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);
51 static void bdev_write_inode(struct inode *inode)
53 spin_lock(&inode->i_lock);
54 while (inode->i_state & I_DIRTY) {
55 spin_unlock(&inode->i_lock);
56 WARN_ON_ONCE(write_inode_now(inode, true));
57 spin_lock(&inode->i_lock);
59 spin_unlock(&inode->i_lock);
62 /* Kill _all_ buffers and pagecache , dirty or not.. */
63 void kill_bdev(struct block_device *bdev)
65 struct address_space *mapping = bdev->bd_inode->i_mapping;
67 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
71 truncate_inode_pages(mapping, 0);
73 EXPORT_SYMBOL(kill_bdev);
75 /* Invalidate clean unused buffers and pagecache. */
76 void invalidate_bdev(struct block_device *bdev)
78 struct address_space *mapping = bdev->bd_inode->i_mapping;
80 if (mapping->nrpages == 0)
84 lru_add_drain_all(); /* make sure all lru add caches are flushed */
85 invalidate_mapping_pages(mapping, 0, -1);
86 /* 99% of the time, we don't need to flush the cleancache on the bdev.
87 * But, for the strange corners, lets be cautious
89 cleancache_invalidate_inode(mapping);
91 EXPORT_SYMBOL(invalidate_bdev);
93 int set_blocksize(struct block_device *bdev, int size)
95 /* Size must be a power of two, and between 512 and PAGE_SIZE */
96 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
99 /* Size cannot be smaller than the size supported by the device */
100 if (size < bdev_logical_block_size(bdev))
103 /* Don't change the size if it is same as current */
104 if (bdev->bd_block_size != size) {
106 bdev->bd_block_size = size;
107 bdev->bd_inode->i_blkbits = blksize_bits(size);
113 EXPORT_SYMBOL(set_blocksize);
115 int sb_set_blocksize(struct super_block *sb, int size)
117 if (set_blocksize(sb->s_bdev, size))
119 /* If we get here, we know size is power of two
120 * and it's value is between 512 and PAGE_SIZE */
121 sb->s_blocksize = size;
122 sb->s_blocksize_bits = blksize_bits(size);
123 return sb->s_blocksize;
126 EXPORT_SYMBOL(sb_set_blocksize);
128 int sb_min_blocksize(struct super_block *sb, int size)
130 int minsize = bdev_logical_block_size(sb->s_bdev);
133 return sb_set_blocksize(sb, size);
136 EXPORT_SYMBOL(sb_min_blocksize);
139 blkdev_get_block(struct inode *inode, sector_t iblock,
140 struct buffer_head *bh, int create)
142 bh->b_bdev = I_BDEV(inode);
143 bh->b_blocknr = iblock;
144 set_buffer_mapped(bh);
149 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
151 struct file *file = iocb->ki_filp;
152 struct inode *inode = file->f_mapping->host;
155 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
156 NULL, DIO_SKIP_DIO_COUNT);
157 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
158 blkdev_get_block, NULL, NULL,
162 int __sync_blockdev(struct block_device *bdev, int wait)
167 return filemap_flush(bdev->bd_inode->i_mapping);
168 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
172 * Write out and wait upon all the dirty data associated with a block
173 * device via its mapping. Does not take the superblock lock.
175 int sync_blockdev(struct block_device *bdev)
177 return __sync_blockdev(bdev, 1);
179 EXPORT_SYMBOL(sync_blockdev);
182 * Write out and wait upon all dirty data associated with this
183 * device. Filesystem data as well as the underlying block
184 * device. Takes the superblock lock.
186 int fsync_bdev(struct block_device *bdev)
188 struct super_block *sb = get_super(bdev);
190 int res = sync_filesystem(sb);
194 return sync_blockdev(bdev);
196 EXPORT_SYMBOL(fsync_bdev);
199 * freeze_bdev -- lock a filesystem and force it into a consistent state
200 * @bdev: blockdevice to lock
202 * If a superblock is found on this device, we take the s_umount semaphore
203 * on it to make sure nobody unmounts until the snapshot creation is done.
204 * The reference counter (bd_fsfreeze_count) guarantees that only the last
205 * unfreeze process can unfreeze the frozen filesystem actually when multiple
206 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
207 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
210 struct super_block *freeze_bdev(struct block_device *bdev)
212 struct super_block *sb;
215 mutex_lock(&bdev->bd_fsfreeze_mutex);
216 if (++bdev->bd_fsfreeze_count > 1) {
218 * We don't even need to grab a reference - the first call
219 * to freeze_bdev grab an active reference and only the last
220 * thaw_bdev drops it.
222 sb = get_super(bdev);
224 mutex_unlock(&bdev->bd_fsfreeze_mutex);
228 sb = get_active_super(bdev);
231 if (sb->s_op->freeze_super)
232 error = sb->s_op->freeze_super(sb);
234 error = freeze_super(sb);
236 deactivate_super(sb);
237 bdev->bd_fsfreeze_count--;
238 mutex_unlock(&bdev->bd_fsfreeze_mutex);
239 return ERR_PTR(error);
241 deactivate_super(sb);
244 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 return sb; /* thaw_bdev releases s->s_umount */
247 EXPORT_SYMBOL(freeze_bdev);
250 * thaw_bdev -- unlock filesystem
251 * @bdev: blockdevice to unlock
252 * @sb: associated superblock
254 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
256 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
260 mutex_lock(&bdev->bd_fsfreeze_mutex);
261 if (!bdev->bd_fsfreeze_count)
265 if (--bdev->bd_fsfreeze_count > 0)
271 if (sb->s_op->thaw_super)
272 error = sb->s_op->thaw_super(sb);
274 error = thaw_super(sb);
276 bdev->bd_fsfreeze_count++;
277 mutex_unlock(&bdev->bd_fsfreeze_mutex);
281 mutex_unlock(&bdev->bd_fsfreeze_mutex);
284 EXPORT_SYMBOL(thaw_bdev);
286 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
288 return block_write_full_page(page, blkdev_get_block, wbc);
291 static int blkdev_readpage(struct file * file, struct page * page)
293 return block_read_full_page(page, blkdev_get_block);
296 static int blkdev_readpages(struct file *file, struct address_space *mapping,
297 struct list_head *pages, unsigned nr_pages)
299 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
302 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
303 loff_t pos, unsigned len, unsigned flags,
304 struct page **pagep, void **fsdata)
306 return block_write_begin(mapping, pos, len, flags, pagep,
310 static int blkdev_write_end(struct file *file, struct address_space *mapping,
311 loff_t pos, unsigned len, unsigned copied,
312 struct page *page, void *fsdata)
315 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
318 page_cache_release(page);
325 * for a block special file file_inode(file)->i_size is zero
326 * so we compute the size by hand (just as in block_read/write above)
328 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
330 struct inode *bd_inode = file->f_mapping->host;
333 mutex_lock(&bd_inode->i_mutex);
334 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
335 mutex_unlock(&bd_inode->i_mutex);
339 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
341 struct inode *bd_inode = filp->f_mapping->host;
342 struct block_device *bdev = I_BDEV(bd_inode);
345 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
350 * There is no need to serialise calls to blkdev_issue_flush with
351 * i_mutex and doing so causes performance issues with concurrent
352 * O_SYNC writers to a block device.
354 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
355 if (error == -EOPNOTSUPP)
360 EXPORT_SYMBOL(blkdev_fsync);
363 * bdev_read_page() - Start reading a page from a block device
364 * @bdev: The device to read the page from
365 * @sector: The offset on the device to read the page to (need not be aligned)
366 * @page: The page to read
368 * On entry, the page should be locked. It will be unlocked when the page
369 * has been read. If the block driver implements rw_page synchronously,
370 * that will be true on exit from this function, but it need not be.
372 * Errors returned by this function are usually "soft", eg out of memory, or
373 * queue full; callers should try a different route to read this page rather
374 * than propagate an error back up the stack.
376 * Return: negative errno if an error occurs, 0 if submission was successful.
378 int bdev_read_page(struct block_device *bdev, sector_t sector,
381 const struct block_device_operations *ops = bdev->bd_disk->fops;
384 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
386 EXPORT_SYMBOL_GPL(bdev_read_page);
389 * bdev_write_page() - Start writing a page to a block device
390 * @bdev: The device to write the page to
391 * @sector: The offset on the device to write the page to (need not be aligned)
392 * @page: The page to write
393 * @wbc: The writeback_control for the write
395 * On entry, the page should be locked and not currently under writeback.
396 * On exit, if the write started successfully, the page will be unlocked and
397 * under writeback. If the write failed already (eg the driver failed to
398 * queue the page to the device), the page will still be locked. If the
399 * caller is a ->writepage implementation, it will need to unlock the page.
401 * Errors returned by this function are usually "soft", eg out of memory, or
402 * queue full; callers should try a different route to write this page rather
403 * than propagate an error back up the stack.
405 * Return: negative errno if an error occurs, 0 if submission was successful.
407 int bdev_write_page(struct block_device *bdev, sector_t sector,
408 struct page *page, struct writeback_control *wbc)
411 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
412 const struct block_device_operations *ops = bdev->bd_disk->fops;
415 set_page_writeback(page);
416 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
418 end_page_writeback(page);
423 EXPORT_SYMBOL_GPL(bdev_write_page);
426 * bdev_direct_access() - Get the address for directly-accessibly memory
427 * @bdev: The device containing the memory
428 * @sector: The offset within the device
429 * @addr: Where to put the address of the memory
430 * @pfn: The Page Frame Number for the memory
431 * @size: The number of bytes requested
433 * If a block device is made up of directly addressable memory, this function
434 * will tell the caller the PFN and the address of the memory. The address
435 * may be directly dereferenced within the kernel without the need to call
436 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
439 * Return: negative errno if an error occurs, otherwise the number of bytes
440 * accessible at this address.
442 long bdev_direct_access(struct block_device *bdev, sector_t sector,
443 void **addr, unsigned long *pfn, long size)
446 const struct block_device_operations *ops = bdev->bd_disk->fops;
450 if (!ops->direct_access)
452 if ((sector + DIV_ROUND_UP(size, 512)) >
453 part_nr_sects_read(bdev->bd_part))
455 sector += get_start_sect(bdev);
456 if (sector % (PAGE_SIZE / 512))
458 avail = ops->direct_access(bdev, sector, addr, pfn, size);
461 return min(avail, size);
463 EXPORT_SYMBOL_GPL(bdev_direct_access);
469 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
470 static struct kmem_cache * bdev_cachep __read_mostly;
472 static struct inode *bdev_alloc_inode(struct super_block *sb)
474 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
477 return &ei->vfs_inode;
480 static void bdev_i_callback(struct rcu_head *head)
482 struct inode *inode = container_of(head, struct inode, i_rcu);
483 struct bdev_inode *bdi = BDEV_I(inode);
485 kmem_cache_free(bdev_cachep, bdi);
488 static void bdev_destroy_inode(struct inode *inode)
490 call_rcu(&inode->i_rcu, bdev_i_callback);
493 static void init_once(void *foo)
495 struct bdev_inode *ei = (struct bdev_inode *) foo;
496 struct block_device *bdev = &ei->bdev;
498 memset(bdev, 0, sizeof(*bdev));
499 mutex_init(&bdev->bd_mutex);
500 INIT_LIST_HEAD(&bdev->bd_inodes);
501 INIT_LIST_HEAD(&bdev->bd_list);
503 INIT_LIST_HEAD(&bdev->bd_holder_disks);
505 inode_init_once(&ei->vfs_inode);
506 /* Initialize mutex for freeze. */
507 mutex_init(&bdev->bd_fsfreeze_mutex);
510 static inline void __bd_forget(struct inode *inode)
512 list_del_init(&inode->i_devices);
513 inode->i_bdev = NULL;
514 inode->i_mapping = &inode->i_data;
517 static void bdev_evict_inode(struct inode *inode)
519 struct block_device *bdev = &BDEV_I(inode)->bdev;
521 truncate_inode_pages_final(&inode->i_data);
522 invalidate_inode_buffers(inode); /* is it needed here? */
524 spin_lock(&bdev_lock);
525 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
526 __bd_forget(list_entry(p, struct inode, i_devices));
528 list_del_init(&bdev->bd_list);
529 spin_unlock(&bdev_lock);
532 static const struct super_operations bdev_sops = {
533 .statfs = simple_statfs,
534 .alloc_inode = bdev_alloc_inode,
535 .destroy_inode = bdev_destroy_inode,
536 .drop_inode = generic_delete_inode,
537 .evict_inode = bdev_evict_inode,
540 static struct dentry *bd_mount(struct file_system_type *fs_type,
541 int flags, const char *dev_name, void *data)
543 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
546 static struct file_system_type bd_type = {
549 .kill_sb = kill_anon_super,
552 static struct super_block *blockdev_superblock __read_mostly;
554 void __init bdev_cache_init(void)
557 static struct vfsmount *bd_mnt;
559 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
560 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
561 SLAB_MEM_SPREAD|SLAB_PANIC),
563 err = register_filesystem(&bd_type);
565 panic("Cannot register bdev pseudo-fs");
566 bd_mnt = kern_mount(&bd_type);
568 panic("Cannot create bdev pseudo-fs");
569 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
573 * Most likely _very_ bad one - but then it's hardly critical for small
574 * /dev and can be fixed when somebody will need really large one.
575 * Keep in mind that it will be fed through icache hash function too.
577 static inline unsigned long hash(dev_t dev)
579 return MAJOR(dev)+MINOR(dev);
582 static int bdev_test(struct inode *inode, void *data)
584 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
587 static int bdev_set(struct inode *inode, void *data)
589 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
593 static LIST_HEAD(all_bdevs);
595 struct block_device *bdget(dev_t dev)
597 struct block_device *bdev;
600 inode = iget5_locked(blockdev_superblock, hash(dev),
601 bdev_test, bdev_set, &dev);
606 bdev = &BDEV_I(inode)->bdev;
608 if (inode->i_state & I_NEW) {
609 bdev->bd_contains = NULL;
610 bdev->bd_super = NULL;
611 bdev->bd_inode = inode;
612 bdev->bd_block_size = (1 << inode->i_blkbits);
613 bdev->bd_part_count = 0;
614 bdev->bd_invalidated = 0;
615 inode->i_mode = S_IFBLK;
617 inode->i_bdev = bdev;
618 inode->i_data.a_ops = &def_blk_aops;
619 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
620 spin_lock(&bdev_lock);
621 list_add(&bdev->bd_list, &all_bdevs);
622 spin_unlock(&bdev_lock);
623 unlock_new_inode(inode);
628 EXPORT_SYMBOL(bdget);
631 * bdgrab -- Grab a reference to an already referenced block device
632 * @bdev: Block device to grab a reference to.
634 struct block_device *bdgrab(struct block_device *bdev)
636 ihold(bdev->bd_inode);
639 EXPORT_SYMBOL(bdgrab);
641 long nr_blockdev_pages(void)
643 struct block_device *bdev;
645 spin_lock(&bdev_lock);
646 list_for_each_entry(bdev, &all_bdevs, bd_list) {
647 ret += bdev->bd_inode->i_mapping->nrpages;
649 spin_unlock(&bdev_lock);
653 void bdput(struct block_device *bdev)
655 iput(bdev->bd_inode);
658 EXPORT_SYMBOL(bdput);
660 static struct block_device *bd_acquire(struct inode *inode)
662 struct block_device *bdev;
664 spin_lock(&bdev_lock);
665 bdev = inode->i_bdev;
667 ihold(bdev->bd_inode);
668 spin_unlock(&bdev_lock);
671 spin_unlock(&bdev_lock);
673 bdev = bdget(inode->i_rdev);
675 spin_lock(&bdev_lock);
676 if (!inode->i_bdev) {
678 * We take an additional reference to bd_inode,
679 * and it's released in clear_inode() of inode.
680 * So, we can access it via ->i_mapping always
683 ihold(bdev->bd_inode);
684 inode->i_bdev = bdev;
685 inode->i_mapping = bdev->bd_inode->i_mapping;
686 list_add(&inode->i_devices, &bdev->bd_inodes);
688 spin_unlock(&bdev_lock);
693 int sb_is_blkdev_sb(struct super_block *sb)
695 return sb == blockdev_superblock;
698 /* Call when you free inode */
700 void bd_forget(struct inode *inode)
702 struct block_device *bdev = NULL;
704 spin_lock(&bdev_lock);
705 if (!sb_is_blkdev_sb(inode->i_sb))
706 bdev = inode->i_bdev;
708 spin_unlock(&bdev_lock);
711 iput(bdev->bd_inode);
715 * bd_may_claim - test whether a block device can be claimed
716 * @bdev: block device of interest
717 * @whole: whole block device containing @bdev, may equal @bdev
718 * @holder: holder trying to claim @bdev
720 * Test whether @bdev can be claimed by @holder.
723 * spin_lock(&bdev_lock).
726 * %true if @bdev can be claimed, %false otherwise.
728 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
731 if (bdev->bd_holder == holder)
732 return true; /* already a holder */
733 else if (bdev->bd_holder != NULL)
734 return false; /* held by someone else */
735 else if (bdev->bd_contains == bdev)
736 return true; /* is a whole device which isn't held */
738 else if (whole->bd_holder == bd_may_claim)
739 return true; /* is a partition of a device that is being partitioned */
740 else if (whole->bd_holder != NULL)
741 return false; /* is a partition of a held device */
743 return true; /* is a partition of an un-held device */
747 * bd_prepare_to_claim - prepare to claim a block device
748 * @bdev: block device of interest
749 * @whole: the whole device containing @bdev, may equal @bdev
750 * @holder: holder trying to claim @bdev
752 * Prepare to claim @bdev. This function fails if @bdev is already
753 * claimed by another holder and waits if another claiming is in
754 * progress. This function doesn't actually claim. On successful
755 * return, the caller has ownership of bd_claiming and bd_holder[s].
758 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
762 * 0 if @bdev can be claimed, -EBUSY otherwise.
764 static int bd_prepare_to_claim(struct block_device *bdev,
765 struct block_device *whole, void *holder)
768 /* if someone else claimed, fail */
769 if (!bd_may_claim(bdev, whole, holder))
772 /* if claiming is already in progress, wait for it to finish */
773 if (whole->bd_claiming) {
774 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
777 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
778 spin_unlock(&bdev_lock);
780 finish_wait(wq, &wait);
781 spin_lock(&bdev_lock);
790 * bd_start_claiming - start claiming a block device
791 * @bdev: block device of interest
792 * @holder: holder trying to claim @bdev
794 * @bdev is about to be opened exclusively. Check @bdev can be opened
795 * exclusively and mark that an exclusive open is in progress. Each
796 * successful call to this function must be matched with a call to
797 * either bd_finish_claiming() or bd_abort_claiming() (which do not
800 * This function is used to gain exclusive access to the block device
801 * without actually causing other exclusive open attempts to fail. It
802 * should be used when the open sequence itself requires exclusive
803 * access but may subsequently fail.
809 * Pointer to the block device containing @bdev on success, ERR_PTR()
812 static struct block_device *bd_start_claiming(struct block_device *bdev,
815 struct gendisk *disk;
816 struct block_device *whole;
822 * @bdev might not have been initialized properly yet, look up
823 * and grab the outer block device the hard way.
825 disk = get_gendisk(bdev->bd_dev, &partno);
827 return ERR_PTR(-ENXIO);
830 * Normally, @bdev should equal what's returned from bdget_disk()
831 * if partno is 0; however, some drivers (floppy) use multiple
832 * bdev's for the same physical device and @bdev may be one of the
833 * aliases. Keep @bdev if partno is 0. This means claimer
834 * tracking is broken for those devices but it has always been that
838 whole = bdget_disk(disk, 0);
840 whole = bdgrab(bdev);
842 module_put(disk->fops->owner);
845 return ERR_PTR(-ENOMEM);
847 /* prepare to claim, if successful, mark claiming in progress */
848 spin_lock(&bdev_lock);
850 err = bd_prepare_to_claim(bdev, whole, holder);
852 whole->bd_claiming = holder;
853 spin_unlock(&bdev_lock);
856 spin_unlock(&bdev_lock);
863 struct bd_holder_disk {
864 struct list_head list;
865 struct gendisk *disk;
869 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
870 struct gendisk *disk)
872 struct bd_holder_disk *holder;
874 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
875 if (holder->disk == disk)
880 static int add_symlink(struct kobject *from, struct kobject *to)
882 return sysfs_create_link(from, to, kobject_name(to));
885 static void del_symlink(struct kobject *from, struct kobject *to)
887 sysfs_remove_link(from, kobject_name(to));
891 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
892 * @bdev: the claimed slave bdev
893 * @disk: the holding disk
895 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
897 * This functions creates the following sysfs symlinks.
899 * - from "slaves" directory of the holder @disk to the claimed @bdev
900 * - from "holders" directory of the @bdev to the holder @disk
902 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
903 * passed to bd_link_disk_holder(), then:
905 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
906 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
908 * The caller must have claimed @bdev before calling this function and
909 * ensure that both @bdev and @disk are valid during the creation and
910 * lifetime of these symlinks.
916 * 0 on success, -errno on failure.
918 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
920 struct bd_holder_disk *holder;
923 mutex_lock(&bdev->bd_mutex);
925 WARN_ON_ONCE(!bdev->bd_holder);
927 /* FIXME: remove the following once add_disk() handles errors */
928 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
931 holder = bd_find_holder_disk(bdev, disk);
937 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
943 INIT_LIST_HEAD(&holder->list);
947 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
951 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
955 * bdev could be deleted beneath us which would implicitly destroy
956 * the holder directory. Hold on to it.
958 kobject_get(bdev->bd_part->holder_dir);
960 list_add(&holder->list, &bdev->bd_holder_disks);
964 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
968 mutex_unlock(&bdev->bd_mutex);
971 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
974 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
975 * @bdev: the calimed slave bdev
976 * @disk: the holding disk
978 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
983 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
985 struct bd_holder_disk *holder;
987 mutex_lock(&bdev->bd_mutex);
989 holder = bd_find_holder_disk(bdev, disk);
991 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
992 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
993 del_symlink(bdev->bd_part->holder_dir,
994 &disk_to_dev(disk)->kobj);
995 kobject_put(bdev->bd_part->holder_dir);
996 list_del_init(&holder->list);
1000 mutex_unlock(&bdev->bd_mutex);
1002 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1006 * flush_disk - invalidates all buffer-cache entries on a disk
1008 * @bdev: struct block device to be flushed
1009 * @kill_dirty: flag to guide handling of dirty inodes
1011 * Invalidates all buffer-cache entries on a disk. It should be called
1012 * when a disk has been changed -- either by a media change or online
1015 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1017 if (__invalidate_device(bdev, kill_dirty)) {
1018 char name[BDEVNAME_SIZE] = "";
1021 disk_name(bdev->bd_disk, 0, name);
1022 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1023 "resized disk %s\n", name);
1028 if (disk_part_scan_enabled(bdev->bd_disk))
1029 bdev->bd_invalidated = 1;
1033 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1034 * @disk: struct gendisk to check
1035 * @bdev: struct bdev to adjust.
1037 * This routine checks to see if the bdev size does not match the disk size
1038 * and adjusts it if it differs.
1040 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1042 loff_t disk_size, bdev_size;
1044 disk_size = (loff_t)get_capacity(disk) << 9;
1045 bdev_size = i_size_read(bdev->bd_inode);
1046 if (disk_size != bdev_size) {
1047 char name[BDEVNAME_SIZE];
1049 disk_name(disk, 0, name);
1051 "%s: detected capacity change from %lld to %lld\n",
1052 name, bdev_size, disk_size);
1053 i_size_write(bdev->bd_inode, disk_size);
1054 flush_disk(bdev, false);
1057 EXPORT_SYMBOL(check_disk_size_change);
1060 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1061 * @disk: struct gendisk to be revalidated
1063 * This routine is a wrapper for lower-level driver's revalidate_disk
1064 * call-backs. It is used to do common pre and post operations needed
1065 * for all revalidate_disk operations.
1067 int revalidate_disk(struct gendisk *disk)
1069 struct block_device *bdev;
1072 if (disk->fops->revalidate_disk)
1073 ret = disk->fops->revalidate_disk(disk);
1075 bdev = bdget_disk(disk, 0);
1079 mutex_lock(&bdev->bd_mutex);
1080 check_disk_size_change(disk, bdev);
1081 bdev->bd_invalidated = 0;
1082 mutex_unlock(&bdev->bd_mutex);
1086 EXPORT_SYMBOL(revalidate_disk);
1089 * This routine checks whether a removable media has been changed,
1090 * and invalidates all buffer-cache-entries in that case. This
1091 * is a relatively slow routine, so we have to try to minimize using
1092 * it. Thus it is called only upon a 'mount' or 'open'. This
1093 * is the best way of combining speed and utility, I think.
1094 * People changing diskettes in the middle of an operation deserve
1097 int check_disk_change(struct block_device *bdev)
1099 struct gendisk *disk = bdev->bd_disk;
1100 const struct block_device_operations *bdops = disk->fops;
1101 unsigned int events;
1103 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1104 DISK_EVENT_EJECT_REQUEST);
1105 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1108 flush_disk(bdev, true);
1109 if (bdops->revalidate_disk)
1110 bdops->revalidate_disk(bdev->bd_disk);
1114 EXPORT_SYMBOL(check_disk_change);
1116 void bd_set_size(struct block_device *bdev, loff_t size)
1118 unsigned bsize = bdev_logical_block_size(bdev);
1120 mutex_lock(&bdev->bd_inode->i_mutex);
1121 i_size_write(bdev->bd_inode, size);
1122 mutex_unlock(&bdev->bd_inode->i_mutex);
1123 while (bsize < PAGE_CACHE_SIZE) {
1128 bdev->bd_block_size = bsize;
1129 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1131 EXPORT_SYMBOL(bd_set_size);
1133 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1138 * mutex_lock(part->bd_mutex)
1139 * mutex_lock_nested(whole->bd_mutex, 1)
1142 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1144 struct gendisk *disk;
1145 struct module *owner;
1150 if (mode & FMODE_READ)
1152 if (mode & FMODE_WRITE)
1155 * hooks: /n/, see "layering violations".
1158 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1168 disk = get_gendisk(bdev->bd_dev, &partno);
1171 owner = disk->fops->owner;
1173 disk_block_events(disk);
1174 mutex_lock_nested(&bdev->bd_mutex, for_part);
1175 if (!bdev->bd_openers) {
1176 bdev->bd_disk = disk;
1177 bdev->bd_queue = disk->queue;
1178 bdev->bd_contains = bdev;
1179 bdev->bd_inode->i_flags = disk->fops->direct_access ? S_DAX : 0;
1182 bdev->bd_part = disk_get_part(disk, partno);
1187 if (disk->fops->open) {
1188 ret = disk->fops->open(bdev, mode);
1189 if (ret == -ERESTARTSYS) {
1190 /* Lost a race with 'disk' being
1191 * deleted, try again.
1194 disk_put_part(bdev->bd_part);
1195 bdev->bd_part = NULL;
1196 bdev->bd_disk = NULL;
1197 bdev->bd_queue = NULL;
1198 mutex_unlock(&bdev->bd_mutex);
1199 disk_unblock_events(disk);
1207 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1210 * If the device is invalidated, rescan partition
1211 * if open succeeded or failed with -ENOMEDIUM.
1212 * The latter is necessary to prevent ghost
1213 * partitions on a removed medium.
1215 if (bdev->bd_invalidated) {
1217 rescan_partitions(disk, bdev);
1218 else if (ret == -ENOMEDIUM)
1219 invalidate_partitions(disk, bdev);
1224 struct block_device *whole;
1225 whole = bdget_disk(disk, 0);
1230 ret = __blkdev_get(whole, mode, 1);
1233 bdev->bd_contains = whole;
1234 bdev->bd_part = disk_get_part(disk, partno);
1235 if (!(disk->flags & GENHD_FL_UP) ||
1236 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1240 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1243 if (bdev->bd_contains == bdev) {
1245 if (bdev->bd_disk->fops->open)
1246 ret = bdev->bd_disk->fops->open(bdev, mode);
1247 /* the same as first opener case, read comment there */
1248 if (bdev->bd_invalidated) {
1250 rescan_partitions(bdev->bd_disk, bdev);
1251 else if (ret == -ENOMEDIUM)
1252 invalidate_partitions(bdev->bd_disk, bdev);
1255 goto out_unlock_bdev;
1257 /* only one opener holds refs to the module and disk */
1263 bdev->bd_part_count++;
1264 mutex_unlock(&bdev->bd_mutex);
1265 disk_unblock_events(disk);
1269 disk_put_part(bdev->bd_part);
1270 bdev->bd_disk = NULL;
1271 bdev->bd_part = NULL;
1272 bdev->bd_queue = NULL;
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 void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1481 struct gendisk *disk = bdev->bd_disk;
1482 struct block_device *victim = NULL;
1484 mutex_lock_nested(&bdev->bd_mutex, for_part);
1486 bdev->bd_part_count--;
1488 if (!--bdev->bd_openers) {
1489 WARN_ON_ONCE(bdev->bd_holders);
1490 sync_blockdev(bdev);
1493 * ->release can cause the queue to disappear, so flush all
1494 * dirty data before.
1496 bdev_write_inode(bdev->bd_inode);
1498 if (bdev->bd_contains == bdev) {
1499 if (disk->fops->release)
1500 disk->fops->release(disk, mode);
1502 if (!bdev->bd_openers) {
1503 struct module *owner = disk->fops->owner;
1505 disk_put_part(bdev->bd_part);
1506 bdev->bd_part = NULL;
1507 bdev->bd_disk = NULL;
1508 if (bdev != bdev->bd_contains)
1509 victim = bdev->bd_contains;
1510 bdev->bd_contains = NULL;
1515 mutex_unlock(&bdev->bd_mutex);
1518 __blkdev_put(victim, mode, 1);
1521 void blkdev_put(struct block_device *bdev, fmode_t mode)
1523 mutex_lock(&bdev->bd_mutex);
1525 if (mode & FMODE_EXCL) {
1529 * Release a claim on the device. The holder fields
1530 * are protected with bdev_lock. bd_mutex is to
1531 * synchronize disk_holder unlinking.
1533 spin_lock(&bdev_lock);
1535 WARN_ON_ONCE(--bdev->bd_holders < 0);
1536 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1538 /* bd_contains might point to self, check in a separate step */
1539 if ((bdev_free = !bdev->bd_holders))
1540 bdev->bd_holder = NULL;
1541 if (!bdev->bd_contains->bd_holders)
1542 bdev->bd_contains->bd_holder = NULL;
1544 spin_unlock(&bdev_lock);
1547 * If this was the last claim, remove holder link and
1548 * unblock evpoll if it was a write holder.
1550 if (bdev_free && bdev->bd_write_holder) {
1551 disk_unblock_events(bdev->bd_disk);
1552 bdev->bd_write_holder = false;
1557 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1558 * event. This is to ensure detection of media removal commanded
1559 * from userland - e.g. eject(1).
1561 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1563 mutex_unlock(&bdev->bd_mutex);
1565 __blkdev_put(bdev, mode, 0);
1567 EXPORT_SYMBOL(blkdev_put);
1569 static int blkdev_close(struct inode * inode, struct file * filp)
1571 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1572 blkdev_put(bdev, filp->f_mode);
1576 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1578 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1579 fmode_t mode = file->f_mode;
1582 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1583 * to updated it before every ioctl.
1585 if (file->f_flags & O_NDELAY)
1586 mode |= FMODE_NDELAY;
1588 mode &= ~FMODE_NDELAY;
1590 return blkdev_ioctl(bdev, mode, cmd, arg);
1594 * Write data to the block device. Only intended for the block device itself
1595 * and the raw driver which basically is a fake block device.
1597 * Does not take i_mutex for the write and thus is not for general purpose
1600 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1602 struct file *file = iocb->ki_filp;
1603 struct inode *bd_inode = file->f_mapping->host;
1604 loff_t size = i_size_read(bd_inode);
1605 struct blk_plug plug;
1608 if (bdev_read_only(I_BDEV(bd_inode)))
1611 if (!iov_iter_count(from))
1614 if (iocb->ki_pos >= size)
1617 iov_iter_truncate(from, size - iocb->ki_pos);
1619 blk_start_plug(&plug);
1620 ret = __generic_file_write_iter(iocb, from);
1623 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1627 blk_finish_plug(&plug);
1630 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1632 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1634 struct file *file = iocb->ki_filp;
1635 struct inode *bd_inode = file->f_mapping->host;
1636 loff_t size = i_size_read(bd_inode);
1637 loff_t pos = iocb->ki_pos;
1643 iov_iter_truncate(to, size);
1644 return generic_file_read_iter(iocb, to);
1646 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1649 * Try to release a page associated with block device when the system
1650 * is under memory pressure.
1652 static int blkdev_releasepage(struct page *page, gfp_t wait)
1654 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1656 if (super && super->s_op->bdev_try_to_free_page)
1657 return super->s_op->bdev_try_to_free_page(super, page, wait);
1659 return try_to_free_buffers(page);
1662 static const struct address_space_operations def_blk_aops = {
1663 .readpage = blkdev_readpage,
1664 .readpages = blkdev_readpages,
1665 .writepage = blkdev_writepage,
1666 .write_begin = blkdev_write_begin,
1667 .write_end = blkdev_write_end,
1668 .writepages = generic_writepages,
1669 .releasepage = blkdev_releasepage,
1670 .direct_IO = blkdev_direct_IO,
1671 .is_dirty_writeback = buffer_check_dirty_writeback,
1674 const struct file_operations def_blk_fops = {
1675 .open = blkdev_open,
1676 .release = blkdev_close,
1677 .llseek = block_llseek,
1678 .read_iter = blkdev_read_iter,
1679 .write_iter = blkdev_write_iter,
1680 .mmap = generic_file_mmap,
1681 .fsync = blkdev_fsync,
1682 .unlocked_ioctl = block_ioctl,
1683 #ifdef CONFIG_COMPAT
1684 .compat_ioctl = compat_blkdev_ioctl,
1686 .splice_read = generic_file_splice_read,
1687 .splice_write = iter_file_splice_write,
1690 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1693 mm_segment_t old_fs = get_fs();
1695 res = blkdev_ioctl(bdev, 0, cmd, arg);
1700 EXPORT_SYMBOL(ioctl_by_bdev);
1703 * lookup_bdev - lookup a struct block_device by name
1704 * @pathname: special file representing the block device
1706 * Get a reference to the blockdevice at @pathname in the current
1707 * namespace if possible and return it. Return ERR_PTR(error)
1710 struct block_device *lookup_bdev(const char *pathname)
1712 struct block_device *bdev;
1713 struct inode *inode;
1717 if (!pathname || !*pathname)
1718 return ERR_PTR(-EINVAL);
1720 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1722 return ERR_PTR(error);
1724 inode = d_backing_inode(path.dentry);
1726 if (!S_ISBLK(inode->i_mode))
1729 if (path.mnt->mnt_flags & MNT_NODEV)
1732 bdev = bd_acquire(inode);
1739 bdev = ERR_PTR(error);
1742 EXPORT_SYMBOL(lookup_bdev);
1744 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1746 struct super_block *sb = get_super(bdev);
1751 * no need to lock the super, get_super holds the
1752 * read mutex so the filesystem cannot go away
1753 * under us (->put_super runs with the write lock
1756 shrink_dcache_sb(sb);
1757 res = invalidate_inodes(sb, kill_dirty);
1760 invalidate_bdev(bdev);
1763 EXPORT_SYMBOL(__invalidate_device);
1765 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1767 struct inode *inode, *old_inode = NULL;
1769 spin_lock(&inode_sb_list_lock);
1770 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1771 struct address_space *mapping = inode->i_mapping;
1773 spin_lock(&inode->i_lock);
1774 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1775 mapping->nrpages == 0) {
1776 spin_unlock(&inode->i_lock);
1780 spin_unlock(&inode->i_lock);
1781 spin_unlock(&inode_sb_list_lock);
1783 * We hold a reference to 'inode' so it couldn't have been
1784 * removed from s_inodes list while we dropped the
1785 * inode_sb_list_lock. We cannot iput the inode now as we can
1786 * be holding the last reference and we cannot iput it under
1787 * inode_sb_list_lock. So we keep the reference and iput it
1793 func(I_BDEV(inode), arg);
1795 spin_lock(&inode_sb_list_lock);
1797 spin_unlock(&inode_sb_list_lock);