4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev
;
41 struct inode vfs_inode
;
44 static const struct address_space_operations def_blk_aops
;
46 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
48 return container_of(inode
, struct bdev_inode
, vfs_inode
);
51 struct block_device
*I_BDEV(struct inode
*inode
)
53 return &BDEV_I(inode
)->bdev
;
55 EXPORT_SYMBOL(I_BDEV
);
57 static void bdev_write_inode(struct block_device
*bdev
)
59 struct inode
*inode
= bdev
->bd_inode
;
62 spin_lock(&inode
->i_lock
);
63 while (inode
->i_state
& I_DIRTY
) {
64 spin_unlock(&inode
->i_lock
);
65 ret
= write_inode_now(inode
, true);
67 char name
[BDEVNAME_SIZE
];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev
, name
), ret
);
72 spin_lock(&inode
->i_lock
);
74 spin_unlock(&inode
->i_lock
);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 void kill_bdev(struct block_device
*bdev
)
80 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
82 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
86 truncate_inode_pages(mapping
, 0);
88 EXPORT_SYMBOL(kill_bdev
);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device
*bdev
)
93 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
95 if (mapping
->nrpages
) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping
, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping
);
105 EXPORT_SYMBOL(invalidate_bdev
);
107 static void set_init_blocksize(struct block_device
*bdev
)
109 unsigned bsize
= bdev_logical_block_size(bdev
);
110 loff_t size
= i_size_read(bdev
->bd_inode
);
112 while (bsize
< PAGE_SIZE
) {
117 bdev
->bd_block_size
= bsize
;
118 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
121 int set_blocksize(struct block_device
*bdev
, int size
)
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
124 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
127 /* Size cannot be smaller than the size supported by the device */
128 if (size
< bdev_logical_block_size(bdev
))
131 /* Don't change the size if it is same as current */
132 if (bdev
->bd_block_size
!= size
) {
134 bdev
->bd_block_size
= size
;
135 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
141 EXPORT_SYMBOL(set_blocksize
);
143 int sb_set_blocksize(struct super_block
*sb
, int size
)
145 if (set_blocksize(sb
->s_bdev
, size
))
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb
->s_blocksize
= size
;
150 sb
->s_blocksize_bits
= blksize_bits(size
);
151 return sb
->s_blocksize
;
154 EXPORT_SYMBOL(sb_set_blocksize
);
156 int sb_min_blocksize(struct super_block
*sb
, int size
)
158 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
161 return sb_set_blocksize(sb
, size
);
164 EXPORT_SYMBOL(sb_min_blocksize
);
167 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
168 struct buffer_head
*bh
, int create
)
170 bh
->b_bdev
= I_BDEV(inode
);
171 bh
->b_blocknr
= iblock
;
172 set_buffer_mapped(bh
);
176 static struct inode
*bdev_file_inode(struct file
*file
)
178 return file
->f_mapping
->host
;
181 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
183 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
185 /* avoid the need for a I/O completion work item */
186 if (iocb
->ki_flags
& IOCB_DSYNC
)
191 #define DIO_INLINE_BIO_VECS 4
193 static void blkdev_bio_end_io_simple(struct bio
*bio
)
195 struct task_struct
*waiter
= bio
->bi_private
;
197 WRITE_ONCE(bio
->bi_private
, NULL
);
198 blk_wake_io_task(waiter
);
202 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
205 struct file
*file
= iocb
->ki_filp
;
206 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
207 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
208 loff_t pos
= iocb
->ki_pos
;
209 bool should_dirty
= false;
215 if ((pos
| iov_iter_alignment(iter
)) &
216 (bdev_logical_block_size(bdev
) - 1))
219 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
222 vecs
= kmalloc_array(nr_pages
, sizeof(struct bio_vec
),
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
);
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
))
246 bio
.bi_opf
= dio_bio_write_op(iocb
);
247 task_io_account_write(ret
);
249 if (iocb
->ki_flags
& IOCB_HIPRI
)
250 bio
.bi_opf
|= REQ_HIPRI
;
252 qc
= submit_bio(&bio
);
254 set_current_state(TASK_UNINTERRUPTIBLE
);
255 if (!READ_ONCE(bio
.bi_private
))
257 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
258 !blk_poll(bdev_get_queue(bdev
), qc
, true))
261 __set_current_state(TASK_RUNNING
);
263 bio_for_each_segment_all(bvec
, &bio
, i
) {
264 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
265 set_page_dirty_lock(bvec
->bv_page
);
266 put_page(bvec
->bv_page
);
269 if (unlikely(bio
.bi_status
))
270 ret
= blk_status_to_errno(bio
.bi_status
);
273 if (vecs
!= inline_vecs
)
284 struct task_struct
*waiter
;
289 bool should_dirty
: 1;
294 static struct bio_set blkdev_dio_pool
;
296 static void blkdev_bio_end_io(struct bio
*bio
)
298 struct blkdev_dio
*dio
= bio
->bi_private
;
299 bool should_dirty
= dio
->should_dirty
;
301 if (dio
->multi_bio
&& !atomic_dec_and_test(&dio
->ref
)) {
302 if (bio
->bi_status
&& !dio
->bio
.bi_status
)
303 dio
->bio
.bi_status
= bio
->bi_status
;
306 struct kiocb
*iocb
= dio
->iocb
;
309 if (likely(!dio
->bio
.bi_status
)) {
313 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
316 dio
->iocb
->ki_complete(iocb
, ret
, 0);
320 struct task_struct
*waiter
= dio
->waiter
;
322 WRITE_ONCE(dio
->waiter
, NULL
);
323 blk_wake_io_task(waiter
);
328 bio_check_pages_dirty(bio
);
330 struct bio_vec
*bvec
;
333 bio_for_each_segment_all(bvec
, bio
, i
)
334 put_page(bvec
->bv_page
);
340 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
342 struct file
*file
= iocb
->ki_filp
;
343 struct inode
*inode
= bdev_file_inode(file
);
344 struct block_device
*bdev
= I_BDEV(inode
);
345 struct blk_plug plug
;
346 struct blkdev_dio
*dio
;
348 bool is_poll
= (iocb
->ki_flags
& IOCB_HIPRI
) != 0;
349 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
350 loff_t pos
= iocb
->ki_pos
;
351 blk_qc_t qc
= BLK_QC_T_NONE
;
354 if ((pos
| iov_iter_alignment(iter
)) &
355 (bdev_logical_block_size(bdev
) - 1))
358 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, &blkdev_dio_pool
);
360 dio
= container_of(bio
, struct blkdev_dio
, bio
);
361 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
363 dio
->waiter
= current
;
370 dio
->multi_bio
= false;
371 dio
->should_dirty
= is_read
&& iter_is_iovec(iter
);
374 * Don't plug for HIPRI/polled IO, as those should go straight
378 blk_start_plug(&plug
);
381 bio_set_dev(bio
, bdev
);
382 bio
->bi_iter
.bi_sector
= pos
>> 9;
383 bio
->bi_write_hint
= iocb
->ki_hint
;
384 bio
->bi_private
= dio
;
385 bio
->bi_end_io
= blkdev_bio_end_io
;
386 bio
->bi_ioprio
= iocb
->ki_ioprio
;
388 ret
= bio_iov_iter_get_pages(bio
, iter
);
390 bio
->bi_status
= BLK_STS_IOERR
;
396 bio
->bi_opf
= REQ_OP_READ
;
397 if (dio
->should_dirty
)
398 bio_set_pages_dirty(bio
);
400 bio
->bi_opf
= dio_bio_write_op(iocb
);
401 task_io_account_write(bio
->bi_iter
.bi_size
);
404 dio
->size
+= bio
->bi_iter
.bi_size
;
405 pos
+= bio
->bi_iter
.bi_size
;
407 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
409 if (iocb
->ki_flags
& IOCB_HIPRI
)
410 bio
->bi_opf
|= REQ_HIPRI
;
412 qc
= submit_bio(bio
);
416 if (!dio
->multi_bio
) {
418 * AIO needs an extra reference to ensure the dio
419 * structure which is embedded into the first bio
424 dio
->multi_bio
= true;
425 atomic_set(&dio
->ref
, 2);
427 atomic_inc(&dio
->ref
);
431 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
435 blk_finish_plug(&plug
);
441 set_current_state(TASK_UNINTERRUPTIBLE
);
442 if (!READ_ONCE(dio
->waiter
))
445 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
446 !blk_poll(bdev_get_queue(bdev
), qc
, true))
449 __set_current_state(TASK_RUNNING
);
452 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
461 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
465 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
468 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
469 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
471 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
474 static __init
int blkdev_init(void)
476 return bioset_init(&blkdev_dio_pool
, 4, offsetof(struct blkdev_dio
, bio
), BIOSET_NEED_BVECS
);
478 module_init(blkdev_init
);
480 int __sync_blockdev(struct block_device
*bdev
, int wait
)
485 return filemap_flush(bdev
->bd_inode
->i_mapping
);
486 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
490 * Write out and wait upon all the dirty data associated with a block
491 * device via its mapping. Does not take the superblock lock.
493 int sync_blockdev(struct block_device
*bdev
)
495 return __sync_blockdev(bdev
, 1);
497 EXPORT_SYMBOL(sync_blockdev
);
500 * Write out and wait upon all dirty data associated with this
501 * device. Filesystem data as well as the underlying block
502 * device. Takes the superblock lock.
504 int fsync_bdev(struct block_device
*bdev
)
506 struct super_block
*sb
= get_super(bdev
);
508 int res
= sync_filesystem(sb
);
512 return sync_blockdev(bdev
);
514 EXPORT_SYMBOL(fsync_bdev
);
517 * freeze_bdev -- lock a filesystem and force it into a consistent state
518 * @bdev: blockdevice to lock
520 * If a superblock is found on this device, we take the s_umount semaphore
521 * on it to make sure nobody unmounts until the snapshot creation is done.
522 * The reference counter (bd_fsfreeze_count) guarantees that only the last
523 * unfreeze process can unfreeze the frozen filesystem actually when multiple
524 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
525 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
528 struct super_block
*freeze_bdev(struct block_device
*bdev
)
530 struct super_block
*sb
;
533 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
534 if (++bdev
->bd_fsfreeze_count
> 1) {
536 * We don't even need to grab a reference - the first call
537 * to freeze_bdev grab an active reference and only the last
538 * thaw_bdev drops it.
540 sb
= get_super(bdev
);
543 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
547 sb
= get_active_super(bdev
);
550 if (sb
->s_op
->freeze_super
)
551 error
= sb
->s_op
->freeze_super(sb
);
553 error
= freeze_super(sb
);
555 deactivate_super(sb
);
556 bdev
->bd_fsfreeze_count
--;
557 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
558 return ERR_PTR(error
);
560 deactivate_super(sb
);
563 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
564 return sb
; /* thaw_bdev releases s->s_umount */
566 EXPORT_SYMBOL(freeze_bdev
);
569 * thaw_bdev -- unlock filesystem
570 * @bdev: blockdevice to unlock
571 * @sb: associated superblock
573 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
575 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
579 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
580 if (!bdev
->bd_fsfreeze_count
)
584 if (--bdev
->bd_fsfreeze_count
> 0)
590 if (sb
->s_op
->thaw_super
)
591 error
= sb
->s_op
->thaw_super(sb
);
593 error
= thaw_super(sb
);
595 bdev
->bd_fsfreeze_count
++;
597 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
600 EXPORT_SYMBOL(thaw_bdev
);
602 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
604 return block_write_full_page(page
, blkdev_get_block
, wbc
);
607 static int blkdev_readpage(struct file
* file
, struct page
* page
)
609 return block_read_full_page(page
, blkdev_get_block
);
612 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
613 struct list_head
*pages
, unsigned nr_pages
)
615 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
618 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
619 loff_t pos
, unsigned len
, unsigned flags
,
620 struct page
**pagep
, void **fsdata
)
622 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
626 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
627 loff_t pos
, unsigned len
, unsigned copied
,
628 struct page
*page
, void *fsdata
)
631 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
641 * for a block special file file_inode(file)->i_size is zero
642 * so we compute the size by hand (just as in block_read/write above)
644 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
646 struct inode
*bd_inode
= bdev_file_inode(file
);
649 inode_lock(bd_inode
);
650 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
651 inode_unlock(bd_inode
);
655 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
657 struct inode
*bd_inode
= bdev_file_inode(filp
);
658 struct block_device
*bdev
= I_BDEV(bd_inode
);
661 error
= file_write_and_wait_range(filp
, start
, end
);
666 * There is no need to serialise calls to blkdev_issue_flush with
667 * i_mutex and doing so causes performance issues with concurrent
668 * O_SYNC writers to a block device.
670 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
671 if (error
== -EOPNOTSUPP
)
676 EXPORT_SYMBOL(blkdev_fsync
);
679 * bdev_read_page() - Start reading a page from a block device
680 * @bdev: The device to read the page from
681 * @sector: The offset on the device to read the page to (need not be aligned)
682 * @page: The page to read
684 * On entry, the page should be locked. It will be unlocked when the page
685 * has been read. If the block driver implements rw_page synchronously,
686 * that will be true on exit from this function, but it need not be.
688 * Errors returned by this function are usually "soft", eg out of memory, or
689 * queue full; callers should try a different route to read this page rather
690 * than propagate an error back up the stack.
692 * Return: negative errno if an error occurs, 0 if submission was successful.
694 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
697 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
698 int result
= -EOPNOTSUPP
;
700 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
703 result
= blk_queue_enter(bdev
->bd_queue
, 0);
706 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
708 blk_queue_exit(bdev
->bd_queue
);
711 EXPORT_SYMBOL_GPL(bdev_read_page
);
714 * bdev_write_page() - Start writing a page to a block device
715 * @bdev: The device to write the page to
716 * @sector: The offset on the device to write the page to (need not be aligned)
717 * @page: The page to write
718 * @wbc: The writeback_control for the write
720 * On entry, the page should be locked and not currently under writeback.
721 * On exit, if the write started successfully, the page will be unlocked and
722 * under writeback. If the write failed already (eg the driver failed to
723 * queue the page to the device), the page will still be locked. If the
724 * caller is a ->writepage implementation, it will need to unlock the page.
726 * Errors returned by this function are usually "soft", eg out of memory, or
727 * queue full; callers should try a different route to write this page rather
728 * than propagate an error back up the stack.
730 * Return: negative errno if an error occurs, 0 if submission was successful.
732 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
733 struct page
*page
, struct writeback_control
*wbc
)
736 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
738 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
740 result
= blk_queue_enter(bdev
->bd_queue
, 0);
744 set_page_writeback(page
);
745 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
748 end_page_writeback(page
);
750 clean_page_buffers(page
);
753 blk_queue_exit(bdev
->bd_queue
);
756 EXPORT_SYMBOL_GPL(bdev_write_page
);
762 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
763 static struct kmem_cache
* bdev_cachep __read_mostly
;
765 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
767 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
770 return &ei
->vfs_inode
;
773 static void bdev_i_callback(struct rcu_head
*head
)
775 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
776 struct bdev_inode
*bdi
= BDEV_I(inode
);
778 kmem_cache_free(bdev_cachep
, bdi
);
781 static void bdev_destroy_inode(struct inode
*inode
)
783 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
786 static void init_once(void *foo
)
788 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
789 struct block_device
*bdev
= &ei
->bdev
;
791 memset(bdev
, 0, sizeof(*bdev
));
792 mutex_init(&bdev
->bd_mutex
);
793 INIT_LIST_HEAD(&bdev
->bd_list
);
795 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
797 bdev
->bd_bdi
= &noop_backing_dev_info
;
798 inode_init_once(&ei
->vfs_inode
);
799 /* Initialize mutex for freeze. */
800 mutex_init(&bdev
->bd_fsfreeze_mutex
);
803 static void bdev_evict_inode(struct inode
*inode
)
805 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
806 truncate_inode_pages_final(&inode
->i_data
);
807 invalidate_inode_buffers(inode
); /* is it needed here? */
809 spin_lock(&bdev_lock
);
810 list_del_init(&bdev
->bd_list
);
811 spin_unlock(&bdev_lock
);
812 /* Detach inode from wb early as bdi_put() may free bdi->wb */
813 inode_detach_wb(inode
);
814 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
815 bdi_put(bdev
->bd_bdi
);
816 bdev
->bd_bdi
= &noop_backing_dev_info
;
820 static const struct super_operations bdev_sops
= {
821 .statfs
= simple_statfs
,
822 .alloc_inode
= bdev_alloc_inode
,
823 .destroy_inode
= bdev_destroy_inode
,
824 .drop_inode
= generic_delete_inode
,
825 .evict_inode
= bdev_evict_inode
,
828 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
829 int flags
, const char *dev_name
, void *data
)
832 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
834 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
838 static struct file_system_type bd_type
= {
841 .kill_sb
= kill_anon_super
,
844 struct super_block
*blockdev_superblock __read_mostly
;
845 EXPORT_SYMBOL_GPL(blockdev_superblock
);
847 void __init
bdev_cache_init(void)
850 static struct vfsmount
*bd_mnt
;
852 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
853 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
854 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
856 err
= register_filesystem(&bd_type
);
858 panic("Cannot register bdev pseudo-fs");
859 bd_mnt
= kern_mount(&bd_type
);
861 panic("Cannot create bdev pseudo-fs");
862 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
866 * Most likely _very_ bad one - but then it's hardly critical for small
867 * /dev and can be fixed when somebody will need really large one.
868 * Keep in mind that it will be fed through icache hash function too.
870 static inline unsigned long hash(dev_t dev
)
872 return MAJOR(dev
)+MINOR(dev
);
875 static int bdev_test(struct inode
*inode
, void *data
)
877 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
880 static int bdev_set(struct inode
*inode
, void *data
)
882 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
886 static LIST_HEAD(all_bdevs
);
889 * If there is a bdev inode for this device, unhash it so that it gets evicted
890 * as soon as last inode reference is dropped.
892 void bdev_unhash_inode(dev_t dev
)
896 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
898 remove_inode_hash(inode
);
903 struct block_device
*bdget(dev_t dev
)
905 struct block_device
*bdev
;
908 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
909 bdev_test
, bdev_set
, &dev
);
914 bdev
= &BDEV_I(inode
)->bdev
;
916 if (inode
->i_state
& I_NEW
) {
917 bdev
->bd_contains
= NULL
;
918 bdev
->bd_super
= NULL
;
919 bdev
->bd_inode
= inode
;
920 bdev
->bd_block_size
= i_blocksize(inode
);
921 bdev
->bd_part_count
= 0;
922 bdev
->bd_invalidated
= 0;
923 inode
->i_mode
= S_IFBLK
;
925 inode
->i_bdev
= bdev
;
926 inode
->i_data
.a_ops
= &def_blk_aops
;
927 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
928 spin_lock(&bdev_lock
);
929 list_add(&bdev
->bd_list
, &all_bdevs
);
930 spin_unlock(&bdev_lock
);
931 unlock_new_inode(inode
);
936 EXPORT_SYMBOL(bdget
);
939 * bdgrab -- Grab a reference to an already referenced block device
940 * @bdev: Block device to grab a reference to.
942 struct block_device
*bdgrab(struct block_device
*bdev
)
944 ihold(bdev
->bd_inode
);
947 EXPORT_SYMBOL(bdgrab
);
949 long nr_blockdev_pages(void)
951 struct block_device
*bdev
;
953 spin_lock(&bdev_lock
);
954 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
955 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
957 spin_unlock(&bdev_lock
);
961 void bdput(struct block_device
*bdev
)
963 iput(bdev
->bd_inode
);
966 EXPORT_SYMBOL(bdput
);
968 static struct block_device
*bd_acquire(struct inode
*inode
)
970 struct block_device
*bdev
;
972 spin_lock(&bdev_lock
);
973 bdev
= inode
->i_bdev
;
974 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
976 spin_unlock(&bdev_lock
);
979 spin_unlock(&bdev_lock
);
982 * i_bdev references block device inode that was already shut down
983 * (corresponding device got removed). Remove the reference and look
984 * up block device inode again just in case new device got
985 * reestablished under the same device number.
990 bdev
= bdget(inode
->i_rdev
);
992 spin_lock(&bdev_lock
);
993 if (!inode
->i_bdev
) {
995 * We take an additional reference to bd_inode,
996 * and it's released in clear_inode() of inode.
997 * So, we can access it via ->i_mapping always
1001 inode
->i_bdev
= bdev
;
1002 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
1004 spin_unlock(&bdev_lock
);
1009 /* Call when you free inode */
1011 void bd_forget(struct inode
*inode
)
1013 struct block_device
*bdev
= NULL
;
1015 spin_lock(&bdev_lock
);
1016 if (!sb_is_blkdev_sb(inode
->i_sb
))
1017 bdev
= inode
->i_bdev
;
1018 inode
->i_bdev
= NULL
;
1019 inode
->i_mapping
= &inode
->i_data
;
1020 spin_unlock(&bdev_lock
);
1027 * bd_may_claim - test whether a block device can be claimed
1028 * @bdev: block device of interest
1029 * @whole: whole block device containing @bdev, may equal @bdev
1030 * @holder: holder trying to claim @bdev
1032 * Test whether @bdev can be claimed by @holder.
1035 * spin_lock(&bdev_lock).
1038 * %true if @bdev can be claimed, %false otherwise.
1040 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1043 if (bdev
->bd_holder
== holder
)
1044 return true; /* already a holder */
1045 else if (bdev
->bd_holder
!= NULL
)
1046 return false; /* held by someone else */
1047 else if (whole
== bdev
)
1048 return true; /* is a whole device which isn't held */
1050 else if (whole
->bd_holder
== bd_may_claim
)
1051 return true; /* is a partition of a device that is being partitioned */
1052 else if (whole
->bd_holder
!= NULL
)
1053 return false; /* is a partition of a held device */
1055 return true; /* is a partition of an un-held device */
1059 * bd_prepare_to_claim - prepare to claim a block device
1060 * @bdev: block device of interest
1061 * @whole: the whole device containing @bdev, may equal @bdev
1062 * @holder: holder trying to claim @bdev
1064 * Prepare to claim @bdev. This function fails if @bdev is already
1065 * claimed by another holder and waits if another claiming is in
1066 * progress. This function doesn't actually claim. On successful
1067 * return, the caller has ownership of bd_claiming and bd_holder[s].
1070 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1071 * it multiple times.
1074 * 0 if @bdev can be claimed, -EBUSY otherwise.
1076 static int bd_prepare_to_claim(struct block_device
*bdev
,
1077 struct block_device
*whole
, void *holder
)
1080 /* if someone else claimed, fail */
1081 if (!bd_may_claim(bdev
, whole
, holder
))
1084 /* if claiming is already in progress, wait for it to finish */
1085 if (whole
->bd_claiming
) {
1086 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1089 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1090 spin_unlock(&bdev_lock
);
1092 finish_wait(wq
, &wait
);
1093 spin_lock(&bdev_lock
);
1101 static struct gendisk
*bdev_get_gendisk(struct block_device
*bdev
, int *partno
)
1103 struct gendisk
*disk
= get_gendisk(bdev
->bd_dev
, partno
);
1108 * Now that we hold gendisk reference we make sure bdev we looked up is
1109 * not stale. If it is, it means device got removed and created before
1110 * we looked up gendisk and we fail open in such case. Associating
1111 * unhashed bdev with newly created gendisk could lead to two bdevs
1112 * (and thus two independent caches) being associated with one device
1115 if (inode_unhashed(bdev
->bd_inode
)) {
1116 put_disk_and_module(disk
);
1123 * bd_start_claiming - start claiming a block device
1124 * @bdev: block device of interest
1125 * @holder: holder trying to claim @bdev
1127 * @bdev is about to be opened exclusively. Check @bdev can be opened
1128 * exclusively and mark that an exclusive open is in progress. Each
1129 * successful call to this function must be matched with a call to
1130 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1133 * This function is used to gain exclusive access to the block device
1134 * without actually causing other exclusive open attempts to fail. It
1135 * should be used when the open sequence itself requires exclusive
1136 * access but may subsequently fail.
1142 * Pointer to the block device containing @bdev on success, ERR_PTR()
1145 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1148 struct gendisk
*disk
;
1149 struct block_device
*whole
;
1155 * @bdev might not have been initialized properly yet, look up
1156 * and grab the outer block device the hard way.
1158 disk
= bdev_get_gendisk(bdev
, &partno
);
1160 return ERR_PTR(-ENXIO
);
1163 * Normally, @bdev should equal what's returned from bdget_disk()
1164 * if partno is 0; however, some drivers (floppy) use multiple
1165 * bdev's for the same physical device and @bdev may be one of the
1166 * aliases. Keep @bdev if partno is 0. This means claimer
1167 * tracking is broken for those devices but it has always been that
1171 whole
= bdget_disk(disk
, 0);
1173 whole
= bdgrab(bdev
);
1175 put_disk_and_module(disk
);
1177 return ERR_PTR(-ENOMEM
);
1179 /* prepare to claim, if successful, mark claiming in progress */
1180 spin_lock(&bdev_lock
);
1182 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1184 whole
->bd_claiming
= holder
;
1185 spin_unlock(&bdev_lock
);
1188 spin_unlock(&bdev_lock
);
1190 return ERR_PTR(err
);
1195 struct bd_holder_disk
{
1196 struct list_head list
;
1197 struct gendisk
*disk
;
1201 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1202 struct gendisk
*disk
)
1204 struct bd_holder_disk
*holder
;
1206 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1207 if (holder
->disk
== disk
)
1212 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1214 return sysfs_create_link(from
, to
, kobject_name(to
));
1217 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1219 sysfs_remove_link(from
, kobject_name(to
));
1223 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1224 * @bdev: the claimed slave bdev
1225 * @disk: the holding disk
1227 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1229 * This functions creates the following sysfs symlinks.
1231 * - from "slaves" directory of the holder @disk to the claimed @bdev
1232 * - from "holders" directory of the @bdev to the holder @disk
1234 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1235 * passed to bd_link_disk_holder(), then:
1237 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1238 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1240 * The caller must have claimed @bdev before calling this function and
1241 * ensure that both @bdev and @disk are valid during the creation and
1242 * lifetime of these symlinks.
1248 * 0 on success, -errno on failure.
1250 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1252 struct bd_holder_disk
*holder
;
1255 mutex_lock(&bdev
->bd_mutex
);
1257 WARN_ON_ONCE(!bdev
->bd_holder
);
1259 /* FIXME: remove the following once add_disk() handles errors */
1260 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1263 holder
= bd_find_holder_disk(bdev
, disk
);
1269 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1275 INIT_LIST_HEAD(&holder
->list
);
1276 holder
->disk
= disk
;
1279 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1283 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1287 * bdev could be deleted beneath us which would implicitly destroy
1288 * the holder directory. Hold on to it.
1290 kobject_get(bdev
->bd_part
->holder_dir
);
1292 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1296 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1300 mutex_unlock(&bdev
->bd_mutex
);
1303 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1306 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1307 * @bdev: the calimed slave bdev
1308 * @disk: the holding disk
1310 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1315 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1317 struct bd_holder_disk
*holder
;
1319 mutex_lock(&bdev
->bd_mutex
);
1321 holder
= bd_find_holder_disk(bdev
, disk
);
1323 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1324 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1325 del_symlink(bdev
->bd_part
->holder_dir
,
1326 &disk_to_dev(disk
)->kobj
);
1327 kobject_put(bdev
->bd_part
->holder_dir
);
1328 list_del_init(&holder
->list
);
1332 mutex_unlock(&bdev
->bd_mutex
);
1334 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1338 * flush_disk - invalidates all buffer-cache entries on a disk
1340 * @bdev: struct block device to be flushed
1341 * @kill_dirty: flag to guide handling of dirty inodes
1343 * Invalidates all buffer-cache entries on a disk. It should be called
1344 * when a disk has been changed -- either by a media change or online
1347 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1349 if (__invalidate_device(bdev
, kill_dirty
)) {
1350 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1351 "resized disk %s\n",
1352 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1357 if (disk_part_scan_enabled(bdev
->bd_disk
))
1358 bdev
->bd_invalidated
= 1;
1362 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1363 * @disk: struct gendisk to check
1364 * @bdev: struct bdev to adjust.
1365 * @verbose: if %true log a message about a size change if there is any
1367 * This routine checks to see if the bdev size does not match the disk size
1368 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1371 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
,
1374 loff_t disk_size
, bdev_size
;
1376 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1377 bdev_size
= i_size_read(bdev
->bd_inode
);
1378 if (disk_size
!= bdev_size
) {
1381 "%s: detected capacity change from %lld to %lld\n",
1382 disk
->disk_name
, bdev_size
, disk_size
);
1384 i_size_write(bdev
->bd_inode
, disk_size
);
1385 if (bdev_size
> disk_size
)
1386 flush_disk(bdev
, false);
1391 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1392 * @disk: struct gendisk to be revalidated
1394 * This routine is a wrapper for lower-level driver's revalidate_disk
1395 * call-backs. It is used to do common pre and post operations needed
1396 * for all revalidate_disk operations.
1398 int revalidate_disk(struct gendisk
*disk
)
1400 struct block_device
*bdev
;
1403 if (disk
->fops
->revalidate_disk
)
1404 ret
= disk
->fops
->revalidate_disk(disk
);
1405 bdev
= bdget_disk(disk
, 0);
1409 mutex_lock(&bdev
->bd_mutex
);
1410 check_disk_size_change(disk
, bdev
, ret
== 0);
1411 bdev
->bd_invalidated
= 0;
1412 mutex_unlock(&bdev
->bd_mutex
);
1416 EXPORT_SYMBOL(revalidate_disk
);
1419 * This routine checks whether a removable media has been changed,
1420 * and invalidates all buffer-cache-entries in that case. This
1421 * is a relatively slow routine, so we have to try to minimize using
1422 * it. Thus it is called only upon a 'mount' or 'open'. This
1423 * is the best way of combining speed and utility, I think.
1424 * People changing diskettes in the middle of an operation deserve
1427 int check_disk_change(struct block_device
*bdev
)
1429 struct gendisk
*disk
= bdev
->bd_disk
;
1430 const struct block_device_operations
*bdops
= disk
->fops
;
1431 unsigned int events
;
1433 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1434 DISK_EVENT_EJECT_REQUEST
);
1435 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1438 flush_disk(bdev
, true);
1439 if (bdops
->revalidate_disk
)
1440 bdops
->revalidate_disk(bdev
->bd_disk
);
1444 EXPORT_SYMBOL(check_disk_change
);
1446 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1448 inode_lock(bdev
->bd_inode
);
1449 i_size_write(bdev
->bd_inode
, size
);
1450 inode_unlock(bdev
->bd_inode
);
1452 EXPORT_SYMBOL(bd_set_size
);
1454 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1459 * mutex_lock(part->bd_mutex)
1460 * mutex_lock_nested(whole->bd_mutex, 1)
1463 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1465 struct gendisk
*disk
;
1469 bool first_open
= false;
1471 if (mode
& FMODE_READ
)
1473 if (mode
& FMODE_WRITE
)
1476 * hooks: /n/, see "layering violations".
1479 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1489 disk
= bdev_get_gendisk(bdev
, &partno
);
1493 disk_block_events(disk
);
1494 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1495 if (!bdev
->bd_openers
) {
1497 bdev
->bd_disk
= disk
;
1498 bdev
->bd_queue
= disk
->queue
;
1499 bdev
->bd_contains
= bdev
;
1500 bdev
->bd_partno
= partno
;
1504 bdev
->bd_part
= disk_get_part(disk
, partno
);
1509 if (disk
->fops
->open
) {
1510 ret
= disk
->fops
->open(bdev
, mode
);
1511 if (ret
== -ERESTARTSYS
) {
1512 /* Lost a race with 'disk' being
1513 * deleted, try again.
1516 disk_put_part(bdev
->bd_part
);
1517 bdev
->bd_part
= NULL
;
1518 bdev
->bd_disk
= NULL
;
1519 bdev
->bd_queue
= NULL
;
1520 mutex_unlock(&bdev
->bd_mutex
);
1521 disk_unblock_events(disk
);
1522 put_disk_and_module(disk
);
1528 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1529 set_init_blocksize(bdev
);
1533 * If the device is invalidated, rescan partition
1534 * if open succeeded or failed with -ENOMEDIUM.
1535 * The latter is necessary to prevent ghost
1536 * partitions on a removed medium.
1538 if (bdev
->bd_invalidated
) {
1540 rescan_partitions(disk
, bdev
);
1541 else if (ret
== -ENOMEDIUM
)
1542 invalidate_partitions(disk
, bdev
);
1548 struct block_device
*whole
;
1549 whole
= bdget_disk(disk
, 0);
1554 ret
= __blkdev_get(whole
, mode
, 1);
1557 bdev
->bd_contains
= whole
;
1558 bdev
->bd_part
= disk_get_part(disk
, partno
);
1559 if (!(disk
->flags
& GENHD_FL_UP
) ||
1560 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1564 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1565 set_init_blocksize(bdev
);
1568 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1569 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1571 if (bdev
->bd_contains
== bdev
) {
1573 if (bdev
->bd_disk
->fops
->open
)
1574 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1575 /* the same as first opener case, read comment there */
1576 if (bdev
->bd_invalidated
) {
1578 rescan_partitions(bdev
->bd_disk
, bdev
);
1579 else if (ret
== -ENOMEDIUM
)
1580 invalidate_partitions(bdev
->bd_disk
, bdev
);
1583 goto out_unlock_bdev
;
1588 bdev
->bd_part_count
++;
1589 mutex_unlock(&bdev
->bd_mutex
);
1590 disk_unblock_events(disk
);
1591 /* only one opener holds refs to the module and disk */
1593 put_disk_and_module(disk
);
1597 disk_put_part(bdev
->bd_part
);
1598 bdev
->bd_disk
= NULL
;
1599 bdev
->bd_part
= NULL
;
1600 bdev
->bd_queue
= NULL
;
1601 if (bdev
!= bdev
->bd_contains
)
1602 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1603 bdev
->bd_contains
= NULL
;
1605 mutex_unlock(&bdev
->bd_mutex
);
1606 disk_unblock_events(disk
);
1607 put_disk_and_module(disk
);
1615 * blkdev_get - open a block device
1616 * @bdev: block_device to open
1617 * @mode: FMODE_* mask
1618 * @holder: exclusive holder identifier
1620 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1621 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1622 * @holder is invalid. Exclusive opens may nest for the same @holder.
1624 * On success, the reference count of @bdev is unchanged. On failure,
1631 * 0 on success, -errno on failure.
1633 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1635 struct block_device
*whole
= NULL
;
1638 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1640 if ((mode
& FMODE_EXCL
) && holder
) {
1641 whole
= bd_start_claiming(bdev
, holder
);
1642 if (IS_ERR(whole
)) {
1644 return PTR_ERR(whole
);
1648 res
= __blkdev_get(bdev
, mode
, 0);
1651 struct gendisk
*disk
= whole
->bd_disk
;
1653 /* finish claiming */
1654 mutex_lock(&bdev
->bd_mutex
);
1655 spin_lock(&bdev_lock
);
1658 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1660 * Note that for a whole device bd_holders
1661 * will be incremented twice, and bd_holder
1662 * will be set to bd_may_claim before being
1665 whole
->bd_holders
++;
1666 whole
->bd_holder
= bd_may_claim
;
1668 bdev
->bd_holder
= holder
;
1671 /* tell others that we're done */
1672 BUG_ON(whole
->bd_claiming
!= holder
);
1673 whole
->bd_claiming
= NULL
;
1674 wake_up_bit(&whole
->bd_claiming
, 0);
1676 spin_unlock(&bdev_lock
);
1679 * Block event polling for write claims if requested. Any
1680 * write holder makes the write_holder state stick until
1681 * all are released. This is good enough and tracking
1682 * individual writeable reference is too fragile given the
1683 * way @mode is used in blkdev_get/put().
1685 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1686 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1687 bdev
->bd_write_holder
= true;
1688 disk_block_events(disk
);
1691 mutex_unlock(&bdev
->bd_mutex
);
1697 EXPORT_SYMBOL(blkdev_get
);
1700 * blkdev_get_by_path - open a block device by name
1701 * @path: path to the block device to open
1702 * @mode: FMODE_* mask
1703 * @holder: exclusive holder identifier
1705 * Open the blockdevice described by the device file at @path. @mode
1706 * and @holder are identical to blkdev_get().
1708 * On success, the returned block_device has reference count of one.
1714 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1716 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1719 struct block_device
*bdev
;
1722 bdev
= lookup_bdev(path
);
1726 err
= blkdev_get(bdev
, mode
, holder
);
1728 return ERR_PTR(err
);
1730 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1731 blkdev_put(bdev
, mode
);
1732 return ERR_PTR(-EACCES
);
1737 EXPORT_SYMBOL(blkdev_get_by_path
);
1740 * blkdev_get_by_dev - open a block device by device number
1741 * @dev: device number of block device to open
1742 * @mode: FMODE_* mask
1743 * @holder: exclusive holder identifier
1745 * Open the blockdevice described by device number @dev. @mode and
1746 * @holder are identical to blkdev_get().
1748 * Use it ONLY if you really do not have anything better - i.e. when
1749 * you are behind a truly sucky interface and all you are given is a
1750 * device number. _Never_ to be used for internal purposes. If you
1751 * ever need it - reconsider your API.
1753 * On success, the returned block_device has reference count of one.
1759 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1761 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1763 struct block_device
*bdev
;
1768 return ERR_PTR(-ENOMEM
);
1770 err
= blkdev_get(bdev
, mode
, holder
);
1772 return ERR_PTR(err
);
1776 EXPORT_SYMBOL(blkdev_get_by_dev
);
1778 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1780 struct block_device
*bdev
;
1783 * Preserve backwards compatibility and allow large file access
1784 * even if userspace doesn't ask for it explicitly. Some mkfs
1785 * binary needs it. We might want to drop this workaround
1786 * during an unstable branch.
1788 filp
->f_flags
|= O_LARGEFILE
;
1790 filp
->f_mode
|= FMODE_NOWAIT
;
1792 if (filp
->f_flags
& O_NDELAY
)
1793 filp
->f_mode
|= FMODE_NDELAY
;
1794 if (filp
->f_flags
& O_EXCL
)
1795 filp
->f_mode
|= FMODE_EXCL
;
1796 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1797 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1799 bdev
= bd_acquire(inode
);
1803 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1804 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
1806 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1809 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1811 struct gendisk
*disk
= bdev
->bd_disk
;
1812 struct block_device
*victim
= NULL
;
1814 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1816 bdev
->bd_part_count
--;
1818 if (!--bdev
->bd_openers
) {
1819 WARN_ON_ONCE(bdev
->bd_holders
);
1820 sync_blockdev(bdev
);
1823 bdev_write_inode(bdev
);
1825 if (bdev
->bd_contains
== bdev
) {
1826 if (disk
->fops
->release
)
1827 disk
->fops
->release(disk
, mode
);
1829 if (!bdev
->bd_openers
) {
1830 disk_put_part(bdev
->bd_part
);
1831 bdev
->bd_part
= NULL
;
1832 bdev
->bd_disk
= NULL
;
1833 if (bdev
!= bdev
->bd_contains
)
1834 victim
= bdev
->bd_contains
;
1835 bdev
->bd_contains
= NULL
;
1837 put_disk_and_module(disk
);
1839 mutex_unlock(&bdev
->bd_mutex
);
1842 __blkdev_put(victim
, mode
, 1);
1845 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1847 mutex_lock(&bdev
->bd_mutex
);
1849 if (mode
& FMODE_EXCL
) {
1853 * Release a claim on the device. The holder fields
1854 * are protected with bdev_lock. bd_mutex is to
1855 * synchronize disk_holder unlinking.
1857 spin_lock(&bdev_lock
);
1859 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1860 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1862 /* bd_contains might point to self, check in a separate step */
1863 if ((bdev_free
= !bdev
->bd_holders
))
1864 bdev
->bd_holder
= NULL
;
1865 if (!bdev
->bd_contains
->bd_holders
)
1866 bdev
->bd_contains
->bd_holder
= NULL
;
1868 spin_unlock(&bdev_lock
);
1871 * If this was the last claim, remove holder link and
1872 * unblock evpoll if it was a write holder.
1874 if (bdev_free
&& bdev
->bd_write_holder
) {
1875 disk_unblock_events(bdev
->bd_disk
);
1876 bdev
->bd_write_holder
= false;
1881 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1882 * event. This is to ensure detection of media removal commanded
1883 * from userland - e.g. eject(1).
1885 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1887 mutex_unlock(&bdev
->bd_mutex
);
1889 __blkdev_put(bdev
, mode
, 0);
1891 EXPORT_SYMBOL(blkdev_put
);
1893 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1895 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1896 blkdev_put(bdev
, filp
->f_mode
);
1900 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1902 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1903 fmode_t mode
= file
->f_mode
;
1906 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1907 * to updated it before every ioctl.
1909 if (file
->f_flags
& O_NDELAY
)
1910 mode
|= FMODE_NDELAY
;
1912 mode
&= ~FMODE_NDELAY
;
1914 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1918 * Write data to the block device. Only intended for the block device itself
1919 * and the raw driver which basically is a fake block device.
1921 * Does not take i_mutex for the write and thus is not for general purpose
1924 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1926 struct file
*file
= iocb
->ki_filp
;
1927 struct inode
*bd_inode
= bdev_file_inode(file
);
1928 loff_t size
= i_size_read(bd_inode
);
1929 struct blk_plug plug
;
1932 if (bdev_read_only(I_BDEV(bd_inode
)))
1935 if (!iov_iter_count(from
))
1938 if (iocb
->ki_pos
>= size
)
1941 if ((iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_DIRECT
)) == IOCB_NOWAIT
)
1944 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1946 blk_start_plug(&plug
);
1947 ret
= __generic_file_write_iter(iocb
, from
);
1949 ret
= generic_write_sync(iocb
, ret
);
1950 blk_finish_plug(&plug
);
1953 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1955 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1957 struct file
*file
= iocb
->ki_filp
;
1958 struct inode
*bd_inode
= bdev_file_inode(file
);
1959 loff_t size
= i_size_read(bd_inode
);
1960 loff_t pos
= iocb
->ki_pos
;
1966 iov_iter_truncate(to
, size
);
1967 return generic_file_read_iter(iocb
, to
);
1969 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1972 * Try to release a page associated with block device when the system
1973 * is under memory pressure.
1975 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1977 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1979 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1980 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1982 return try_to_free_buffers(page
);
1985 static int blkdev_writepages(struct address_space
*mapping
,
1986 struct writeback_control
*wbc
)
1988 return generic_writepages(mapping
, wbc
);
1991 static const struct address_space_operations def_blk_aops
= {
1992 .readpage
= blkdev_readpage
,
1993 .readpages
= blkdev_readpages
,
1994 .writepage
= blkdev_writepage
,
1995 .write_begin
= blkdev_write_begin
,
1996 .write_end
= blkdev_write_end
,
1997 .writepages
= blkdev_writepages
,
1998 .releasepage
= blkdev_releasepage
,
1999 .direct_IO
= blkdev_direct_IO
,
2000 .migratepage
= buffer_migrate_page_norefs
,
2001 .is_dirty_writeback
= buffer_check_dirty_writeback
,
2004 #define BLKDEV_FALLOC_FL_SUPPORTED \
2005 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2006 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2008 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
2011 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
2012 struct address_space
*mapping
;
2013 loff_t end
= start
+ len
- 1;
2017 /* Fail if we don't recognize the flags. */
2018 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
2021 /* Don't go off the end of the device. */
2022 isize
= i_size_read(bdev
->bd_inode
);
2026 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2027 len
= isize
- start
;
2028 end
= start
+ len
- 1;
2034 * Don't allow IO that isn't aligned to logical block size.
2036 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2039 /* Invalidate the page cache, including dirty pages. */
2040 mapping
= bdev
->bd_inode
->i_mapping
;
2041 truncate_inode_pages_range(mapping
, start
, end
);
2044 case FALLOC_FL_ZERO_RANGE
:
2045 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2046 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2047 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
2049 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2050 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2051 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
2053 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2054 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2064 * Invalidate again; if someone wandered in and dirtied a page,
2065 * the caller will be given -EBUSY. The third argument is
2066 * inclusive, so the rounding here is safe.
2068 return invalidate_inode_pages2_range(mapping
,
2069 start
>> PAGE_SHIFT
,
2073 const struct file_operations def_blk_fops
= {
2074 .open
= blkdev_open
,
2075 .release
= blkdev_close
,
2076 .llseek
= block_llseek
,
2077 .read_iter
= blkdev_read_iter
,
2078 .write_iter
= blkdev_write_iter
,
2079 .mmap
= generic_file_mmap
,
2080 .fsync
= blkdev_fsync
,
2081 .unlocked_ioctl
= block_ioctl
,
2082 #ifdef CONFIG_COMPAT
2083 .compat_ioctl
= compat_blkdev_ioctl
,
2085 .splice_read
= generic_file_splice_read
,
2086 .splice_write
= iter_file_splice_write
,
2087 .fallocate
= blkdev_fallocate
,
2090 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2093 mm_segment_t old_fs
= get_fs();
2095 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2100 EXPORT_SYMBOL(ioctl_by_bdev
);
2103 * lookup_bdev - lookup a struct block_device by name
2104 * @pathname: special file representing the block device
2106 * Get a reference to the blockdevice at @pathname in the current
2107 * namespace if possible and return it. Return ERR_PTR(error)
2110 struct block_device
*lookup_bdev(const char *pathname
)
2112 struct block_device
*bdev
;
2113 struct inode
*inode
;
2117 if (!pathname
|| !*pathname
)
2118 return ERR_PTR(-EINVAL
);
2120 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2122 return ERR_PTR(error
);
2124 inode
= d_backing_inode(path
.dentry
);
2126 if (!S_ISBLK(inode
->i_mode
))
2129 if (!may_open_dev(&path
))
2132 bdev
= bd_acquire(inode
);
2139 bdev
= ERR_PTR(error
);
2142 EXPORT_SYMBOL(lookup_bdev
);
2144 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2146 struct super_block
*sb
= get_super(bdev
);
2151 * no need to lock the super, get_super holds the
2152 * read mutex so the filesystem cannot go away
2153 * under us (->put_super runs with the write lock
2156 shrink_dcache_sb(sb
);
2157 res
= invalidate_inodes(sb
, kill_dirty
);
2160 invalidate_bdev(bdev
);
2163 EXPORT_SYMBOL(__invalidate_device
);
2165 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2167 struct inode
*inode
, *old_inode
= NULL
;
2169 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2170 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2171 struct address_space
*mapping
= inode
->i_mapping
;
2172 struct block_device
*bdev
;
2174 spin_lock(&inode
->i_lock
);
2175 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2176 mapping
->nrpages
== 0) {
2177 spin_unlock(&inode
->i_lock
);
2181 spin_unlock(&inode
->i_lock
);
2182 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2184 * We hold a reference to 'inode' so it couldn't have been
2185 * removed from s_inodes list while we dropped the
2186 * s_inode_list_lock We cannot iput the inode now as we can
2187 * be holding the last reference and we cannot iput it under
2188 * s_inode_list_lock. So we keep the reference and iput it
2193 bdev
= I_BDEV(inode
);
2195 mutex_lock(&bdev
->bd_mutex
);
2196 if (bdev
->bd_openers
)
2198 mutex_unlock(&bdev
->bd_mutex
);
2200 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2202 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);