s\n", name);
1022 if (disk_part_scan_enabled(bdev->bd_disk))
1023 bdev->bd_invalidated = 1;
1027 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1028 * @disk: struct gendisk to check
1029 * @bdev: struct bdev to adjust.
1031 * This routine checks to see if the bdev size does not match the disk size
1032 * and adjusts it if it differs.
1034 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1036 loff_t disk_size, bdev_size;
1038 disk_size = (loff_t)get_capacity(disk) << 9;
1039 bdev_size = i_size_read(bdev->bd_inode);
1040 if (disk_size != bdev_size) {
1041 char name[BDEVNAME_SIZE];
1043 disk_name(disk, 0, name);
1045 "%s: detected capacity change from %lld to %lld\n",
1046 name, bdev_size, disk_size);
1047 i_size_write(bdev->bd_inode, disk_size);
1048 flush_disk(bdev, false);
1051 EXPORT_SYMBOL(check_disk_size_change);
1054 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1055 * @disk: struct gendisk to be revalidated
1057 * This routine is a wrapper for lower-level driver's revalidate_disk
1058 * call-backs. It is used to do common pre and post operations needed
1059 * for all revalidate_disk operations.
1061 int revalidate_disk(struct gendisk *disk)
1063 struct block_device *bdev;
1066 if (disk->fops->revalidate_disk)
1067 ret = disk->fops->revalidate_disk(disk);
1069 bdev = bdget_disk(disk, 0);
1073 mutex_lock(&bdev->bd_mutex);
1074 check_disk_size_change(disk, bdev);
1075 mutex_unlock(&bdev->bd_mutex);
1079 EXPORT_SYMBOL(revalidate_disk);
1082 * This routine checks whether a removable media has been changed,
1083 * and invalidates all buffer-cache-entries in that case. This
1084 * is a relatively slow routine, so we have to try to minimize using
1085 * it. Thus it is called only upon a 'mount' or 'open'. This
1086 * is the best way of combining speed and utility, I think.
1087 * People changing diskettes in the middle of an operation deserve
1090 int check_disk_change(struct block_device *bdev)
1092 struct gendisk *disk = bdev->bd_disk;
1093 const struct block_device_operations *bdops = disk->fops;
1094 unsigned int events;
1096 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1097 DISK_EVENT_EJECT_REQUEST);
1098 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1101 flush_disk(bdev, true);
1102 if (bdops->revalidate_disk)
1103 bdops->revalidate_disk(bdev->bd_disk);
1107 EXPORT_SYMBOL(check_disk_change);
1109 void bd_set_size(struct block_device *bdev, loff_t size)
1111 unsigned bsize = bdev_logical_block_size(bdev);
1113 bdev->bd_inode->i_size = size;
1114 while (bsize < PAGE_CACHE_SIZE) {
1119 bdev->bd_block_size = bsize;
1120 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1122 EXPORT_SYMBOL(bd_set_size);
1124 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1129 * mutex_lock(part->bd_mutex)
1130 * mutex_lock_nested(whole->bd_mutex, 1)
1133 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1135 struct gendisk *disk;
1136 struct module *owner;
1141 if (mode & FMODE_READ)
1143 if (mode & FMODE_WRITE)
1146 * hooks: /n/, see "layering violations".
1149 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1159 disk = get_gendisk(bdev->bd_dev, &partno);
1162 owner = disk->fops->owner;
1164 disk_block_events(disk);
1165 mutex_lock_nested(&bdev->bd_mutex, for_part);
1166 if (!bdev->bd_openers) {
1167 bdev->bd_disk = disk;
1168 bdev->bd_queue = disk->queue;
1169 bdev->bd_contains = bdev;
1171 struct backing_dev_info *bdi;
1174 bdev->bd_part = disk_get_part(disk, partno);
1179 if (disk->fops->open) {
1180 ret = disk->fops->open(bdev, mode);
1181 if (ret == -ERESTARTSYS) {
1182 /* Lost a race with 'disk' being
1183 * deleted, try again.
1186 disk_put_part(bdev->bd_part);
1187 bdev->bd_part = NULL;
1188 bdev->bd_disk = NULL;
1189 bdev->bd_queue = NULL;
1190 mutex_unlock(&bdev->bd_mutex);
1191 disk_unblock_events(disk);
1198 if (!ret && !bdev->bd_openers) {
1199 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1200 bdi = blk_get_backing_dev_info(bdev);
1202 bdi = &default_backing_dev_info;
1203 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1207 * If the device is invalidated, rescan partition
1208 * if open succeeded or failed with -ENOMEDIUM.
1209 * The latter is necessary to prevent ghost
1210 * partitions on a removed medium.
1212 if (bdev->bd_invalidated) {
1214 rescan_partitions(disk, bdev);
1215 else if (ret == -ENOMEDIUM)
1216 invalidate_partitions(disk, bdev);
1221 struct block_device *whole;
1222 whole = bdget_disk(disk, 0);
1227 ret = __blkdev_get(whole, mode, 1);
1230 bdev->bd_contains = whole;
1231 bdev_inode_switch_bdi(bdev->bd_inode,
1232 whole->bd_inode->i_data.backing_dev_info);
1233 bdev->bd_part = disk_get_part(disk, partno);
1234 if (!(disk->flags & GENHD_FL_UP) ||
1235 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1239 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1242 if (bdev->bd_contains == bdev) {
1244 if (bdev->bd_disk->fops->open)
1245 ret = bdev->bd_disk->fops->open(bdev, mode);
1246 /* the same as first opener case, read comment there */
1247 if (bdev->bd_invalidated) {
1249 rescan_partitions(bdev->bd_disk, bdev);
1250 else if (ret == -ENOMEDIUM)
1251 invalidate_partitions(bdev->bd_disk, bdev);
1254 goto out_unlock_bdev;
1256 /* only one opener holds refs to the module and disk */
1262 bdev->bd_part_count++;
1263 mutex_unlock(&bdev->bd_mutex);
1264 disk_unblock_events(disk);
1268 disk_put_part(bdev->bd_part);
1269 bdev->bd_disk = NULL;
1270 bdev->bd_part = NULL;
1271 bdev->bd_queue = NULL;
1272 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1273 if (bdev != bdev->bd_contains)
1274 __blkdev_put(bdev->bd_contains, mode, 1);
1275 bdev->bd_contains = NULL;
1277 mutex_unlock(&bdev->bd_mutex);
1278 disk_unblock_events(disk);
1288 * blkdev_get - open a block device
1289 * @bdev: block_device to open
1290 * @mode: FMODE_* mask
1291 * @holder: exclusive holder identifier
1293 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1294 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1295 * @holder is invalid. Exclusive opens may nest for the same @holder.
1297 * On success, the reference count of @bdev is unchanged. On failure,
1304 * 0 on success, -errno on failure.
1306 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1308 struct block_device *whole = NULL;
1311 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1313 if ((mode & FMODE_EXCL) && holder) {
1314 whole = bd_start_claiming(bdev, holder);
1315 if (IS_ERR(whole)) {
1317 return PTR_ERR(whole);
1321 res = __blkdev_get(bdev, mode, 0);
1324 struct gendisk *disk = whole->bd_disk;
1326 /* finish claiming */
1327 mutex_lock(&bdev->bd_mutex);
1328 spin_lock(&bdev_lock);
1331 BUG_ON(!bd_may_claim(bdev, whole, holder));
1333 * Note that for a whole device bd_holders
1334 * will be incremented twice, and bd_holder
1335 * will be set to bd_may_claim before being
1338 whole->bd_holders++;
1339 whole->bd_holder = bd_may_claim;
1341 bdev->bd_holder = holder;
1344 /* tell others that we're done */
1345 BUG_ON(whole->bd_claiming != holder);
1346 whole->bd_claiming = NULL;
1347 wake_up_bit(&whole->bd_claiming, 0);
1349 spin_unlock(&bdev_lock);
1352 * Block event polling for write claims if requested. Any
1353 * write holder makes the write_holder state stick until
1354 * all are released. This is good enough and tracking
1355 * individual writeable reference is too fragile given the
1356 * way @mode is used in blkdev_get/put().
1358 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1359 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1360 bdev->bd_write_holder = true;
1361 disk_block_events(disk);
1364 mutex_unlock(&bdev->bd_mutex);
1370 EXPORT_SYMBOL(blkdev_get);
1373 * blkdev_get_by_path - open a block device by name
1374 * @path: path to the block device to open
1375 * @mode: FMODE_* mask
1376 * @holder: exclusive holder identifier
1378 * Open the blockdevice described by the device file at @path. @mode
1379 * and @holder are identical to blkdev_get().
1381 * On success, the returned block_device has reference count of one.
1387 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1389 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1392 struct block_device *bdev;
1395 bdev = lookup_bdev(path);
1399 err = blkdev_get(bdev, mode, holder);
1401 return ERR_PTR(err);
1403 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1404 blkdev_put(bdev, mode);
1405 return ERR_PTR(-EACCES);
1410 EXPORT_SYMBOL(blkdev_get_by_path);
1413 * blkdev_get_by_dev - open a block device by device number
1414 * @dev: device number of block device to open
1415 * @mode: FMODE_* mask
1416 * @holder: exclusive holder identifier
1418 * Open the blockdevice described by device number @dev. @mode and
1419 * @holder are identical to blkdev_get().
1421 * Use it ONLY if you really do not have anything better - i.e. when
1422 * you are behind a truly sucky interface and all you are given is a
1423 * device number. _Never_ to be used for internal purposes. If you
1424 * ever need it - reconsider your API.
1426 * On success, the returned block_device has reference count of one.
1432 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1434 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1436 struct block_device *bdev;
1441 return ERR_PTR(-ENOMEM);
1443 err = blkdev_get(bdev, mode, holder);
1445 return ERR_PTR(err);
1449 EXPORT_SYMBOL(blkdev_get_by_dev);
1451 static int blkdev_open(struct inode * inode, struct file * filp)
1453 struct block_device *bdev;
1456 * Preserve backwards compatibility and allow large file access
1457 * even if userspace doesn't ask for it explicitly. Some mkfs
1458 * binary needs it. We might want to drop this workaround
1459 * during an unstable branch.
1461 filp->f_flags |= O_LARGEFILE;
1463 if (filp->f_flags & O_NDELAY)
1464 filp->f_mode |= FMODE_NDELAY;
1465 if (filp->f_flags & O_EXCL)
1466 filp->f_mode |= FMODE_EXCL;
1467 if ((filp->f_flags & O_ACCMODE) == 3)
1468 filp->f_mode |= FMODE_WRITE_IOCTL;
1470 bdev = bd_acquire(inode);
1474 filp->f_mapping = bdev->bd_inode->i_mapping;
1476 return blkdev_get(bdev, filp->f_mode, filp);
1479 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1482 struct gendisk *disk = bdev->bd_disk;
1483 struct block_device *victim = NULL;
1485 mutex_lock_nested(&bdev->bd_mutex, for_part);
1487 bdev->bd_part_count--;
1489 if (!--bdev->bd_openers) {
1490 WARN_ON_ONCE(bdev->bd_holders);
1491 sync_blockdev(bdev);
1493 /* ->release can cause the old bdi to disappear,
1494 * so must switch it out first
1496 bdev_inode_switch_bdi(bdev->bd_inode,
1497 &default_backing_dev_info);
1499 if (bdev->bd_contains == bdev) {
1500 if (disk->fops->release)
1501 ret = disk->fops->release(disk, mode);
1503 if (!bdev->bd_openers) {
1504 struct module *owner = disk->fops->owner;
1506 disk_put_part(bdev->bd_part);
1507 bdev->bd_part = NULL;
1508 bdev->bd_disk = NULL;
1509 if (bdev != bdev->bd_contains)
1510 victim = bdev->bd_contains;
1511 bdev->bd_contains = NULL;
1516 mutex_unlock(&bdev->bd_mutex);
1519 __blkdev_put(victim, mode, 1);
1523 int blkdev_put(struct block_device *bdev, fmode_t mode)
1525 mutex_lock(&bdev->bd_mutex);
1527 if (mode & FMODE_EXCL) {
1531 * Release a claim on the device. The holder fields
1532 * are protected with bdev_lock. bd_mutex is to
1533 * synchronize disk_holder unlinking.
1535 spin_lock(&bdev_lock);
1537 WARN_ON_ONCE(--bdev->bd_holders < 0);
1538 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1540 /* bd_contains might point to self, check in a separate step */
1541 if ((bdev_free = !bdev->bd_holders))
1542 bdev->bd_holder = NULL;
1543 if (!bdev->bd_contains->bd_holders)
1544 bdev->bd_contains->bd_holder = NULL;
1546 spin_unlock(&bdev_lock);
1549 * If this was the last claim, remove holder link and
1550 * unblock evpoll if it was a write holder.
1552 if (bdev_free && bdev->bd_write_holder) {
1553 disk_unblock_events(bdev->bd_disk);
1554 bdev->bd_write_holder = false;
1559 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1560 * event. This is to ensure detection of media removal commanded
1561 * from userland - e.g. eject(1).
1563 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1565 mutex_unlock(&bdev->bd_mutex);
1567 return __blkdev_put(bdev, mode, 0);
1569 EXPORT_SYMBOL(blkdev_put);
1571 static int blkdev_close(struct inode * inode, struct file * filp)
1573 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1575 return blkdev_put(bdev, filp->f_mode);
1578 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1580 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1581 fmode_t mode = file->f_mode;
1584 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1585 * to updated it before every ioctl.
1587 if (file->f_flags & O_NDELAY)
1588 mode |= FMODE_NDELAY;
1590 mode &= ~FMODE_NDELAY;
1592 return blkdev_ioctl(bdev, mode, cmd, arg);
1595 ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1596 unsigned long nr_segs, loff_t pos)
1599 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
1601 percpu_down_read(&bdev->bd_block_size_semaphore);
1603 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
1605 percpu_up_read(&bdev->bd_block_size_semaphore);
1609 EXPORT_SYMBOL_GPL(blkdev_aio_read);
1612 * Write data to the block device. Only intended for the block device itself
1613 * and the raw driver which basically is a fake block device.
1615 * Does not take i_mutex for the write and thus is not for general purpose
1618 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1619 unsigned long nr_segs, loff_t pos)
1621 struct file *file = iocb->ki_filp;
1622 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1623 struct blk_plug plug;
1626 BUG_ON(iocb->ki_pos != pos);
1628 blk_start_plug(&plug);
1630 percpu_down_read(&bdev->bd_block_size_semaphore);
1632 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1633 if (ret > 0 || ret == -EIOCBQUEUED) {
1636 err = generic_write_sync(file, pos, ret);
1637 if (err < 0 && ret > 0)
1641 percpu_up_read(&bdev->bd_block_size_semaphore);
1643 blk_finish_plug(&plug);
1647 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1649 static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1652 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1654 percpu_down_read(&bdev->bd_block_size_semaphore);
1656 ret = generic_file_mmap(file, vma);
1658 percpu_up_read(&bdev->bd_block_size_semaphore);
1664 * Try to release a page associated with block device when the system
1665 * is under memory pressure.
1667 static int blkdev_releasepage(struct page *page, gfp_t wait)
1669 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1671 if (super && super->s_op->bdev_try_to_free_page)
1672 return super->s_op->bdev_try_to_free_page(super, page, wait);
1674 return try_to_free_buffers(page);
1677 static const struct address_space_operations def_blk_aops = {
1678 .readpage = blkdev_readpage,
1679 .writepage = blkdev_writepage,
1680 .write_begin = blkdev_write_begin,
1681 .write_end = blkdev_write_end,
1682 .writepages = generic_writepages,
1683 .releasepage = blkdev_releasepage,
1684 .direct_IO = blkdev_direct_IO,
1687 const struct file_operations def_blk_fops = {
1688 .open = blkdev_open,
1689 .release = blkdev_close,
1690 .llseek = block_llseek,
1691 .read = do_sync_read,
1692 .write = do_sync_write,
1693 .aio_read = blkdev_aio_read,
1694 .aio_write = blkdev_aio_write,
1695 .mmap = blkdev_mmap,
1696 .fsync = blkdev_fsync,
1697 .unlocked_ioctl = block_ioctl,
1698 #ifdef CONFIG_COMPAT
1699 .compat_ioctl = compat_blkdev_ioctl,
1701 .splice_read = generic_file_splice_read,
1702 .splice_write = generic_file_splice_write,
1705 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1708 mm_segment_t old_fs = get_fs();
1710 res = blkdev_ioctl(bdev, 0, cmd, arg);
1715 EXPORT_SYMBOL(ioctl_by_bdev);
1718 * lookup_bdev - lookup a struct block_device by name
1719 * @pathname: special file representing the block device
1721 * Get a reference to the blockdevice at @pathname in the current
1722 * namespace if possible and return it. Return ERR_PTR(error)
1725 struct block_device *lookup_bdev(const char *pathname)
1727 struct block_device *bdev;
1728 struct inode *inode;
1732 if (!pathname || !*pathname)
1733 return ERR_PTR(-EINVAL);
1735 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1737 return ERR_PTR(error);
1739 inode = path.dentry->d_inode;
1741 if (!S_ISBLK(inode->i_mode))
1744 if (path.mnt->mnt_flags & MNT_NODEV)
1747 bdev = bd_acquire(inode);
1754 bdev = ERR_PTR(error);
1757 EXPORT_SYMBOL(lookup_bdev);
1759 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1761 struct super_block *sb = get_super(bdev);
1766 * no need to lock the super, get_super holds the
1767 * read mutex so the filesystem cannot go away
1768 * under us (->put_super runs with the write lock
1771 shrink_dcache_sb(sb);
1772 res = invalidate_inodes(sb, kill_dirty);
1775 invalidate_bdev(bdev);
1778 EXPORT_SYMBOL(__invalidate_device);
1780 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1782 struct inode *inode, *old_inode = NULL;
1784 spin_lock(&inode_sb_list_lock);
1785 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1786 struct address_space *mapping = inode->i_mapping;
1788 spin_lock(&inode->i_lock);
1789 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1790 mapping->nrpages == 0) {
1791 spin_unlock(&inode->i_lock);
1795 spin_unlock(&inode->i_lock);
1796 spin_unlock(&inode_sb_list_lock);
1798 * We hold a reference to 'inode' so it couldn't have been
1799 * removed from s_inodes list while we dropped the
1800 * inode_sb_list_lock. We cannot iput the inode now as we can
1801 * be holding the last reference and we cannot iput it under
1802 * inode_sb_list_lock. So we keep the reference and iput it
1808 func(I_BDEV(inode), arg);
1810 spin_lock(&inode_sb_list_lock);
1812 spin_unlock(&inode_sb_list_lock);