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/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <asm/uaccess.h>
35 struct block_device bdev;
36 struct inode vfs_inode;
39 static const struct address_space_operations def_blk_aops;
41 static inline struct bdev_inode *BDEV_I(struct inode *inode)
43 return container_of(inode, struct bdev_inode, vfs_inode);
46 inline struct block_device *I_BDEV(struct inode *inode)
48 return &BDEV_I(inode)->bdev;
50 EXPORT_SYMBOL(I_BDEV);
52 static void bdev_write_inode(struct inode *inode)
54 spin_lock(&inode->i_lock);
55 while (inode->i_state & I_DIRTY) {
56 spin_unlock(&inode->i_lock);
57 WARN_ON_ONCE(write_inode_now(inode, true));
58 spin_lock(&inode->i_lock);
60 spin_unlock(&inode->i_lock);
63 /* Kill _all_ buffers and pagecache , dirty or not.. */
64 void kill_bdev(struct block_device *bdev)
66 struct address_space *mapping = bdev->bd_inode->i_mapping;
68 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
72 truncate_inode_pages(mapping, 0);
74 EXPORT_SYMBOL(kill_bdev);
76 /* Invalidate clean unused buffers and pagecache. */
77 void invalidate_bdev(struct block_device *bdev)
79 struct address_space *mapping = bdev->bd_inode->i_mapping;
81 if (mapping->nrpages == 0)
85 lru_add_drain_all(); /* make sure all lru add caches are flushed */
86 invalidate_mapping_pages(mapping, 0, -1);
87 /* 99% of the time, we don't need to flush the cleancache on the bdev.
88 * But, for the strange corners, lets be cautious
90 cleancache_invalidate_inode(mapping);
92 EXPORT_SYMBOL(invalidate_bdev);
94 int set_blocksize(struct block_device *bdev, int size)
96 /* Size must be a power of two, and between 512 and PAGE_SIZE */
97 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
100 /* Size cannot be smaller than the size supported by the device */
101 if (size < bdev_logical_block_size(bdev))
104 /* Don't change the size if it is same as current */
105 if (bdev->bd_block_size != size) {
107 bdev->bd_block_size = size;
108 bdev->bd_inode->i_blkbits = blksize_bits(size);
114 EXPORT_SYMBOL(set_blocksize);
116 int sb_set_blocksize(struct super_block *sb, int size)
118 if (set_blocksize(sb->s_bdev, size))
120 /* If we get here, we know size is power of two
121 * and it's value is between 512 and PAGE_SIZE */
122 sb->s_blocksize = size;
123 sb->s_blocksize_bits = blksize_bits(size);
124 return sb->s_blocksize;
127 EXPORT_SYMBOL(sb_set_blocksize);
129 int sb_min_blocksize(struct super_block *sb, int size)
131 int minsize = bdev_logical_block_size(sb->s_bdev);
134 return sb_set_blocksize(sb, size);
137 EXPORT_SYMBOL(sb_min_blocksize);
140 blkdev_get_block(struct inode *inode, sector_t iblock,
141 struct buffer_head *bh, int create)
143 bh->b_bdev = I_BDEV(inode);
144 bh->b_blocknr = iblock;
145 set_buffer_mapped(bh);
150 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
152 struct file *file = iocb->ki_filp;
153 struct inode *inode = file->f_mapping->host;
155 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
156 blkdev_get_block, NULL, NULL,
160 int __sync_blockdev(struct block_device *bdev, int wait)
165 return filemap_flush(bdev->bd_inode->i_mapping);
166 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
170 * Write out and wait upon all the dirty data associated with a block
171 * device via its mapping. Does not take the superblock lock.
173 int sync_blockdev(struct block_device *bdev)
175 return __sync_blockdev(bdev, 1);
177 EXPORT_SYMBOL(sync_blockdev);
180 * Write out and wait upon all dirty data associated with this
181 * device. Filesystem data as well as the underlying block
182 * device. Takes the superblock lock.
184 int fsync_bdev(struct block_device *bdev)
186 struct super_block *sb = get_super(bdev);
188 int res = sync_filesystem(sb);
192 return sync_blockdev(bdev);
194 EXPORT_SYMBOL(fsync_bdev);
197 * freeze_bdev -- lock a filesystem and force it into a consistent state
198 * @bdev: blockdevice to lock
200 * If a superblock is found on this device, we take the s_umount semaphore
201 * on it to make sure nobody unmounts until the snapshot creation is done.
202 * The reference counter (bd_fsfreeze_count) guarantees that only the last
203 * unfreeze process can unfreeze the frozen filesystem actually when multiple
204 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
205 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
208 struct super_block *freeze_bdev(struct block_device *bdev)
210 struct super_block *sb;
213 mutex_lock(&bdev->bd_fsfreeze_mutex);
214 if (++bdev->bd_fsfreeze_count > 1) {
216 * We don't even need to grab a reference - the first call
217 * to freeze_bdev grab an active reference and only the last
218 * thaw_bdev drops it.
220 sb = get_super(bdev);
222 mutex_unlock(&bdev->bd_fsfreeze_mutex);
226 sb = get_active_super(bdev);
229 if (sb->s_op->freeze_super)
230 error = sb->s_op->freeze_super(sb);
232 error = freeze_super(sb);
234 deactivate_super(sb);
235 bdev->bd_fsfreeze_count--;
236 mutex_unlock(&bdev->bd_fsfreeze_mutex);
237 return ERR_PTR(error);
239 deactivate_super(sb);
242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
243 return sb; /* thaw_bdev releases s->s_umount */
245 EXPORT_SYMBOL(freeze_bdev);
248 * thaw_bdev -- unlock filesystem
249 * @bdev: blockdevice to unlock
250 * @sb: associated superblock
252 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
254 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
258 mutex_lock(&bdev->bd_fsfreeze_mutex);
259 if (!bdev->bd_fsfreeze_count)
263 if (--bdev->bd_fsfreeze_count > 0)
269 if (sb->s_op->thaw_super)
270 error = sb->s_op->thaw_super(sb);
272 error = thaw_super(sb);
274 bdev->bd_fsfreeze_count++;
275 mutex_unlock(&bdev->bd_fsfreeze_mutex);
279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 EXPORT_SYMBOL(thaw_bdev);
284 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
286 return block_write_full_page(page, blkdev_get_block, wbc);
289 static int blkdev_readpage(struct file * file, struct page * page)
291 return block_read_full_page(page, blkdev_get_block);
294 static int blkdev_readpages(struct file *file, struct address_space *mapping,
295 struct list_head *pages, unsigned nr_pages)
297 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
300 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
301 loff_t pos, unsigned len, unsigned flags,
302 struct page **pagep, void **fsdata)
304 return block_write_begin(mapping, pos, len, flags, pagep,
308 static int blkdev_write_end(struct file *file, struct address_space *mapping,
309 loff_t pos, unsigned len, unsigned copied,
310 struct page *page, void *fsdata)
313 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
316 page_cache_release(page);
323 * for a block special file file_inode(file)->i_size is zero
324 * so we compute the size by hand (just as in block_read/write above)
326 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
328 struct inode *bd_inode = file->f_mapping->host;
331 mutex_lock(&bd_inode->i_mutex);
332 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
333 mutex_unlock(&bd_inode->i_mutex);
337 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
339 struct inode *bd_inode = filp->f_mapping->host;
340 struct block_device *bdev = I_BDEV(bd_inode);
343 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
348 * There is no need to serialise calls to blkdev_issue_flush with
349 * i_mutex and doing so causes performance issues with concurrent
350 * O_SYNC writers to a block device.
352 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
353 if (error == -EOPNOTSUPP)
358 EXPORT_SYMBOL(blkdev_fsync);
361 * bdev_read_page() - Start reading a page from a block device
362 * @bdev: The device to read the page from
363 * @sector: The offset on the device to read the page to (need not be aligned)
364 * @page: The page to read
366 * On entry, the page should be locked. It will be unlocked when the page
367 * has been read. If the block driver implements rw_page synchronously,
368 * that will be true on exit from this function, but it need not be.
370 * Errors returned by this function are usually "soft", eg out of memory, or
371 * queue full; callers should try a different route to read this page rather
372 * than propagate an error back up the stack.
374 * Return: negative errno if an error occurs, 0 if submission was successful.
376 int bdev_read_page(struct block_device *bdev, sector_t sector,
379 const struct block_device_operations *ops = bdev->bd_disk->fops;
382 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
384 EXPORT_SYMBOL_GPL(bdev_read_page);
387 * bdev_write_page() - Start writing a page to a block device
388 * @bdev: The device to write the page to
389 * @sector: The offset on the device to write the page to (need not be aligned)
390 * @page: The page to write
391 * @wbc: The writeback_control for the write
393 * On entry, the page should be locked and not currently under writeback.
394 * On exit, if the write started successfully, the page will be unlocked and
395 * under writeback. If the write failed already (eg the driver failed to
396 * queue the page to the device), the page will still be locked. If the
397 * caller is a ->writepage implementation, it will need to unlock the page.
399 * Errors returned by this function are usually "soft", eg out of memory, or
400 * queue full; callers should try a different route to write this page rather
401 * than propagate an error back up the stack.
403 * Return: negative errno if an error occurs, 0 if submission was successful.
405 int bdev_write_page(struct block_device *bdev, sector_t sector,
406 struct page *page, struct writeback_control *wbc)
409 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
410 const struct block_device_operations *ops = bdev->bd_disk->fops;
413 set_page_writeback(page);
414 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
416 end_page_writeback(page);
421 EXPORT_SYMBOL_GPL(bdev_write_page);
424 * bdev_direct_access() - Get the address for directly-accessibly memory
425 * @bdev: The device containing the memory
426 * @sector: The offset within the device
427 * @addr: Where to put the address of the memory
428 * @pfn: The Page Frame Number for the memory
429 * @size: The number of bytes requested
431 * If a block device is made up of directly addressable memory, this function
432 * will tell the caller the PFN and the address of the memory. The address
433 * may be directly dereferenced within the kernel without the need to call
434 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
437 * Return: negative errno if an error occurs, otherwise the number of bytes
438 * accessible at this address.
440 long bdev_direct_access(struct block_device *bdev, sector_t sector,
441 void **addr, unsigned long *pfn, long size)
444 const struct block_device_operations *ops = bdev->bd_disk->fops;
448 if (!ops->direct_access)
450 if ((sector + DIV_ROUND_UP(size, 512)) >
451 part_nr_sects_read(bdev->bd_part))
453 sector += get_start_sect(bdev);
454 if (sector % (PAGE_SIZE / 512))
456 avail = ops->direct_access(bdev, sector, addr, pfn, size);
459 return min(avail, size);
461 EXPORT_SYMBOL_GPL(bdev_direct_access);
467 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
468 static struct kmem_cache * bdev_cachep __read_mostly;
470 static struct inode *bdev_alloc_inode(struct super_block *sb)
472 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
475 return &ei->vfs_inode;
478 static void bdev_i_callback(struct rcu_head *head)
480 struct inode *inode = container_of(head, struct inode, i_rcu);
481 struct bdev_inode *bdi = BDEV_I(inode);
483 kmem_cache_free(bdev_cachep, bdi);
486 static void bdev_destroy_inode(struct inode *inode)
488 call_rcu(&inode->i_rcu, bdev_i_callback);
491 static void init_once(void *foo)
493 struct bdev_inode *ei = (struct bdev_inode *) foo;
494 struct block_device *bdev = &ei->bdev;
496 memset(bdev, 0, sizeof(*bdev));
497 mutex_init(&bdev->bd_mutex);
498 INIT_LIST_HEAD(&bdev->bd_inodes);
499 INIT_LIST_HEAD(&bdev->bd_list);
501 INIT_LIST_HEAD(&bdev->bd_holder_disks);
503 inode_init_once(&ei->vfs_inode);
504 /* Initialize mutex for freeze. */
505 mutex_init(&bdev->bd_fsfreeze_mutex);
508 static inline void __bd_forget(struct inode *inode)
510 list_del_init(&inode->i_devices);
511 inode->i_bdev = NULL;
512 inode->i_mapping = &inode->i_data;
515 static void bdev_evict_inode(struct inode *inode)
517 struct block_device *bdev = &BDEV_I(inode)->bdev;
519 truncate_inode_pages_final(&inode->i_data);
520 invalidate_inode_buffers(inode); /* is it needed here? */
522 spin_lock(&bdev_lock);
523 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
524 __bd_forget(list_entry(p, struct inode, i_devices));
526 list_del_init(&bdev->bd_list);
527 spin_unlock(&bdev_lock);
530 static const struct super_operations bdev_sops = {
531 .statfs = simple_statfs,
532 .alloc_inode = bdev_alloc_inode,
533 .destroy_inode = bdev_destroy_inode,
534 .drop_inode = generic_delete_inode,
535 .evict_inode = bdev_evict_inode,
538 static struct dentry *bd_mount(struct file_system_type *fs_type,
539 int flags, const char *dev_name, void *data)
541 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
544 static struct file_system_type bd_type = {
547 .kill_sb = kill_anon_super,
550 struct super_block *blockdev_superblock __read_mostly;
551 EXPORT_SYMBOL_GPL(blockdev_superblock);
553 void __init bdev_cache_init(void)
556 static struct vfsmount *bd_mnt;
558 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
559 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
560 SLAB_MEM_SPREAD|SLAB_PANIC),
562 err = register_filesystem(&bd_type);
564 panic("Cannot register bdev pseudo-fs");
565 bd_mnt = kern_mount(&bd_type);
567 panic("Cannot create bdev pseudo-fs");
568 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
572 * Most likely _very_ bad one - but then it's hardly critical for small
573 * /dev and can be fixed when somebody will need really large one.
574 * Keep in mind that it will be fed through icache hash function too.
576 static inline unsigned long hash(dev_t dev)
578 return MAJOR(dev)+MINOR(dev);
581 static int bdev_test(struct inode *inode, void *data)
583 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
586 static int bdev_set(struct inode *inode, void *data)
588 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
592 static LIST_HEAD(all_bdevs);
594 struct block_device *bdget(dev_t dev)
596 struct block_device *bdev;
599 inode = iget5_locked(blockdev_superblock, hash(dev),
600 bdev_test, bdev_set, &dev);
605 bdev = &BDEV_I(inode)->bdev;
607 if (inode->i_state & I_NEW) {
608 bdev->bd_contains = NULL;
609 bdev->bd_super = NULL;
610 bdev->bd_inode = inode;
611 bdev->bd_block_size = (1 << inode->i_blkbits);
612 bdev->bd_part_count = 0;
613 bdev->bd_invalidated = 0;
614 inode->i_mode = S_IFBLK;
616 inode->i_bdev = bdev;
617 inode->i_data.a_ops = &def_blk_aops;
618 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
619 spin_lock(&bdev_lock);
620 list_add(&bdev->bd_list, &all_bdevs);
621 spin_unlock(&bdev_lock);
622 unlock_new_inode(inode);
627 EXPORT_SYMBOL(bdget);
630 * bdgrab -- Grab a reference to an already referenced block device
631 * @bdev: Block device to grab a reference to.
633 struct block_device *bdgrab(struct block_device *bdev)
635 ihold(bdev->bd_inode);
638 EXPORT_SYMBOL(bdgrab);
640 long nr_blockdev_pages(void)
642 struct block_device *bdev;
644 spin_lock(&bdev_lock);
645 list_for_each_entry(bdev, &all_bdevs, bd_list) {
646 ret += bdev->bd_inode->i_mapping->nrpages;
648 spin_unlock(&bdev_lock);
652 void bdput(struct block_device *bdev)
654 iput(bdev->bd_inode);
657 EXPORT_SYMBOL(bdput);
659 static struct block_device *bd_acquire(struct inode *inode)
661 struct block_device *bdev;
663 spin_lock(&bdev_lock);
664 bdev = inode->i_bdev;
666 ihold(bdev->bd_inode);
667 spin_unlock(&bdev_lock);
670 spin_unlock(&bdev_lock);
672 bdev = bdget(inode->i_rdev);
674 spin_lock(&bdev_lock);
675 if (!inode->i_bdev) {
677 * We take an additional reference to bd_inode,
678 * and it's released in clear_inode() of inode.
679 * So, we can access it via ->i_mapping always
682 ihold(bdev->bd_inode);
683 inode->i_bdev = bdev;
684 inode->i_mapping = bdev->bd_inode->i_mapping;
685 list_add(&inode->i_devices, &bdev->bd_inodes);
687 spin_unlock(&bdev_lock);
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 bdev->bd_invalidated = 0;
1076 mutex_unlock(&bdev->bd_mutex);
1080 EXPORT_SYMBOL(revalidate_disk);
1083 * This routine checks whether a removable media has been changed,
1084 * and invalidates all buffer-cache-entries in that case. This
1085 * is a relatively slow routine, so we have to try to minimize using
1086 * it. Thus it is called only upon a 'mount' or 'open'. This
1087 * is the best way of combining speed and utility, I think.
1088 * People changing diskettes in the middle of an operation deserve
1091 int check_disk_change(struct block_device *bdev)
1093 struct gendisk *disk = bdev->bd_disk;
1094 const struct block_device_operations *bdops = disk->fops;
1095 unsigned int events;
1097 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1098 DISK_EVENT_EJECT_REQUEST);
1099 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1102 flush_disk(bdev, true);
1103 if (bdops->revalidate_disk)
1104 bdops->revalidate_disk(bdev->bd_disk);
1108 EXPORT_SYMBOL(check_disk_change);
1110 void bd_set_size(struct block_device *bdev, loff_t size)
1112 unsigned bsize = bdev_logical_block_size(bdev);
1114 mutex_lock(&bdev->bd_inode->i_mutex);
1115 i_size_write(bdev->bd_inode, size);
1116 mutex_unlock(&bdev->bd_inode->i_mutex);
1117 while (bsize < PAGE_CACHE_SIZE) {
1122 bdev->bd_block_size = bsize;
1123 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1125 EXPORT_SYMBOL(bd_set_size);
1127 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1132 * mutex_lock(part->bd_mutex)
1133 * mutex_lock_nested(whole->bd_mutex, 1)
1136 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1138 struct gendisk *disk;
1139 struct module *owner;
1144 if (mode & FMODE_READ)
1146 if (mode & FMODE_WRITE)
1149 * hooks: /n/, see "layering violations".
1152 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1162 disk = get_gendisk(bdev->bd_dev, &partno);
1165 owner = disk->fops->owner;
1167 disk_block_events(disk);
1168 mutex_lock_nested(&bdev->bd_mutex, for_part);
1169 if (!bdev->bd_openers) {
1170 bdev->bd_disk = disk;
1171 bdev->bd_queue = disk->queue;
1172 bdev->bd_contains = bdev;
1175 bdev->bd_part = disk_get_part(disk, partno);
1180 if (disk->fops->open) {
1181 ret = disk->fops->open(bdev, mode);
1182 if (ret == -ERESTARTSYS) {
1183 /* Lost a race with 'disk' being
1184 * deleted, try again.
1187 disk_put_part(bdev->bd_part);
1188 bdev->bd_part = NULL;
1189 bdev->bd_disk = NULL;
1190 bdev->bd_queue = NULL;
1191 mutex_unlock(&bdev->bd_mutex);
1192 disk_unblock_events(disk);
1200 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1203 * If the device is invalidated, rescan partition
1204 * if open succeeded or failed with -ENOMEDIUM.
1205 * The latter is necessary to prevent ghost
1206 * partitions on a removed medium.
1208 if (bdev->bd_invalidated) {
1210 rescan_partitions(disk, bdev);
1211 else if (ret == -ENOMEDIUM)
1212 invalidate_partitions(disk, bdev);
1217 struct block_device *whole;
1218 whole = bdget_disk(disk, 0);
1223 ret = __blkdev_get(whole, mode, 1);
1226 bdev->bd_contains = whole;
1227 bdev->bd_part = disk_get_part(disk, partno);
1228 if (!(disk->flags & GENHD_FL_UP) ||
1229 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1233 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1236 if (bdev->bd_contains == bdev) {
1238 if (bdev->bd_disk->fops->open)
1239 ret = bdev->bd_disk->fops->open(bdev, mode);
1240 /* the same as first opener case, read comment there */
1241 if (bdev->bd_invalidated) {
1243 rescan_partitions(bdev->bd_disk, bdev);
1244 else if (ret == -ENOMEDIUM)
1245 invalidate_partitions(bdev->bd_disk, bdev);
1248 goto out_unlock_bdev;
1250 /* only one opener holds refs to the module and disk */
1256 bdev->bd_part_count++;
1257 mutex_unlock(&bdev->bd_mutex);
1258 disk_unblock_events(disk);
1262 disk_put_part(bdev->bd_part);
1263 bdev->bd_disk = NULL;
1264 bdev->bd_part = NULL;
1265 bdev->bd_queue = NULL;
1266 if (bdev != bdev->bd_contains)
1267 __blkdev_put(bdev->bd_contains, mode, 1);
1268 bdev->bd_contains = NULL;
1270 mutex_unlock(&bdev->bd_mutex);
1271 disk_unblock_events(disk);
1281 * blkdev_get - open a block device
1282 * @bdev: block_device to open
1283 * @mode: FMODE_* mask
1284 * @holder: exclusive holder identifier
1286 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1287 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1288 * @holder is invalid. Exclusive opens may nest for the same @holder.
1290 * On success, the reference count of @bdev is unchanged. On failure,
1297 * 0 on success, -errno on failure.
1299 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1301 struct block_device *whole = NULL;
1304 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1306 if ((mode & FMODE_EXCL) && holder) {
1307 whole = bd_start_claiming(bdev, holder);
1308 if (IS_ERR(whole)) {
1310 return PTR_ERR(whole);
1314 res = __blkdev_get(bdev, mode, 0);
1317 struct gendisk *disk = whole->bd_disk;
1319 /* finish claiming */
1320 mutex_lock(&bdev->bd_mutex);
1321 spin_lock(&bdev_lock);
1324 BUG_ON(!bd_may_claim(bdev, whole, holder));
1326 * Note that for a whole device bd_holders
1327 * will be incremented twice, and bd_holder
1328 * will be set to bd_may_claim before being
1331 whole->bd_holders++;
1332 whole->bd_holder = bd_may_claim;
1334 bdev->bd_holder = holder;
1337 /* tell others that we're done */
1338 BUG_ON(whole->bd_claiming != holder);
1339 whole->bd_claiming = NULL;
1340 wake_up_bit(&whole->bd_claiming, 0);
1342 spin_unlock(&bdev_lock);
1345 * Block event polling for write claims if requested. Any
1346 * write holder makes the write_holder state stick until
1347 * all are released. This is good enough and tracking
1348 * individual writeable reference is too fragile given the
1349 * way @mode is used in blkdev_get/put().
1351 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1352 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1353 bdev->bd_write_holder = true;
1354 disk_block_events(disk);
1357 mutex_unlock(&bdev->bd_mutex);
1363 EXPORT_SYMBOL(blkdev_get);
1366 * blkdev_get_by_path - open a block device by name
1367 * @path: path to the block device to open
1368 * @mode: FMODE_* mask
1369 * @holder: exclusive holder identifier
1371 * Open the blockdevice described by the device file at @path. @mode
1372 * and @holder are identical to blkdev_get().
1374 * On success, the returned block_device has reference count of one.
1380 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1382 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1385 struct block_device *bdev;
1388 bdev = lookup_bdev(path);
1392 err = blkdev_get(bdev, mode, holder);
1394 return ERR_PTR(err);
1396 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1397 blkdev_put(bdev, mode);
1398 return ERR_PTR(-EACCES);
1403 EXPORT_SYMBOL(blkdev_get_by_path);
1406 * blkdev_get_by_dev - open a block device by device number
1407 * @dev: device number of block device to open
1408 * @mode: FMODE_* mask
1409 * @holder: exclusive holder identifier
1411 * Open the blockdevice described by device number @dev. @mode and
1412 * @holder are identical to blkdev_get().
1414 * Use it ONLY if you really do not have anything better - i.e. when
1415 * you are behind a truly sucky interface and all you are given is a
1416 * device number. _Never_ to be used for internal purposes. If you
1417 * ever need it - reconsider your API.
1419 * On success, the returned block_device has reference count of one.
1425 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1427 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1429 struct block_device *bdev;
1434 return ERR_PTR(-ENOMEM);
1436 err = blkdev_get(bdev, mode, holder);
1438 return ERR_PTR(err);
1442 EXPORT_SYMBOL(blkdev_get_by_dev);
1444 static int blkdev_open(struct inode * inode, struct file * filp)
1446 struct block_device *bdev;
1449 * Preserve backwards compatibility and allow large file access
1450 * even if userspace doesn't ask for it explicitly. Some mkfs
1451 * binary needs it. We might want to drop this workaround
1452 * during an unstable branch.
1454 filp->f_flags |= O_LARGEFILE;
1456 if (filp->f_flags & O_NDELAY)
1457 filp->f_mode |= FMODE_NDELAY;
1458 if (filp->f_flags & O_EXCL)
1459 filp->f_mode |= FMODE_EXCL;
1460 if ((filp->f_flags & O_ACCMODE) == 3)
1461 filp->f_mode |= FMODE_WRITE_IOCTL;
1463 bdev = bd_acquire(inode);
1467 filp->f_mapping = bdev->bd_inode->i_mapping;
1469 return blkdev_get(bdev, filp->f_mode, filp);
1472 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1474 struct gendisk *disk = bdev->bd_disk;
1475 struct block_device *victim = NULL;
1477 mutex_lock_nested(&bdev->bd_mutex, for_part);
1479 bdev->bd_part_count--;
1481 if (!--bdev->bd_openers) {
1482 WARN_ON_ONCE(bdev->bd_holders);
1483 sync_blockdev(bdev);
1486 * ->release can cause the queue to disappear, so flush all
1487 * dirty data before.
1489 bdev_write_inode(bdev->bd_inode);
1491 if (bdev->bd_contains == bdev) {
1492 if (disk->fops->release)
1493 disk->fops->release(disk, mode);
1495 if (!bdev->bd_openers) {
1496 struct module *owner = disk->fops->owner;
1498 disk_put_part(bdev->bd_part);
1499 bdev->bd_part = NULL;
1500 bdev->bd_disk = NULL;
1501 if (bdev != bdev->bd_contains)
1502 victim = bdev->bd_contains;
1503 bdev->bd_contains = NULL;
1508 mutex_unlock(&bdev->bd_mutex);
1511 __blkdev_put(victim, mode, 1);
1514 void blkdev_put(struct block_device *bdev, fmode_t mode)
1516 mutex_lock(&bdev->bd_mutex);
1518 if (mode & FMODE_EXCL) {
1522 * Release a claim on the device. The holder fields
1523 * are protected with bdev_lock. bd_mutex is to
1524 * synchronize disk_holder unlinking.
1526 spin_lock(&bdev_lock);
1528 WARN_ON_ONCE(--bdev->bd_holders < 0);
1529 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1531 /* bd_contains might point to self, check in a separate step */
1532 if ((bdev_free = !bdev->bd_holders))
1533 bdev->bd_holder = NULL;
1534 if (!bdev->bd_contains->bd_holders)
1535 bdev->bd_contains->bd_holder = NULL;
1537 spin_unlock(&bdev_lock);
1540 * If this was the last claim, remove holder link and
1541 * unblock evpoll if it was a write holder.
1543 if (bdev_free && bdev->bd_write_holder) {
1544 disk_unblock_events(bdev->bd_disk);
1545 bdev->bd_write_holder = false;
1550 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1551 * event. This is to ensure detection of media removal commanded
1552 * from userland - e.g. eject(1).
1554 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1556 mutex_unlock(&bdev->bd_mutex);
1558 __blkdev_put(bdev, mode, 0);
1560 EXPORT_SYMBOL(blkdev_put);
1562 static int blkdev_close(struct inode * inode, struct file * filp)
1564 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1565 blkdev_put(bdev, filp->f_mode);
1569 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1571 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1572 fmode_t mode = file->f_mode;
1575 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1576 * to updated it before every ioctl.
1578 if (file->f_flags & O_NDELAY)
1579 mode |= FMODE_NDELAY;
1581 mode &= ~FMODE_NDELAY;
1583 return blkdev_ioctl(bdev, mode, cmd, arg);
1587 * Write data to the block device. Only intended for the block device itself
1588 * and the raw driver which basically is a fake block device.
1590 * Does not take i_mutex for the write and thus is not for general purpose
1593 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1595 struct file *file = iocb->ki_filp;
1596 struct inode *bd_inode = file->f_mapping->host;
1597 loff_t size = i_size_read(bd_inode);
1598 struct blk_plug plug;
1601 if (bdev_read_only(I_BDEV(bd_inode)))
1604 if (!iov_iter_count(from))
1607 if (iocb->ki_pos >= size)
1610 iov_iter_truncate(from, size - iocb->ki_pos);
1612 blk_start_plug(&plug);
1613 ret = __generic_file_write_iter(iocb, from);
1616 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1620 blk_finish_plug(&plug);
1623 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1625 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1627 struct file *file = iocb->ki_filp;
1628 struct inode *bd_inode = file->f_mapping->host;
1629 loff_t size = i_size_read(bd_inode);
1630 loff_t pos = iocb->ki_pos;
1636 iov_iter_truncate(to, size);
1637 return generic_file_read_iter(iocb, to);
1639 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1642 * Try to release a page associated with block device when the system
1643 * is under memory pressure.
1645 static int blkdev_releasepage(struct page *page, gfp_t wait)
1647 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1649 if (super && super->s_op->bdev_try_to_free_page)
1650 return super->s_op->bdev_try_to_free_page(super, page, wait);
1652 return try_to_free_buffers(page);
1655 static const struct address_space_operations def_blk_aops = {
1656 .readpage = blkdev_readpage,
1657 .readpages = blkdev_readpages,
1658 .writepage = blkdev_writepage,
1659 .write_begin = blkdev_write_begin,
1660 .write_end = blkdev_write_end,
1661 .writepages = generic_writepages,
1662 .releasepage = blkdev_releasepage,
1663 .direct_IO = blkdev_direct_IO,
1664 .is_dirty_writeback = buffer_check_dirty_writeback,
1667 const struct file_operations def_blk_fops = {
1668 .open = blkdev_open,
1669 .release = blkdev_close,
1670 .llseek = block_llseek,
1671 .read_iter = blkdev_read_iter,
1672 .write_iter = blkdev_write_iter,
1673 .mmap = generic_file_mmap,
1674 .fsync = blkdev_fsync,
1675 .unlocked_ioctl = block_ioctl,
1676 #ifdef CONFIG_COMPAT
1677 .compat_ioctl = compat_blkdev_ioctl,
1679 .splice_read = generic_file_splice_read,
1680 .splice_write = iter_file_splice_write,
1683 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1686 mm_segment_t old_fs = get_fs();
1688 res = blkdev_ioctl(bdev, 0, cmd, arg);
1693 EXPORT_SYMBOL(ioctl_by_bdev);
1696 * lookup_bdev - lookup a struct block_device by name
1697 * @pathname: special file representing the block device
1699 * Get a reference to the blockdevice at @pathname in the current
1700 * namespace if possible and return it. Return ERR_PTR(error)
1703 struct block_device *lookup_bdev(const char *pathname)
1705 struct block_device *bdev;
1706 struct inode *inode;
1710 if (!pathname || !*pathname)
1711 return ERR_PTR(-EINVAL);
1713 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1715 return ERR_PTR(error);
1717 inode = d_backing_inode(path.dentry);
1719 if (!S_ISBLK(inode->i_mode))
1722 if (path.mnt->mnt_flags & MNT_NODEV)
1725 bdev = bd_acquire(inode);
1732 bdev = ERR_PTR(error);
1735 EXPORT_SYMBOL(lookup_bdev);
1737 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1739 struct super_block *sb = get_super(bdev);
1744 * no need to lock the super, get_super holds the
1745 * read mutex so the filesystem cannot go away
1746 * under us (->put_super runs with the write lock
1749 shrink_dcache_sb(sb);
1750 res = invalidate_inodes(sb, kill_dirty);
1753 invalidate_bdev(bdev);
1756 EXPORT_SYMBOL(__invalidate_device);
1758 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1760 struct inode *inode, *old_inode = NULL;
1762 spin_lock(&inode_sb_list_lock);
1763 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1764 struct address_space *mapping = inode->i_mapping;
1766 spin_lock(&inode->i_lock);
1767 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1768 mapping->nrpages == 0) {
1769 spin_unlock(&inode->i_lock);
1773 spin_unlock(&inode->i_lock);
1774 spin_unlock(&inode_sb_list_lock);
1776 * We hold a reference to 'inode' so it couldn't have been
1777 * removed from s_inodes list while we dropped the
1778 * inode_sb_list_lock. We cannot iput the inode now as we can
1779 * be holding the last reference and we cannot iput it under
1780 * inode_sb_list_lock. So we keep the reference and iput it
1786 func(I_BDEV(inode), arg);
1788 spin_lock(&inode_sb_list_lock);
1790 spin_unlock(&inode_sb_list_lock);