gpio: rcar: Fix runtime PM imbalance on error
[linux/fpc-iii.git] / fs / block_dev.c
blob52b6f646cdbd0af27d827d9e4d983ec1455c87f0
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/fs/block_dev.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
7 */
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/dax.h>
23 #include <linux/buffer_head.h>
24 #include <linux/swap.h>
25 #include <linux/pagevec.h>
26 #include <linux/writeback.h>
27 #include <linux/mpage.h>
28 #include <linux/mount.h>
29 #include <linux/pseudo_fs.h>
30 #include <linux/uio.h>
31 #include <linux/namei.h>
32 #include <linux/log2.h>
33 #include <linux/cleancache.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
37 #include <linux/suspend.h>
38 #include "internal.h"
40 struct bdev_inode {
41 struct block_device bdev;
42 struct inode vfs_inode;
45 static const struct address_space_operations def_blk_aops;
47 static inline struct bdev_inode *BDEV_I(struct inode *inode)
49 return container_of(inode, struct bdev_inode, vfs_inode);
52 struct block_device *I_BDEV(struct inode *inode)
54 return &BDEV_I(inode)->bdev;
56 EXPORT_SYMBOL(I_BDEV);
58 static void bdev_write_inode(struct block_device *bdev)
60 struct inode *inode = bdev->bd_inode;
61 int ret;
63 spin_lock(&inode->i_lock);
64 while (inode->i_state & I_DIRTY) {
65 spin_unlock(&inode->i_lock);
66 ret = write_inode_now(inode, true);
67 if (ret) {
68 char name[BDEVNAME_SIZE];
69 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
70 "for block device %s (err=%d).\n",
71 bdevname(bdev, name), ret);
73 spin_lock(&inode->i_lock);
75 spin_unlock(&inode->i_lock);
78 /* Kill _all_ buffers and pagecache , dirty or not.. */
79 void kill_bdev(struct block_device *bdev)
81 struct address_space *mapping = bdev->bd_inode->i_mapping;
83 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
84 return;
86 invalidate_bh_lrus();
87 truncate_inode_pages(mapping, 0);
89 EXPORT_SYMBOL(kill_bdev);
91 /* Invalidate clean unused buffers and pagecache. */
92 void invalidate_bdev(struct block_device *bdev)
94 struct address_space *mapping = bdev->bd_inode->i_mapping;
96 if (mapping->nrpages) {
97 invalidate_bh_lrus();
98 lru_add_drain_all(); /* make sure all lru add caches are flushed */
99 invalidate_mapping_pages(mapping, 0, -1);
101 /* 99% of the time, we don't need to flush the cleancache on the bdev.
102 * But, for the strange corners, lets be cautious
104 cleancache_invalidate_inode(mapping);
106 EXPORT_SYMBOL(invalidate_bdev);
108 static void set_init_blocksize(struct block_device *bdev)
110 unsigned bsize = bdev_logical_block_size(bdev);
111 loff_t size = i_size_read(bdev->bd_inode);
113 while (bsize < PAGE_SIZE) {
114 if (size & bsize)
115 break;
116 bsize <<= 1;
118 bdev->bd_block_size = bsize;
119 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
122 int set_blocksize(struct block_device *bdev, int size)
124 /* Size must be a power of two, and between 512 and PAGE_SIZE */
125 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
126 return -EINVAL;
128 /* Size cannot be smaller than the size supported by the device */
129 if (size < bdev_logical_block_size(bdev))
130 return -EINVAL;
132 /* Don't change the size if it is same as current */
133 if (bdev->bd_block_size != size) {
134 sync_blockdev(bdev);
135 bdev->bd_block_size = size;
136 bdev->bd_inode->i_blkbits = blksize_bits(size);
137 kill_bdev(bdev);
139 return 0;
142 EXPORT_SYMBOL(set_blocksize);
144 int sb_set_blocksize(struct super_block *sb, int size)
146 if (set_blocksize(sb->s_bdev, size))
147 return 0;
148 /* If we get here, we know size is power of two
149 * and it's value is between 512 and PAGE_SIZE */
150 sb->s_blocksize = size;
151 sb->s_blocksize_bits = blksize_bits(size);
152 return sb->s_blocksize;
155 EXPORT_SYMBOL(sb_set_blocksize);
157 int sb_min_blocksize(struct super_block *sb, int size)
159 int minsize = bdev_logical_block_size(sb->s_bdev);
160 if (size < minsize)
161 size = minsize;
162 return sb_set_blocksize(sb, size);
165 EXPORT_SYMBOL(sb_min_blocksize);
167 static int
168 blkdev_get_block(struct inode *inode, sector_t iblock,
169 struct buffer_head *bh, int create)
171 bh->b_bdev = I_BDEV(inode);
172 bh->b_blocknr = iblock;
173 set_buffer_mapped(bh);
174 return 0;
177 static struct inode *bdev_file_inode(struct file *file)
179 return file->f_mapping->host;
182 static unsigned int dio_bio_write_op(struct kiocb *iocb)
184 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
186 /* avoid the need for a I/O completion work item */
187 if (iocb->ki_flags & IOCB_DSYNC)
188 op |= REQ_FUA;
189 return op;
192 #define DIO_INLINE_BIO_VECS 4
194 static void blkdev_bio_end_io_simple(struct bio *bio)
196 struct task_struct *waiter = bio->bi_private;
198 WRITE_ONCE(bio->bi_private, NULL);
199 blk_wake_io_task(waiter);
202 static ssize_t
203 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
204 int nr_pages)
206 struct file *file = iocb->ki_filp;
207 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
208 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
209 loff_t pos = iocb->ki_pos;
210 bool should_dirty = false;
211 struct bio bio;
212 ssize_t ret;
213 blk_qc_t qc;
215 if ((pos | iov_iter_alignment(iter)) &
216 (bdev_logical_block_size(bdev) - 1))
217 return -EINVAL;
219 if (nr_pages <= DIO_INLINE_BIO_VECS)
220 vecs = inline_vecs;
221 else {
222 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
223 GFP_KERNEL);
224 if (!vecs)
225 return -ENOMEM;
228 bio_init(&bio, vecs, nr_pages);
229 bio_set_dev(&bio, bdev);
230 bio.bi_iter.bi_sector = pos >> 9;
231 bio.bi_write_hint = iocb->ki_hint;
232 bio.bi_private = current;
233 bio.bi_end_io = blkdev_bio_end_io_simple;
234 bio.bi_ioprio = iocb->ki_ioprio;
236 ret = bio_iov_iter_get_pages(&bio, iter);
237 if (unlikely(ret))
238 goto out;
239 ret = bio.bi_iter.bi_size;
241 if (iov_iter_rw(iter) == READ) {
242 bio.bi_opf = REQ_OP_READ;
243 if (iter_is_iovec(iter))
244 should_dirty = true;
245 } else {
246 bio.bi_opf = dio_bio_write_op(iocb);
247 task_io_account_write(ret);
249 if (iocb->ki_flags & IOCB_HIPRI)
250 bio_set_polled(&bio, iocb);
252 qc = submit_bio(&bio);
253 for (;;) {
254 set_current_state(TASK_UNINTERRUPTIBLE);
255 if (!READ_ONCE(bio.bi_private))
256 break;
257 if (!(iocb->ki_flags & IOCB_HIPRI) ||
258 !blk_poll(bdev_get_queue(bdev), qc, true))
259 io_schedule();
261 __set_current_state(TASK_RUNNING);
263 bio_release_pages(&bio, should_dirty);
264 if (unlikely(bio.bi_status))
265 ret = blk_status_to_errno(bio.bi_status);
267 out:
268 if (vecs != inline_vecs)
269 kfree(vecs);
271 bio_uninit(&bio);
273 return ret;
276 struct blkdev_dio {
277 union {
278 struct kiocb *iocb;
279 struct task_struct *waiter;
281 size_t size;
282 atomic_t ref;
283 bool multi_bio : 1;
284 bool should_dirty : 1;
285 bool is_sync : 1;
286 struct bio bio;
289 static struct bio_set blkdev_dio_pool;
291 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
293 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
294 struct request_queue *q = bdev_get_queue(bdev);
296 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
299 static void blkdev_bio_end_io(struct bio *bio)
301 struct blkdev_dio *dio = bio->bi_private;
302 bool should_dirty = dio->should_dirty;
304 if (bio->bi_status && !dio->bio.bi_status)
305 dio->bio.bi_status = bio->bi_status;
307 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
308 if (!dio->is_sync) {
309 struct kiocb *iocb = dio->iocb;
310 ssize_t ret;
312 if (likely(!dio->bio.bi_status)) {
313 ret = dio->size;
314 iocb->ki_pos += ret;
315 } else {
316 ret = blk_status_to_errno(dio->bio.bi_status);
319 dio->iocb->ki_complete(iocb, ret, 0);
320 if (dio->multi_bio)
321 bio_put(&dio->bio);
322 } else {
323 struct task_struct *waiter = dio->waiter;
325 WRITE_ONCE(dio->waiter, NULL);
326 blk_wake_io_task(waiter);
330 if (should_dirty) {
331 bio_check_pages_dirty(bio);
332 } else {
333 bio_release_pages(bio, false);
334 bio_put(bio);
338 static ssize_t
339 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
341 struct file *file = iocb->ki_filp;
342 struct inode *inode = bdev_file_inode(file);
343 struct block_device *bdev = I_BDEV(inode);
344 struct blk_plug plug;
345 struct blkdev_dio *dio;
346 struct bio *bio;
347 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
348 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
349 loff_t pos = iocb->ki_pos;
350 blk_qc_t qc = BLK_QC_T_NONE;
351 int ret = 0;
353 if ((pos | iov_iter_alignment(iter)) &
354 (bdev_logical_block_size(bdev) - 1))
355 return -EINVAL;
357 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
359 dio = container_of(bio, struct blkdev_dio, bio);
360 dio->is_sync = is_sync = is_sync_kiocb(iocb);
361 if (dio->is_sync) {
362 dio->waiter = current;
363 bio_get(bio);
364 } else {
365 dio->iocb = iocb;
368 dio->size = 0;
369 dio->multi_bio = false;
370 dio->should_dirty = is_read && iter_is_iovec(iter);
373 * Don't plug for HIPRI/polled IO, as those should go straight
374 * to issue
376 if (!is_poll)
377 blk_start_plug(&plug);
379 for (;;) {
380 bio_set_dev(bio, bdev);
381 bio->bi_iter.bi_sector = pos >> 9;
382 bio->bi_write_hint = iocb->ki_hint;
383 bio->bi_private = dio;
384 bio->bi_end_io = blkdev_bio_end_io;
385 bio->bi_ioprio = iocb->ki_ioprio;
387 ret = bio_iov_iter_get_pages(bio, iter);
388 if (unlikely(ret)) {
389 bio->bi_status = BLK_STS_IOERR;
390 bio_endio(bio);
391 break;
394 if (is_read) {
395 bio->bi_opf = REQ_OP_READ;
396 if (dio->should_dirty)
397 bio_set_pages_dirty(bio);
398 } else {
399 bio->bi_opf = dio_bio_write_op(iocb);
400 task_io_account_write(bio->bi_iter.bi_size);
403 dio->size += bio->bi_iter.bi_size;
404 pos += bio->bi_iter.bi_size;
406 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
407 if (!nr_pages) {
408 bool polled = false;
410 if (iocb->ki_flags & IOCB_HIPRI) {
411 bio_set_polled(bio, iocb);
412 polled = true;
415 qc = submit_bio(bio);
417 if (polled)
418 WRITE_ONCE(iocb->ki_cookie, qc);
419 break;
422 if (!dio->multi_bio) {
424 * AIO needs an extra reference to ensure the dio
425 * structure which is embedded into the first bio
426 * stays around.
428 if (!is_sync)
429 bio_get(bio);
430 dio->multi_bio = true;
431 atomic_set(&dio->ref, 2);
432 } else {
433 atomic_inc(&dio->ref);
436 submit_bio(bio);
437 bio = bio_alloc(GFP_KERNEL, nr_pages);
440 if (!is_poll)
441 blk_finish_plug(&plug);
443 if (!is_sync)
444 return -EIOCBQUEUED;
446 for (;;) {
447 set_current_state(TASK_UNINTERRUPTIBLE);
448 if (!READ_ONCE(dio->waiter))
449 break;
451 if (!(iocb->ki_flags & IOCB_HIPRI) ||
452 !blk_poll(bdev_get_queue(bdev), qc, true))
453 io_schedule();
455 __set_current_state(TASK_RUNNING);
457 if (!ret)
458 ret = blk_status_to_errno(dio->bio.bi_status);
459 if (likely(!ret))
460 ret = dio->size;
462 bio_put(&dio->bio);
463 return ret;
466 static ssize_t
467 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
469 int nr_pages;
471 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
472 if (!nr_pages)
473 return 0;
474 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
475 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
477 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
480 static __init int blkdev_init(void)
482 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
484 module_init(blkdev_init);
486 int __sync_blockdev(struct block_device *bdev, int wait)
488 if (!bdev)
489 return 0;
490 if (!wait)
491 return filemap_flush(bdev->bd_inode->i_mapping);
492 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
496 * Write out and wait upon all the dirty data associated with a block
497 * device via its mapping. Does not take the superblock lock.
499 int sync_blockdev(struct block_device *bdev)
501 return __sync_blockdev(bdev, 1);
503 EXPORT_SYMBOL(sync_blockdev);
506 * Write out and wait upon all dirty data associated with this
507 * device. Filesystem data as well as the underlying block
508 * device. Takes the superblock lock.
510 int fsync_bdev(struct block_device *bdev)
512 struct super_block *sb = get_super(bdev);
513 if (sb) {
514 int res = sync_filesystem(sb);
515 drop_super(sb);
516 return res;
518 return sync_blockdev(bdev);
520 EXPORT_SYMBOL(fsync_bdev);
523 * freeze_bdev -- lock a filesystem and force it into a consistent state
524 * @bdev: blockdevice to lock
526 * If a superblock is found on this device, we take the s_umount semaphore
527 * on it to make sure nobody unmounts until the snapshot creation is done.
528 * The reference counter (bd_fsfreeze_count) guarantees that only the last
529 * unfreeze process can unfreeze the frozen filesystem actually when multiple
530 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
531 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
532 * actually.
534 struct super_block *freeze_bdev(struct block_device *bdev)
536 struct super_block *sb;
537 int error = 0;
539 mutex_lock(&bdev->bd_fsfreeze_mutex);
540 if (++bdev->bd_fsfreeze_count > 1) {
542 * We don't even need to grab a reference - the first call
543 * to freeze_bdev grab an active reference and only the last
544 * thaw_bdev drops it.
546 sb = get_super(bdev);
547 if (sb)
548 drop_super(sb);
549 mutex_unlock(&bdev->bd_fsfreeze_mutex);
550 return sb;
553 sb = get_active_super(bdev);
554 if (!sb)
555 goto out;
556 if (sb->s_op->freeze_super)
557 error = sb->s_op->freeze_super(sb);
558 else
559 error = freeze_super(sb);
560 if (error) {
561 deactivate_super(sb);
562 bdev->bd_fsfreeze_count--;
563 mutex_unlock(&bdev->bd_fsfreeze_mutex);
564 return ERR_PTR(error);
566 deactivate_super(sb);
567 out:
568 sync_blockdev(bdev);
569 mutex_unlock(&bdev->bd_fsfreeze_mutex);
570 return sb; /* thaw_bdev releases s->s_umount */
572 EXPORT_SYMBOL(freeze_bdev);
575 * thaw_bdev -- unlock filesystem
576 * @bdev: blockdevice to unlock
577 * @sb: associated superblock
579 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
581 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
583 int error = -EINVAL;
585 mutex_lock(&bdev->bd_fsfreeze_mutex);
586 if (!bdev->bd_fsfreeze_count)
587 goto out;
589 error = 0;
590 if (--bdev->bd_fsfreeze_count > 0)
591 goto out;
593 if (!sb)
594 goto out;
596 if (sb->s_op->thaw_super)
597 error = sb->s_op->thaw_super(sb);
598 else
599 error = thaw_super(sb);
600 if (error)
601 bdev->bd_fsfreeze_count++;
602 out:
603 mutex_unlock(&bdev->bd_fsfreeze_mutex);
604 return error;
606 EXPORT_SYMBOL(thaw_bdev);
608 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
610 return block_write_full_page(page, blkdev_get_block, wbc);
613 static int blkdev_readpage(struct file * file, struct page * page)
615 return block_read_full_page(page, blkdev_get_block);
618 static int blkdev_readpages(struct file *file, struct address_space *mapping,
619 struct list_head *pages, unsigned nr_pages)
621 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
624 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
625 loff_t pos, unsigned len, unsigned flags,
626 struct page **pagep, void **fsdata)
628 return block_write_begin(mapping, pos, len, flags, pagep,
629 blkdev_get_block);
632 static int blkdev_write_end(struct file *file, struct address_space *mapping,
633 loff_t pos, unsigned len, unsigned copied,
634 struct page *page, void *fsdata)
636 int ret;
637 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
639 unlock_page(page);
640 put_page(page);
642 return ret;
646 * private llseek:
647 * for a block special file file_inode(file)->i_size is zero
648 * so we compute the size by hand (just as in block_read/write above)
650 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
652 struct inode *bd_inode = bdev_file_inode(file);
653 loff_t retval;
655 inode_lock(bd_inode);
656 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
657 inode_unlock(bd_inode);
658 return retval;
661 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
663 struct inode *bd_inode = bdev_file_inode(filp);
664 struct block_device *bdev = I_BDEV(bd_inode);
665 int error;
667 error = file_write_and_wait_range(filp, start, end);
668 if (error)
669 return error;
672 * There is no need to serialise calls to blkdev_issue_flush with
673 * i_mutex and doing so causes performance issues with concurrent
674 * O_SYNC writers to a block device.
676 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
677 if (error == -EOPNOTSUPP)
678 error = 0;
680 return error;
682 EXPORT_SYMBOL(blkdev_fsync);
685 * bdev_read_page() - Start reading a page from a block device
686 * @bdev: The device to read the page from
687 * @sector: The offset on the device to read the page to (need not be aligned)
688 * @page: The page to read
690 * On entry, the page should be locked. It will be unlocked when the page
691 * has been read. If the block driver implements rw_page synchronously,
692 * that will be true on exit from this function, but it need not be.
694 * Errors returned by this function are usually "soft", eg out of memory, or
695 * queue full; callers should try a different route to read this page rather
696 * than propagate an error back up the stack.
698 * Return: negative errno if an error occurs, 0 if submission was successful.
700 int bdev_read_page(struct block_device *bdev, sector_t sector,
701 struct page *page)
703 const struct block_device_operations *ops = bdev->bd_disk->fops;
704 int result = -EOPNOTSUPP;
706 if (!ops->rw_page || bdev_get_integrity(bdev))
707 return result;
709 result = blk_queue_enter(bdev->bd_queue, 0);
710 if (result)
711 return result;
712 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
713 REQ_OP_READ);
714 blk_queue_exit(bdev->bd_queue);
715 return result;
717 EXPORT_SYMBOL_GPL(bdev_read_page);
720 * bdev_write_page() - Start writing a page to a block device
721 * @bdev: The device to write the page to
722 * @sector: The offset on the device to write the page to (need not be aligned)
723 * @page: The page to write
724 * @wbc: The writeback_control for the write
726 * On entry, the page should be locked and not currently under writeback.
727 * On exit, if the write started successfully, the page will be unlocked and
728 * under writeback. If the write failed already (eg the driver failed to
729 * queue the page to the device), the page will still be locked. If the
730 * caller is a ->writepage implementation, it will need to unlock the page.
732 * Errors returned by this function are usually "soft", eg out of memory, or
733 * queue full; callers should try a different route to write this page rather
734 * than propagate an error back up the stack.
736 * Return: negative errno if an error occurs, 0 if submission was successful.
738 int bdev_write_page(struct block_device *bdev, sector_t sector,
739 struct page *page, struct writeback_control *wbc)
741 int result;
742 const struct block_device_operations *ops = bdev->bd_disk->fops;
744 if (!ops->rw_page || bdev_get_integrity(bdev))
745 return -EOPNOTSUPP;
746 result = blk_queue_enter(bdev->bd_queue, 0);
747 if (result)
748 return result;
750 set_page_writeback(page);
751 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
752 REQ_OP_WRITE);
753 if (result) {
754 end_page_writeback(page);
755 } else {
756 clean_page_buffers(page);
757 unlock_page(page);
759 blk_queue_exit(bdev->bd_queue);
760 return result;
762 EXPORT_SYMBOL_GPL(bdev_write_page);
765 * pseudo-fs
768 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
769 static struct kmem_cache * bdev_cachep __read_mostly;
771 static struct inode *bdev_alloc_inode(struct super_block *sb)
773 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
774 if (!ei)
775 return NULL;
776 return &ei->vfs_inode;
779 static void bdev_free_inode(struct inode *inode)
781 kmem_cache_free(bdev_cachep, BDEV_I(inode));
784 static void init_once(void *foo)
786 struct bdev_inode *ei = (struct bdev_inode *) foo;
787 struct block_device *bdev = &ei->bdev;
789 memset(bdev, 0, sizeof(*bdev));
790 mutex_init(&bdev->bd_mutex);
791 INIT_LIST_HEAD(&bdev->bd_list);
792 #ifdef CONFIG_SYSFS
793 INIT_LIST_HEAD(&bdev->bd_holder_disks);
794 #endif
795 bdev->bd_bdi = &noop_backing_dev_info;
796 inode_init_once(&ei->vfs_inode);
797 /* Initialize mutex for freeze. */
798 mutex_init(&bdev->bd_fsfreeze_mutex);
801 static void bdev_evict_inode(struct inode *inode)
803 struct block_device *bdev = &BDEV_I(inode)->bdev;
804 truncate_inode_pages_final(&inode->i_data);
805 invalidate_inode_buffers(inode); /* is it needed here? */
806 clear_inode(inode);
807 spin_lock(&bdev_lock);
808 list_del_init(&bdev->bd_list);
809 spin_unlock(&bdev_lock);
810 /* Detach inode from wb early as bdi_put() may free bdi->wb */
811 inode_detach_wb(inode);
812 if (bdev->bd_bdi != &noop_backing_dev_info) {
813 bdi_put(bdev->bd_bdi);
814 bdev->bd_bdi = &noop_backing_dev_info;
818 static const struct super_operations bdev_sops = {
819 .statfs = simple_statfs,
820 .alloc_inode = bdev_alloc_inode,
821 .free_inode = bdev_free_inode,
822 .drop_inode = generic_delete_inode,
823 .evict_inode = bdev_evict_inode,
826 static int bd_init_fs_context(struct fs_context *fc)
828 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
829 if (!ctx)
830 return -ENOMEM;
831 fc->s_iflags |= SB_I_CGROUPWB;
832 ctx->ops = &bdev_sops;
833 return 0;
836 static struct file_system_type bd_type = {
837 .name = "bdev",
838 .init_fs_context = bd_init_fs_context,
839 .kill_sb = kill_anon_super,
842 struct super_block *blockdev_superblock __read_mostly;
843 EXPORT_SYMBOL_GPL(blockdev_superblock);
845 void __init bdev_cache_init(void)
847 int err;
848 static struct vfsmount *bd_mnt;
850 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
851 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
852 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
853 init_once);
854 err = register_filesystem(&bd_type);
855 if (err)
856 panic("Cannot register bdev pseudo-fs");
857 bd_mnt = kern_mount(&bd_type);
858 if (IS_ERR(bd_mnt))
859 panic("Cannot create bdev pseudo-fs");
860 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
864 * Most likely _very_ bad one - but then it's hardly critical for small
865 * /dev and can be fixed when somebody will need really large one.
866 * Keep in mind that it will be fed through icache hash function too.
868 static inline unsigned long hash(dev_t dev)
870 return MAJOR(dev)+MINOR(dev);
873 static int bdev_test(struct inode *inode, void *data)
875 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
878 static int bdev_set(struct inode *inode, void *data)
880 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
881 return 0;
884 static LIST_HEAD(all_bdevs);
887 * If there is a bdev inode for this device, unhash it so that it gets evicted
888 * as soon as last inode reference is dropped.
890 void bdev_unhash_inode(dev_t dev)
892 struct inode *inode;
894 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
895 if (inode) {
896 remove_inode_hash(inode);
897 iput(inode);
901 struct block_device *bdget(dev_t dev)
903 struct block_device *bdev;
904 struct inode *inode;
906 inode = iget5_locked(blockdev_superblock, hash(dev),
907 bdev_test, bdev_set, &dev);
909 if (!inode)
910 return NULL;
912 bdev = &BDEV_I(inode)->bdev;
914 if (inode->i_state & I_NEW) {
915 bdev->bd_contains = NULL;
916 bdev->bd_super = NULL;
917 bdev->bd_inode = inode;
918 bdev->bd_block_size = i_blocksize(inode);
919 bdev->bd_part_count = 0;
920 bdev->bd_invalidated = 0;
921 inode->i_mode = S_IFBLK;
922 inode->i_rdev = dev;
923 inode->i_bdev = bdev;
924 inode->i_data.a_ops = &def_blk_aops;
925 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
926 spin_lock(&bdev_lock);
927 list_add(&bdev->bd_list, &all_bdevs);
928 spin_unlock(&bdev_lock);
929 unlock_new_inode(inode);
931 return bdev;
934 EXPORT_SYMBOL(bdget);
937 * bdgrab -- Grab a reference to an already referenced block device
938 * @bdev: Block device to grab a reference to.
940 struct block_device *bdgrab(struct block_device *bdev)
942 ihold(bdev->bd_inode);
943 return bdev;
945 EXPORT_SYMBOL(bdgrab);
947 long nr_blockdev_pages(void)
949 struct block_device *bdev;
950 long ret = 0;
951 spin_lock(&bdev_lock);
952 list_for_each_entry(bdev, &all_bdevs, bd_list) {
953 ret += bdev->bd_inode->i_mapping->nrpages;
955 spin_unlock(&bdev_lock);
956 return ret;
959 void bdput(struct block_device *bdev)
961 iput(bdev->bd_inode);
964 EXPORT_SYMBOL(bdput);
966 static struct block_device *bd_acquire(struct inode *inode)
968 struct block_device *bdev;
970 spin_lock(&bdev_lock);
971 bdev = inode->i_bdev;
972 if (bdev && !inode_unhashed(bdev->bd_inode)) {
973 bdgrab(bdev);
974 spin_unlock(&bdev_lock);
975 return bdev;
977 spin_unlock(&bdev_lock);
980 * i_bdev references block device inode that was already shut down
981 * (corresponding device got removed). Remove the reference and look
982 * up block device inode again just in case new device got
983 * reestablished under the same device number.
985 if (bdev)
986 bd_forget(inode);
988 bdev = bdget(inode->i_rdev);
989 if (bdev) {
990 spin_lock(&bdev_lock);
991 if (!inode->i_bdev) {
993 * We take an additional reference to bd_inode,
994 * and it's released in clear_inode() of inode.
995 * So, we can access it via ->i_mapping always
996 * without igrab().
998 bdgrab(bdev);
999 inode->i_bdev = bdev;
1000 inode->i_mapping = bdev->bd_inode->i_mapping;
1002 spin_unlock(&bdev_lock);
1004 return bdev;
1007 /* Call when you free inode */
1009 void bd_forget(struct inode *inode)
1011 struct block_device *bdev = NULL;
1013 spin_lock(&bdev_lock);
1014 if (!sb_is_blkdev_sb(inode->i_sb))
1015 bdev = inode->i_bdev;
1016 inode->i_bdev = NULL;
1017 inode->i_mapping = &inode->i_data;
1018 spin_unlock(&bdev_lock);
1020 if (bdev)
1021 bdput(bdev);
1025 * bd_may_claim - test whether a block device can be claimed
1026 * @bdev: block device of interest
1027 * @whole: whole block device containing @bdev, may equal @bdev
1028 * @holder: holder trying to claim @bdev
1030 * Test whether @bdev can be claimed by @holder.
1032 * CONTEXT:
1033 * spin_lock(&bdev_lock).
1035 * RETURNS:
1036 * %true if @bdev can be claimed, %false otherwise.
1038 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1039 void *holder)
1041 if (bdev->bd_holder == holder)
1042 return true; /* already a holder */
1043 else if (bdev->bd_holder != NULL)
1044 return false; /* held by someone else */
1045 else if (whole == bdev)
1046 return true; /* is a whole device which isn't held */
1048 else if (whole->bd_holder == bd_may_claim)
1049 return true; /* is a partition of a device that is being partitioned */
1050 else if (whole->bd_holder != NULL)
1051 return false; /* is a partition of a held device */
1052 else
1053 return true; /* is a partition of an un-held device */
1057 * bd_prepare_to_claim - prepare to claim a block device
1058 * @bdev: block device of interest
1059 * @whole: the whole device containing @bdev, may equal @bdev
1060 * @holder: holder trying to claim @bdev
1062 * Prepare to claim @bdev. This function fails if @bdev is already
1063 * claimed by another holder and waits if another claiming is in
1064 * progress. This function doesn't actually claim. On successful
1065 * return, the caller has ownership of bd_claiming and bd_holder[s].
1067 * CONTEXT:
1068 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1069 * it multiple times.
1071 * RETURNS:
1072 * 0 if @bdev can be claimed, -EBUSY otherwise.
1074 static int bd_prepare_to_claim(struct block_device *bdev,
1075 struct block_device *whole, void *holder)
1077 retry:
1078 /* if someone else claimed, fail */
1079 if (!bd_may_claim(bdev, whole, holder))
1080 return -EBUSY;
1082 /* if claiming is already in progress, wait for it to finish */
1083 if (whole->bd_claiming) {
1084 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1085 DEFINE_WAIT(wait);
1087 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1088 spin_unlock(&bdev_lock);
1089 schedule();
1090 finish_wait(wq, &wait);
1091 spin_lock(&bdev_lock);
1092 goto retry;
1095 /* yay, all mine */
1096 return 0;
1099 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1101 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1103 if (!disk)
1104 return NULL;
1106 * Now that we hold gendisk reference we make sure bdev we looked up is
1107 * not stale. If it is, it means device got removed and created before
1108 * we looked up gendisk and we fail open in such case. Associating
1109 * unhashed bdev with newly created gendisk could lead to two bdevs
1110 * (and thus two independent caches) being associated with one device
1111 * which is bad.
1113 if (inode_unhashed(bdev->bd_inode)) {
1114 put_disk_and_module(disk);
1115 return NULL;
1117 return disk;
1121 * bd_start_claiming - start claiming a block device
1122 * @bdev: block device of interest
1123 * @holder: holder trying to claim @bdev
1125 * @bdev is about to be opened exclusively. Check @bdev can be opened
1126 * exclusively and mark that an exclusive open is in progress. Each
1127 * successful call to this function must be matched with a call to
1128 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1129 * fail).
1131 * This function is used to gain exclusive access to the block device
1132 * without actually causing other exclusive open attempts to fail. It
1133 * should be used when the open sequence itself requires exclusive
1134 * access but may subsequently fail.
1136 * CONTEXT:
1137 * Might sleep.
1139 * RETURNS:
1140 * Pointer to the block device containing @bdev on success, ERR_PTR()
1141 * value on failure.
1143 struct block_device *bd_start_claiming(struct block_device *bdev, void *holder)
1145 struct gendisk *disk;
1146 struct block_device *whole;
1147 int partno, err;
1149 might_sleep();
1152 * @bdev might not have been initialized properly yet, look up
1153 * and grab the outer block device the hard way.
1155 disk = bdev_get_gendisk(bdev, &partno);
1156 if (!disk)
1157 return ERR_PTR(-ENXIO);
1160 * Normally, @bdev should equal what's returned from bdget_disk()
1161 * if partno is 0; however, some drivers (floppy) use multiple
1162 * bdev's for the same physical device and @bdev may be one of the
1163 * aliases. Keep @bdev if partno is 0. This means claimer
1164 * tracking is broken for those devices but it has always been that
1165 * way.
1167 if (partno)
1168 whole = bdget_disk(disk, 0);
1169 else
1170 whole = bdgrab(bdev);
1172 put_disk_and_module(disk);
1173 if (!whole)
1174 return ERR_PTR(-ENOMEM);
1176 /* prepare to claim, if successful, mark claiming in progress */
1177 spin_lock(&bdev_lock);
1179 err = bd_prepare_to_claim(bdev, whole, holder);
1180 if (err == 0) {
1181 whole->bd_claiming = holder;
1182 spin_unlock(&bdev_lock);
1183 return whole;
1184 } else {
1185 spin_unlock(&bdev_lock);
1186 bdput(whole);
1187 return ERR_PTR(err);
1190 EXPORT_SYMBOL(bd_start_claiming);
1192 static void bd_clear_claiming(struct block_device *whole, void *holder)
1194 lockdep_assert_held(&bdev_lock);
1195 /* tell others that we're done */
1196 BUG_ON(whole->bd_claiming != holder);
1197 whole->bd_claiming = NULL;
1198 wake_up_bit(&whole->bd_claiming, 0);
1202 * bd_finish_claiming - finish claiming of a block device
1203 * @bdev: block device of interest
1204 * @whole: whole block device (returned from bd_start_claiming())
1205 * @holder: holder that has claimed @bdev
1207 * Finish exclusive open of a block device. Mark the device as exlusively
1208 * open by the holder and wake up all waiters for exclusive open to finish.
1210 void bd_finish_claiming(struct block_device *bdev, struct block_device *whole,
1211 void *holder)
1213 spin_lock(&bdev_lock);
1214 BUG_ON(!bd_may_claim(bdev, whole, holder));
1216 * Note that for a whole device bd_holders will be incremented twice,
1217 * and bd_holder will be set to bd_may_claim before being set to holder
1219 whole->bd_holders++;
1220 whole->bd_holder = bd_may_claim;
1221 bdev->bd_holders++;
1222 bdev->bd_holder = holder;
1223 bd_clear_claiming(whole, holder);
1224 spin_unlock(&bdev_lock);
1226 EXPORT_SYMBOL(bd_finish_claiming);
1229 * bd_abort_claiming - abort claiming of a block device
1230 * @bdev: block device of interest
1231 * @whole: whole block device (returned from bd_start_claiming())
1232 * @holder: holder that has claimed @bdev
1234 * Abort claiming of a block device when the exclusive open failed. This can be
1235 * also used when exclusive open is not actually desired and we just needed
1236 * to block other exclusive openers for a while.
1238 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1239 void *holder)
1241 spin_lock(&bdev_lock);
1242 bd_clear_claiming(whole, holder);
1243 spin_unlock(&bdev_lock);
1245 EXPORT_SYMBOL(bd_abort_claiming);
1247 #ifdef CONFIG_SYSFS
1248 struct bd_holder_disk {
1249 struct list_head list;
1250 struct gendisk *disk;
1251 int refcnt;
1254 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1255 struct gendisk *disk)
1257 struct bd_holder_disk *holder;
1259 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1260 if (holder->disk == disk)
1261 return holder;
1262 return NULL;
1265 static int add_symlink(struct kobject *from, struct kobject *to)
1267 return sysfs_create_link(from, to, kobject_name(to));
1270 static void del_symlink(struct kobject *from, struct kobject *to)
1272 sysfs_remove_link(from, kobject_name(to));
1276 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1277 * @bdev: the claimed slave bdev
1278 * @disk: the holding disk
1280 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1282 * This functions creates the following sysfs symlinks.
1284 * - from "slaves" directory of the holder @disk to the claimed @bdev
1285 * - from "holders" directory of the @bdev to the holder @disk
1287 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1288 * passed to bd_link_disk_holder(), then:
1290 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1291 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1293 * The caller must have claimed @bdev before calling this function and
1294 * ensure that both @bdev and @disk are valid during the creation and
1295 * lifetime of these symlinks.
1297 * CONTEXT:
1298 * Might sleep.
1300 * RETURNS:
1301 * 0 on success, -errno on failure.
1303 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1305 struct bd_holder_disk *holder;
1306 int ret = 0;
1308 mutex_lock(&bdev->bd_mutex);
1310 WARN_ON_ONCE(!bdev->bd_holder);
1312 /* FIXME: remove the following once add_disk() handles errors */
1313 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1314 goto out_unlock;
1316 holder = bd_find_holder_disk(bdev, disk);
1317 if (holder) {
1318 holder->refcnt++;
1319 goto out_unlock;
1322 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1323 if (!holder) {
1324 ret = -ENOMEM;
1325 goto out_unlock;
1328 INIT_LIST_HEAD(&holder->list);
1329 holder->disk = disk;
1330 holder->refcnt = 1;
1332 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1333 if (ret)
1334 goto out_free;
1336 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1337 if (ret)
1338 goto out_del;
1340 * bdev could be deleted beneath us which would implicitly destroy
1341 * the holder directory. Hold on to it.
1343 kobject_get(bdev->bd_part->holder_dir);
1345 list_add(&holder->list, &bdev->bd_holder_disks);
1346 goto out_unlock;
1348 out_del:
1349 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1350 out_free:
1351 kfree(holder);
1352 out_unlock:
1353 mutex_unlock(&bdev->bd_mutex);
1354 return ret;
1356 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1359 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1360 * @bdev: the calimed slave bdev
1361 * @disk: the holding disk
1363 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1365 * CONTEXT:
1366 * Might sleep.
1368 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1370 struct bd_holder_disk *holder;
1372 mutex_lock(&bdev->bd_mutex);
1374 holder = bd_find_holder_disk(bdev, disk);
1376 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1377 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1378 del_symlink(bdev->bd_part->holder_dir,
1379 &disk_to_dev(disk)->kobj);
1380 kobject_put(bdev->bd_part->holder_dir);
1381 list_del_init(&holder->list);
1382 kfree(holder);
1385 mutex_unlock(&bdev->bd_mutex);
1387 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1388 #endif
1391 * flush_disk - invalidates all buffer-cache entries on a disk
1393 * @bdev: struct block device to be flushed
1394 * @kill_dirty: flag to guide handling of dirty inodes
1396 * Invalidates all buffer-cache entries on a disk. It should be called
1397 * when a disk has been changed -- either by a media change or online
1398 * resize.
1400 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1402 if (__invalidate_device(bdev, kill_dirty)) {
1403 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1404 "resized disk %s\n",
1405 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1407 bdev->bd_invalidated = 1;
1411 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1412 * @disk: struct gendisk to check
1413 * @bdev: struct bdev to adjust.
1414 * @verbose: if %true log a message about a size change if there is any
1416 * This routine checks to see if the bdev size does not match the disk size
1417 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1418 * are freed.
1420 static void check_disk_size_change(struct gendisk *disk,
1421 struct block_device *bdev, bool verbose)
1423 loff_t disk_size, bdev_size;
1425 disk_size = (loff_t)get_capacity(disk) << 9;
1426 bdev_size = i_size_read(bdev->bd_inode);
1427 if (disk_size != bdev_size) {
1428 if (verbose) {
1429 printk(KERN_INFO
1430 "%s: detected capacity change from %lld to %lld\n",
1431 disk->disk_name, bdev_size, disk_size);
1433 i_size_write(bdev->bd_inode, disk_size);
1434 if (bdev_size > disk_size)
1435 flush_disk(bdev, false);
1437 bdev->bd_invalidated = 0;
1441 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1442 * @disk: struct gendisk to be revalidated
1444 * This routine is a wrapper for lower-level driver's revalidate_disk
1445 * call-backs. It is used to do common pre and post operations needed
1446 * for all revalidate_disk operations.
1448 int revalidate_disk(struct gendisk *disk)
1450 int ret = 0;
1452 if (disk->fops->revalidate_disk)
1453 ret = disk->fops->revalidate_disk(disk);
1456 * Hidden disks don't have associated bdev so there's no point in
1457 * revalidating it.
1459 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1460 struct block_device *bdev = bdget_disk(disk, 0);
1462 if (!bdev)
1463 return ret;
1465 mutex_lock(&bdev->bd_mutex);
1466 check_disk_size_change(disk, bdev, ret == 0);
1467 mutex_unlock(&bdev->bd_mutex);
1468 bdput(bdev);
1470 return ret;
1472 EXPORT_SYMBOL(revalidate_disk);
1475 * This routine checks whether a removable media has been changed,
1476 * and invalidates all buffer-cache-entries in that case. This
1477 * is a relatively slow routine, so we have to try to minimize using
1478 * it. Thus it is called only upon a 'mount' or 'open'. This
1479 * is the best way of combining speed and utility, I think.
1480 * People changing diskettes in the middle of an operation deserve
1481 * to lose :-)
1483 int check_disk_change(struct block_device *bdev)
1485 struct gendisk *disk = bdev->bd_disk;
1486 const struct block_device_operations *bdops = disk->fops;
1487 unsigned int events;
1489 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1490 DISK_EVENT_EJECT_REQUEST);
1491 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1492 return 0;
1494 flush_disk(bdev, true);
1495 if (bdops->revalidate_disk)
1496 bdops->revalidate_disk(bdev->bd_disk);
1497 return 1;
1500 EXPORT_SYMBOL(check_disk_change);
1502 void bd_set_size(struct block_device *bdev, loff_t size)
1504 inode_lock(bdev->bd_inode);
1505 i_size_write(bdev->bd_inode, size);
1506 inode_unlock(bdev->bd_inode);
1508 EXPORT_SYMBOL(bd_set_size);
1510 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1512 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1514 struct gendisk *disk = bdev->bd_disk;
1515 int ret;
1517 lockdep_assert_held(&bdev->bd_mutex);
1519 rescan:
1520 ret = blk_drop_partitions(disk, bdev);
1521 if (ret)
1522 return ret;
1525 * Historically we only set the capacity to zero for devices that
1526 * support partitions (independ of actually having partitions created).
1527 * Doing that is rather inconsistent, but changing it broke legacy
1528 * udisks polling for legacy ide-cdrom devices. Use the crude check
1529 * below to get the sane behavior for most device while not breaking
1530 * userspace for this particular setup.
1532 if (invalidate) {
1533 if (disk_part_scan_enabled(disk) ||
1534 !(disk->flags & GENHD_FL_REMOVABLE))
1535 set_capacity(disk, 0);
1536 } else {
1537 if (disk->fops->revalidate_disk)
1538 disk->fops->revalidate_disk(disk);
1541 check_disk_size_change(disk, bdev, !invalidate);
1543 if (get_capacity(disk)) {
1544 ret = blk_add_partitions(disk, bdev);
1545 if (ret == -EAGAIN)
1546 goto rescan;
1547 } else if (invalidate) {
1549 * Tell userspace that the media / partition table may have
1550 * changed.
1552 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1555 return ret;
1558 * Only exported for for loop and dasd for historic reasons. Don't use in new
1559 * code!
1561 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1564 * bd_mutex locking:
1566 * mutex_lock(part->bd_mutex)
1567 * mutex_lock_nested(whole->bd_mutex, 1)
1570 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1572 struct gendisk *disk;
1573 int ret;
1574 int partno;
1575 int perm = 0;
1576 bool first_open = false;
1578 if (mode & FMODE_READ)
1579 perm |= MAY_READ;
1580 if (mode & FMODE_WRITE)
1581 perm |= MAY_WRITE;
1583 * hooks: /n/, see "layering violations".
1585 if (!for_part) {
1586 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1587 if (ret != 0) {
1588 bdput(bdev);
1589 return ret;
1593 restart:
1595 ret = -ENXIO;
1596 disk = bdev_get_gendisk(bdev, &partno);
1597 if (!disk)
1598 goto out;
1600 disk_block_events(disk);
1601 mutex_lock_nested(&bdev->bd_mutex, for_part);
1602 if (!bdev->bd_openers) {
1603 first_open = true;
1604 bdev->bd_disk = disk;
1605 bdev->bd_queue = disk->queue;
1606 bdev->bd_contains = bdev;
1607 bdev->bd_partno = partno;
1609 if (!partno) {
1610 ret = -ENXIO;
1611 bdev->bd_part = disk_get_part(disk, partno);
1612 if (!bdev->bd_part)
1613 goto out_clear;
1615 ret = 0;
1616 if (disk->fops->open) {
1617 ret = disk->fops->open(bdev, mode);
1618 if (ret == -ERESTARTSYS) {
1619 /* Lost a race with 'disk' being
1620 * deleted, try again.
1621 * See md.c
1623 disk_put_part(bdev->bd_part);
1624 bdev->bd_part = NULL;
1625 bdev->bd_disk = NULL;
1626 bdev->bd_queue = NULL;
1627 mutex_unlock(&bdev->bd_mutex);
1628 disk_unblock_events(disk);
1629 put_disk_and_module(disk);
1630 goto restart;
1634 if (!ret) {
1635 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1636 set_init_blocksize(bdev);
1640 * If the device is invalidated, rescan partition
1641 * if open succeeded or failed with -ENOMEDIUM.
1642 * The latter is necessary to prevent ghost
1643 * partitions on a removed medium.
1645 if (bdev->bd_invalidated &&
1646 (!ret || ret == -ENOMEDIUM))
1647 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1649 if (ret)
1650 goto out_clear;
1651 } else {
1652 struct block_device *whole;
1653 whole = bdget_disk(disk, 0);
1654 ret = -ENOMEM;
1655 if (!whole)
1656 goto out_clear;
1657 BUG_ON(for_part);
1658 ret = __blkdev_get(whole, mode, 1);
1659 if (ret)
1660 goto out_clear;
1661 bdev->bd_contains = whole;
1662 bdev->bd_part = disk_get_part(disk, partno);
1663 if (!(disk->flags & GENHD_FL_UP) ||
1664 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1665 ret = -ENXIO;
1666 goto out_clear;
1668 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1669 set_init_blocksize(bdev);
1672 if (bdev->bd_bdi == &noop_backing_dev_info)
1673 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1674 } else {
1675 if (bdev->bd_contains == bdev) {
1676 ret = 0;
1677 if (bdev->bd_disk->fops->open)
1678 ret = bdev->bd_disk->fops->open(bdev, mode);
1679 /* the same as first opener case, read comment there */
1680 if (bdev->bd_invalidated &&
1681 (!ret || ret == -ENOMEDIUM))
1682 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1683 if (ret)
1684 goto out_unlock_bdev;
1687 bdev->bd_openers++;
1688 if (for_part)
1689 bdev->bd_part_count++;
1690 mutex_unlock(&bdev->bd_mutex);
1691 disk_unblock_events(disk);
1692 /* only one opener holds refs to the module and disk */
1693 if (!first_open)
1694 put_disk_and_module(disk);
1695 return 0;
1697 out_clear:
1698 disk_put_part(bdev->bd_part);
1699 bdev->bd_disk = NULL;
1700 bdev->bd_part = NULL;
1701 bdev->bd_queue = NULL;
1702 if (bdev != bdev->bd_contains)
1703 __blkdev_put(bdev->bd_contains, mode, 1);
1704 bdev->bd_contains = NULL;
1705 out_unlock_bdev:
1706 mutex_unlock(&bdev->bd_mutex);
1707 disk_unblock_events(disk);
1708 put_disk_and_module(disk);
1709 out:
1710 bdput(bdev);
1712 return ret;
1716 * blkdev_get - open a block device
1717 * @bdev: block_device to open
1718 * @mode: FMODE_* mask
1719 * @holder: exclusive holder identifier
1721 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1722 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1723 * @holder is invalid. Exclusive opens may nest for the same @holder.
1725 * On success, the reference count of @bdev is unchanged. On failure,
1726 * @bdev is put.
1728 * CONTEXT:
1729 * Might sleep.
1731 * RETURNS:
1732 * 0 on success, -errno on failure.
1734 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1736 struct block_device *whole = NULL;
1737 int res;
1739 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1741 if ((mode & FMODE_EXCL) && holder) {
1742 whole = bd_start_claiming(bdev, holder);
1743 if (IS_ERR(whole)) {
1744 bdput(bdev);
1745 return PTR_ERR(whole);
1749 res = __blkdev_get(bdev, mode, 0);
1751 if (whole) {
1752 struct gendisk *disk = whole->bd_disk;
1754 /* finish claiming */
1755 mutex_lock(&bdev->bd_mutex);
1756 if (!res)
1757 bd_finish_claiming(bdev, whole, holder);
1758 else
1759 bd_abort_claiming(bdev, whole, holder);
1761 * Block event polling for write claims if requested. Any
1762 * write holder makes the write_holder state stick until
1763 * all are released. This is good enough and tracking
1764 * individual writeable reference is too fragile given the
1765 * way @mode is used in blkdev_get/put().
1767 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1768 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1769 bdev->bd_write_holder = true;
1770 disk_block_events(disk);
1773 mutex_unlock(&bdev->bd_mutex);
1774 bdput(whole);
1777 return res;
1779 EXPORT_SYMBOL(blkdev_get);
1782 * blkdev_get_by_path - open a block device by name
1783 * @path: path to the block device to open
1784 * @mode: FMODE_* mask
1785 * @holder: exclusive holder identifier
1787 * Open the blockdevice described by the device file at @path. @mode
1788 * and @holder are identical to blkdev_get().
1790 * On success, the returned block_device has reference count of one.
1792 * CONTEXT:
1793 * Might sleep.
1795 * RETURNS:
1796 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1798 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1799 void *holder)
1801 struct block_device *bdev;
1802 int err;
1804 bdev = lookup_bdev(path);
1805 if (IS_ERR(bdev))
1806 return bdev;
1808 err = blkdev_get(bdev, mode, holder);
1809 if (err)
1810 return ERR_PTR(err);
1812 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1813 blkdev_put(bdev, mode);
1814 return ERR_PTR(-EACCES);
1817 return bdev;
1819 EXPORT_SYMBOL(blkdev_get_by_path);
1822 * blkdev_get_by_dev - open a block device by device number
1823 * @dev: device number of block device to open
1824 * @mode: FMODE_* mask
1825 * @holder: exclusive holder identifier
1827 * Open the blockdevice described by device number @dev. @mode and
1828 * @holder are identical to blkdev_get().
1830 * Use it ONLY if you really do not have anything better - i.e. when
1831 * you are behind a truly sucky interface and all you are given is a
1832 * device number. _Never_ to be used for internal purposes. If you
1833 * ever need it - reconsider your API.
1835 * On success, the returned block_device has reference count of one.
1837 * CONTEXT:
1838 * Might sleep.
1840 * RETURNS:
1841 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1843 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1845 struct block_device *bdev;
1846 int err;
1848 bdev = bdget(dev);
1849 if (!bdev)
1850 return ERR_PTR(-ENOMEM);
1852 err = blkdev_get(bdev, mode, holder);
1853 if (err)
1854 return ERR_PTR(err);
1856 return bdev;
1858 EXPORT_SYMBOL(blkdev_get_by_dev);
1860 static int blkdev_open(struct inode * inode, struct file * filp)
1862 struct block_device *bdev;
1865 * Preserve backwards compatibility and allow large file access
1866 * even if userspace doesn't ask for it explicitly. Some mkfs
1867 * binary needs it. We might want to drop this workaround
1868 * during an unstable branch.
1870 filp->f_flags |= O_LARGEFILE;
1872 filp->f_mode |= FMODE_NOWAIT;
1874 if (filp->f_flags & O_NDELAY)
1875 filp->f_mode |= FMODE_NDELAY;
1876 if (filp->f_flags & O_EXCL)
1877 filp->f_mode |= FMODE_EXCL;
1878 if ((filp->f_flags & O_ACCMODE) == 3)
1879 filp->f_mode |= FMODE_WRITE_IOCTL;
1881 bdev = bd_acquire(inode);
1882 if (bdev == NULL)
1883 return -ENOMEM;
1885 filp->f_mapping = bdev->bd_inode->i_mapping;
1886 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1888 return blkdev_get(bdev, filp->f_mode, filp);
1891 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1893 struct gendisk *disk = bdev->bd_disk;
1894 struct block_device *victim = NULL;
1896 mutex_lock_nested(&bdev->bd_mutex, for_part);
1897 if (for_part)
1898 bdev->bd_part_count--;
1900 if (!--bdev->bd_openers) {
1901 WARN_ON_ONCE(bdev->bd_holders);
1902 sync_blockdev(bdev);
1903 kill_bdev(bdev);
1905 bdev_write_inode(bdev);
1907 if (bdev->bd_contains == bdev) {
1908 if (disk->fops->release)
1909 disk->fops->release(disk, mode);
1911 if (!bdev->bd_openers) {
1912 disk_put_part(bdev->bd_part);
1913 bdev->bd_part = NULL;
1914 bdev->bd_disk = NULL;
1915 if (bdev != bdev->bd_contains)
1916 victim = bdev->bd_contains;
1917 bdev->bd_contains = NULL;
1919 put_disk_and_module(disk);
1921 mutex_unlock(&bdev->bd_mutex);
1922 bdput(bdev);
1923 if (victim)
1924 __blkdev_put(victim, mode, 1);
1927 void blkdev_put(struct block_device *bdev, fmode_t mode)
1929 mutex_lock(&bdev->bd_mutex);
1931 if (mode & FMODE_EXCL) {
1932 bool bdev_free;
1935 * Release a claim on the device. The holder fields
1936 * are protected with bdev_lock. bd_mutex is to
1937 * synchronize disk_holder unlinking.
1939 spin_lock(&bdev_lock);
1941 WARN_ON_ONCE(--bdev->bd_holders < 0);
1942 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1944 /* bd_contains might point to self, check in a separate step */
1945 if ((bdev_free = !bdev->bd_holders))
1946 bdev->bd_holder = NULL;
1947 if (!bdev->bd_contains->bd_holders)
1948 bdev->bd_contains->bd_holder = NULL;
1950 spin_unlock(&bdev_lock);
1953 * If this was the last claim, remove holder link and
1954 * unblock evpoll if it was a write holder.
1956 if (bdev_free && bdev->bd_write_holder) {
1957 disk_unblock_events(bdev->bd_disk);
1958 bdev->bd_write_holder = false;
1963 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1964 * event. This is to ensure detection of media removal commanded
1965 * from userland - e.g. eject(1).
1967 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1969 mutex_unlock(&bdev->bd_mutex);
1971 __blkdev_put(bdev, mode, 0);
1973 EXPORT_SYMBOL(blkdev_put);
1975 static int blkdev_close(struct inode * inode, struct file * filp)
1977 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1978 blkdev_put(bdev, filp->f_mode);
1979 return 0;
1982 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1984 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1985 fmode_t mode = file->f_mode;
1988 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1989 * to updated it before every ioctl.
1991 if (file->f_flags & O_NDELAY)
1992 mode |= FMODE_NDELAY;
1993 else
1994 mode &= ~FMODE_NDELAY;
1996 return blkdev_ioctl(bdev, mode, cmd, arg);
2000 * Write data to the block device. Only intended for the block device itself
2001 * and the raw driver which basically is a fake block device.
2003 * Does not take i_mutex for the write and thus is not for general purpose
2004 * use.
2006 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
2008 struct file *file = iocb->ki_filp;
2009 struct inode *bd_inode = bdev_file_inode(file);
2010 loff_t size = i_size_read(bd_inode);
2011 struct blk_plug plug;
2012 ssize_t ret;
2014 if (bdev_read_only(I_BDEV(bd_inode)))
2015 return -EPERM;
2017 /* uswsusp needs write permission to the swap */
2018 if (IS_SWAPFILE(bd_inode) && !hibernation_available())
2019 return -ETXTBSY;
2021 if (!iov_iter_count(from))
2022 return 0;
2024 if (iocb->ki_pos >= size)
2025 return -ENOSPC;
2027 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
2028 return -EOPNOTSUPP;
2030 iov_iter_truncate(from, size - iocb->ki_pos);
2032 blk_start_plug(&plug);
2033 ret = __generic_file_write_iter(iocb, from);
2034 if (ret > 0)
2035 ret = generic_write_sync(iocb, ret);
2036 blk_finish_plug(&plug);
2037 return ret;
2039 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2041 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2043 struct file *file = iocb->ki_filp;
2044 struct inode *bd_inode = bdev_file_inode(file);
2045 loff_t size = i_size_read(bd_inode);
2046 loff_t pos = iocb->ki_pos;
2048 if (pos >= size)
2049 return 0;
2051 size -= pos;
2052 iov_iter_truncate(to, size);
2053 return generic_file_read_iter(iocb, to);
2055 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2058 * Try to release a page associated with block device when the system
2059 * is under memory pressure.
2061 static int blkdev_releasepage(struct page *page, gfp_t wait)
2063 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2065 if (super && super->s_op->bdev_try_to_free_page)
2066 return super->s_op->bdev_try_to_free_page(super, page, wait);
2068 return try_to_free_buffers(page);
2071 static int blkdev_writepages(struct address_space *mapping,
2072 struct writeback_control *wbc)
2074 return generic_writepages(mapping, wbc);
2077 static const struct address_space_operations def_blk_aops = {
2078 .readpage = blkdev_readpage,
2079 .readpages = blkdev_readpages,
2080 .writepage = blkdev_writepage,
2081 .write_begin = blkdev_write_begin,
2082 .write_end = blkdev_write_end,
2083 .writepages = blkdev_writepages,
2084 .releasepage = blkdev_releasepage,
2085 .direct_IO = blkdev_direct_IO,
2086 .migratepage = buffer_migrate_page_norefs,
2087 .is_dirty_writeback = buffer_check_dirty_writeback,
2090 #define BLKDEV_FALLOC_FL_SUPPORTED \
2091 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2092 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2094 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2095 loff_t len)
2097 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2098 struct address_space *mapping;
2099 loff_t end = start + len - 1;
2100 loff_t isize;
2101 int error;
2103 /* Fail if we don't recognize the flags. */
2104 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2105 return -EOPNOTSUPP;
2107 /* Don't go off the end of the device. */
2108 isize = i_size_read(bdev->bd_inode);
2109 if (start >= isize)
2110 return -EINVAL;
2111 if (end >= isize) {
2112 if (mode & FALLOC_FL_KEEP_SIZE) {
2113 len = isize - start;
2114 end = start + len - 1;
2115 } else
2116 return -EINVAL;
2120 * Don't allow IO that isn't aligned to logical block size.
2122 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2123 return -EINVAL;
2125 /* Invalidate the page cache, including dirty pages. */
2126 mapping = bdev->bd_inode->i_mapping;
2127 truncate_inode_pages_range(mapping, start, end);
2129 switch (mode) {
2130 case FALLOC_FL_ZERO_RANGE:
2131 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2132 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2133 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2134 break;
2135 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2136 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2137 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2138 break;
2139 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2140 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2141 GFP_KERNEL, 0);
2142 break;
2143 default:
2144 return -EOPNOTSUPP;
2146 if (error)
2147 return error;
2150 * Invalidate again; if someone wandered in and dirtied a page,
2151 * the caller will be given -EBUSY. The third argument is
2152 * inclusive, so the rounding here is safe.
2154 return invalidate_inode_pages2_range(mapping,
2155 start >> PAGE_SHIFT,
2156 end >> PAGE_SHIFT);
2159 const struct file_operations def_blk_fops = {
2160 .open = blkdev_open,
2161 .release = blkdev_close,
2162 .llseek = block_llseek,
2163 .read_iter = blkdev_read_iter,
2164 .write_iter = blkdev_write_iter,
2165 .iopoll = blkdev_iopoll,
2166 .mmap = generic_file_mmap,
2167 .fsync = blkdev_fsync,
2168 .unlocked_ioctl = block_ioctl,
2169 #ifdef CONFIG_COMPAT
2170 .compat_ioctl = compat_blkdev_ioctl,
2171 #endif
2172 .splice_read = generic_file_splice_read,
2173 .splice_write = iter_file_splice_write,
2174 .fallocate = blkdev_fallocate,
2177 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2179 int res;
2180 mm_segment_t old_fs = get_fs();
2181 set_fs(KERNEL_DS);
2182 res = blkdev_ioctl(bdev, 0, cmd, arg);
2183 set_fs(old_fs);
2184 return res;
2187 EXPORT_SYMBOL(ioctl_by_bdev);
2190 * lookup_bdev - lookup a struct block_device by name
2191 * @pathname: special file representing the block device
2193 * Get a reference to the blockdevice at @pathname in the current
2194 * namespace if possible and return it. Return ERR_PTR(error)
2195 * otherwise.
2197 struct block_device *lookup_bdev(const char *pathname)
2199 struct block_device *bdev;
2200 struct inode *inode;
2201 struct path path;
2202 int error;
2204 if (!pathname || !*pathname)
2205 return ERR_PTR(-EINVAL);
2207 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2208 if (error)
2209 return ERR_PTR(error);
2211 inode = d_backing_inode(path.dentry);
2212 error = -ENOTBLK;
2213 if (!S_ISBLK(inode->i_mode))
2214 goto fail;
2215 error = -EACCES;
2216 if (!may_open_dev(&path))
2217 goto fail;
2218 error = -ENOMEM;
2219 bdev = bd_acquire(inode);
2220 if (!bdev)
2221 goto fail;
2222 out:
2223 path_put(&path);
2224 return bdev;
2225 fail:
2226 bdev = ERR_PTR(error);
2227 goto out;
2229 EXPORT_SYMBOL(lookup_bdev);
2231 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2233 struct super_block *sb = get_super(bdev);
2234 int res = 0;
2236 if (sb) {
2238 * no need to lock the super, get_super holds the
2239 * read mutex so the filesystem cannot go away
2240 * under us (->put_super runs with the write lock
2241 * hold).
2243 shrink_dcache_sb(sb);
2244 res = invalidate_inodes(sb, kill_dirty);
2245 drop_super(sb);
2247 invalidate_bdev(bdev);
2248 return res;
2250 EXPORT_SYMBOL(__invalidate_device);
2252 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2254 struct inode *inode, *old_inode = NULL;
2256 spin_lock(&blockdev_superblock->s_inode_list_lock);
2257 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2258 struct address_space *mapping = inode->i_mapping;
2259 struct block_device *bdev;
2261 spin_lock(&inode->i_lock);
2262 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2263 mapping->nrpages == 0) {
2264 spin_unlock(&inode->i_lock);
2265 continue;
2267 __iget(inode);
2268 spin_unlock(&inode->i_lock);
2269 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2271 * We hold a reference to 'inode' so it couldn't have been
2272 * removed from s_inodes list while we dropped the
2273 * s_inode_list_lock We cannot iput the inode now as we can
2274 * be holding the last reference and we cannot iput it under
2275 * s_inode_list_lock. So we keep the reference and iput it
2276 * later.
2278 iput(old_inode);
2279 old_inode = inode;
2280 bdev = I_BDEV(inode);
2282 mutex_lock(&bdev->bd_mutex);
2283 if (bdev->bd_openers)
2284 func(bdev, arg);
2285 mutex_unlock(&bdev->bd_mutex);
2287 spin_lock(&blockdev_superblock->s_inode_list_lock);
2289 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2290 iput(old_inode);