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;
214 struct bvec_iter_all iter_all
;
216 if ((pos
| iov_iter_alignment(iter
)) &
217 (bdev_logical_block_size(bdev
) - 1))
220 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
223 vecs
= kmalloc_array(nr_pages
, sizeof(struct bio_vec
),
229 bio_init(&bio
, vecs
, nr_pages
);
230 bio_set_dev(&bio
, bdev
);
231 bio
.bi_iter
.bi_sector
= pos
>> 9;
232 bio
.bi_write_hint
= iocb
->ki_hint
;
233 bio
.bi_private
= current
;
234 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
235 bio
.bi_ioprio
= iocb
->ki_ioprio
;
237 ret
= bio_iov_iter_get_pages(&bio
, iter
);
240 ret
= bio
.bi_iter
.bi_size
;
242 if (iov_iter_rw(iter
) == READ
) {
243 bio
.bi_opf
= REQ_OP_READ
;
244 if (iter_is_iovec(iter
))
247 bio
.bi_opf
= dio_bio_write_op(iocb
);
248 task_io_account_write(ret
);
250 if (iocb
->ki_flags
& IOCB_HIPRI
)
251 bio
.bi_opf
|= REQ_HIPRI
;
253 qc
= submit_bio(&bio
);
255 set_current_state(TASK_UNINTERRUPTIBLE
);
256 if (!READ_ONCE(bio
.bi_private
))
258 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
259 !blk_poll(bdev_get_queue(bdev
), qc
, true))
262 __set_current_state(TASK_RUNNING
);
264 bio_for_each_segment_all(bvec
, &bio
, i
, iter_all
) {
265 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
266 set_page_dirty_lock(bvec
->bv_page
);
267 put_page(bvec
->bv_page
);
270 if (unlikely(bio
.bi_status
))
271 ret
= blk_status_to_errno(bio
.bi_status
);
274 if (vecs
!= inline_vecs
)
285 struct task_struct
*waiter
;
290 bool should_dirty
: 1;
295 static struct bio_set blkdev_dio_pool
;
297 static void blkdev_bio_end_io(struct bio
*bio
)
299 struct blkdev_dio
*dio
= bio
->bi_private
;
300 bool should_dirty
= dio
->should_dirty
;
302 if (dio
->multi_bio
&& !atomic_dec_and_test(&dio
->ref
)) {
303 if (bio
->bi_status
&& !dio
->bio
.bi_status
)
304 dio
->bio
.bi_status
= bio
->bi_status
;
307 struct kiocb
*iocb
= dio
->iocb
;
310 if (likely(!dio
->bio
.bi_status
)) {
314 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
317 dio
->iocb
->ki_complete(iocb
, ret
, 0);
321 struct task_struct
*waiter
= dio
->waiter
;
323 WRITE_ONCE(dio
->waiter
, NULL
);
324 blk_wake_io_task(waiter
);
329 bio_check_pages_dirty(bio
);
331 struct bio_vec
*bvec
;
333 struct bvec_iter_all iter_all
;
335 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
)
336 put_page(bvec
->bv_page
);
342 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
344 struct file
*file
= iocb
->ki_filp
;
345 struct inode
*inode
= bdev_file_inode(file
);
346 struct block_device
*bdev
= I_BDEV(inode
);
347 struct blk_plug plug
;
348 struct blkdev_dio
*dio
;
350 bool is_poll
= (iocb
->ki_flags
& IOCB_HIPRI
) != 0;
351 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
352 loff_t pos
= iocb
->ki_pos
;
353 blk_qc_t qc
= BLK_QC_T_NONE
;
356 if ((pos
| iov_iter_alignment(iter
)) &
357 (bdev_logical_block_size(bdev
) - 1))
360 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, &blkdev_dio_pool
);
362 dio
= container_of(bio
, struct blkdev_dio
, bio
);
363 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
365 dio
->waiter
= current
;
372 dio
->multi_bio
= false;
373 dio
->should_dirty
= is_read
&& iter_is_iovec(iter
);
376 * Don't plug for HIPRI/polled IO, as those should go straight
380 blk_start_plug(&plug
);
383 bio_set_dev(bio
, bdev
);
384 bio
->bi_iter
.bi_sector
= pos
>> 9;
385 bio
->bi_write_hint
= iocb
->ki_hint
;
386 bio
->bi_private
= dio
;
387 bio
->bi_end_io
= blkdev_bio_end_io
;
388 bio
->bi_ioprio
= iocb
->ki_ioprio
;
390 ret
= bio_iov_iter_get_pages(bio
, iter
);
392 bio
->bi_status
= BLK_STS_IOERR
;
398 bio
->bi_opf
= REQ_OP_READ
;
399 if (dio
->should_dirty
)
400 bio_set_pages_dirty(bio
);
402 bio
->bi_opf
= dio_bio_write_op(iocb
);
403 task_io_account_write(bio
->bi_iter
.bi_size
);
406 dio
->size
+= bio
->bi_iter
.bi_size
;
407 pos
+= bio
->bi_iter
.bi_size
;
409 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
411 if (iocb
->ki_flags
& IOCB_HIPRI
)
412 bio
->bi_opf
|= REQ_HIPRI
;
414 qc
= submit_bio(bio
);
418 if (!dio
->multi_bio
) {
420 * AIO needs an extra reference to ensure the dio
421 * structure which is embedded into the first bio
426 dio
->multi_bio
= true;
427 atomic_set(&dio
->ref
, 2);
429 atomic_inc(&dio
->ref
);
433 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
437 blk_finish_plug(&plug
);
443 set_current_state(TASK_UNINTERRUPTIBLE
);
444 if (!READ_ONCE(dio
->waiter
))
447 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
448 !blk_poll(bdev_get_queue(bdev
), qc
, true))
451 __set_current_state(TASK_RUNNING
);
454 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
463 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
467 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
470 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
471 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
473 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
476 static __init
int blkdev_init(void)
478 return bioset_init(&blkdev_dio_pool
, 4, offsetof(struct blkdev_dio
, bio
), BIOSET_NEED_BVECS
);
480 module_init(blkdev_init
);
482 int __sync_blockdev(struct block_device
*bdev
, int wait
)
487 return filemap_flush(bdev
->bd_inode
->i_mapping
);
488 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
492 * Write out and wait upon all the dirty data associated with a block
493 * device via its mapping. Does not take the superblock lock.
495 int sync_blockdev(struct block_device
*bdev
)
497 return __sync_blockdev(bdev
, 1);
499 EXPORT_SYMBOL(sync_blockdev
);
502 * Write out and wait upon all dirty data associated with this
503 * device. Filesystem data as well as the underlying block
504 * device. Takes the superblock lock.
506 int fsync_bdev(struct block_device
*bdev
)
508 struct super_block
*sb
= get_super(bdev
);
510 int res
= sync_filesystem(sb
);
514 return sync_blockdev(bdev
);
516 EXPORT_SYMBOL(fsync_bdev
);
519 * freeze_bdev -- lock a filesystem and force it into a consistent state
520 * @bdev: blockdevice to lock
522 * If a superblock is found on this device, we take the s_umount semaphore
523 * on it to make sure nobody unmounts until the snapshot creation is done.
524 * The reference counter (bd_fsfreeze_count) guarantees that only the last
525 * unfreeze process can unfreeze the frozen filesystem actually when multiple
526 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
527 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
530 struct super_block
*freeze_bdev(struct block_device
*bdev
)
532 struct super_block
*sb
;
535 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
536 if (++bdev
->bd_fsfreeze_count
> 1) {
538 * We don't even need to grab a reference - the first call
539 * to freeze_bdev grab an active reference and only the last
540 * thaw_bdev drops it.
542 sb
= get_super(bdev
);
545 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
549 sb
= get_active_super(bdev
);
552 if (sb
->s_op
->freeze_super
)
553 error
= sb
->s_op
->freeze_super(sb
);
555 error
= freeze_super(sb
);
557 deactivate_super(sb
);
558 bdev
->bd_fsfreeze_count
--;
559 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
560 return ERR_PTR(error
);
562 deactivate_super(sb
);
565 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
566 return sb
; /* thaw_bdev releases s->s_umount */
568 EXPORT_SYMBOL(freeze_bdev
);
571 * thaw_bdev -- unlock filesystem
572 * @bdev: blockdevice to unlock
573 * @sb: associated superblock
575 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
577 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
581 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
582 if (!bdev
->bd_fsfreeze_count
)
586 if (--bdev
->bd_fsfreeze_count
> 0)
592 if (sb
->s_op
->thaw_super
)
593 error
= sb
->s_op
->thaw_super(sb
);
595 error
= thaw_super(sb
);
597 bdev
->bd_fsfreeze_count
++;
599 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
602 EXPORT_SYMBOL(thaw_bdev
);
604 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
606 return block_write_full_page(page
, blkdev_get_block
, wbc
);
609 static int blkdev_readpage(struct file
* file
, struct page
* page
)
611 return block_read_full_page(page
, blkdev_get_block
);
614 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
615 struct list_head
*pages
, unsigned nr_pages
)
617 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
620 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
621 loff_t pos
, unsigned len
, unsigned flags
,
622 struct page
**pagep
, void **fsdata
)
624 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
628 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
629 loff_t pos
, unsigned len
, unsigned copied
,
630 struct page
*page
, void *fsdata
)
633 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
643 * for a block special file file_inode(file)->i_size is zero
644 * so we compute the size by hand (just as in block_read/write above)
646 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
648 struct inode
*bd_inode
= bdev_file_inode(file
);
651 inode_lock(bd_inode
);
652 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
653 inode_unlock(bd_inode
);
657 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
659 struct inode
*bd_inode
= bdev_file_inode(filp
);
660 struct block_device
*bdev
= I_BDEV(bd_inode
);
663 error
= file_write_and_wait_range(filp
, start
, end
);
668 * There is no need to serialise calls to blkdev_issue_flush with
669 * i_mutex and doing so causes performance issues with concurrent
670 * O_SYNC writers to a block device.
672 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
673 if (error
== -EOPNOTSUPP
)
678 EXPORT_SYMBOL(blkdev_fsync
);
681 * bdev_read_page() - Start reading a page from a block device
682 * @bdev: The device to read the page from
683 * @sector: The offset on the device to read the page to (need not be aligned)
684 * @page: The page to read
686 * On entry, the page should be locked. It will be unlocked when the page
687 * has been read. If the block driver implements rw_page synchronously,
688 * that will be true on exit from this function, but it need not be.
690 * Errors returned by this function are usually "soft", eg out of memory, or
691 * queue full; callers should try a different route to read this page rather
692 * than propagate an error back up the stack.
694 * Return: negative errno if an error occurs, 0 if submission was successful.
696 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
699 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
700 int result
= -EOPNOTSUPP
;
702 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
705 result
= blk_queue_enter(bdev
->bd_queue
, 0);
708 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
710 blk_queue_exit(bdev
->bd_queue
);
713 EXPORT_SYMBOL_GPL(bdev_read_page
);
716 * bdev_write_page() - Start writing a page to a block device
717 * @bdev: The device to write the page to
718 * @sector: The offset on the device to write the page to (need not be aligned)
719 * @page: The page to write
720 * @wbc: The writeback_control for the write
722 * On entry, the page should be locked and not currently under writeback.
723 * On exit, if the write started successfully, the page will be unlocked and
724 * under writeback. If the write failed already (eg the driver failed to
725 * queue the page to the device), the page will still be locked. If the
726 * caller is a ->writepage implementation, it will need to unlock the page.
728 * Errors returned by this function are usually "soft", eg out of memory, or
729 * queue full; callers should try a different route to write this page rather
730 * than propagate an error back up the stack.
732 * Return: negative errno if an error occurs, 0 if submission was successful.
734 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
735 struct page
*page
, struct writeback_control
*wbc
)
738 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
740 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
742 result
= blk_queue_enter(bdev
->bd_queue
, 0);
746 set_page_writeback(page
);
747 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
750 end_page_writeback(page
);
752 clean_page_buffers(page
);
755 blk_queue_exit(bdev
->bd_queue
);
758 EXPORT_SYMBOL_GPL(bdev_write_page
);
764 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
765 static struct kmem_cache
* bdev_cachep __read_mostly
;
767 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
769 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
772 return &ei
->vfs_inode
;
775 static void bdev_i_callback(struct rcu_head
*head
)
777 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
778 struct bdev_inode
*bdi
= BDEV_I(inode
);
780 kmem_cache_free(bdev_cachep
, bdi
);
783 static void bdev_destroy_inode(struct inode
*inode
)
785 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
788 static void init_once(void *foo
)
790 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
791 struct block_device
*bdev
= &ei
->bdev
;
793 memset(bdev
, 0, sizeof(*bdev
));
794 mutex_init(&bdev
->bd_mutex
);
795 INIT_LIST_HEAD(&bdev
->bd_list
);
797 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
799 bdev
->bd_bdi
= &noop_backing_dev_info
;
800 inode_init_once(&ei
->vfs_inode
);
801 /* Initialize mutex for freeze. */
802 mutex_init(&bdev
->bd_fsfreeze_mutex
);
805 static void bdev_evict_inode(struct inode
*inode
)
807 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
808 truncate_inode_pages_final(&inode
->i_data
);
809 invalidate_inode_buffers(inode
); /* is it needed here? */
811 spin_lock(&bdev_lock
);
812 list_del_init(&bdev
->bd_list
);
813 spin_unlock(&bdev_lock
);
814 /* Detach inode from wb early as bdi_put() may free bdi->wb */
815 inode_detach_wb(inode
);
816 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
817 bdi_put(bdev
->bd_bdi
);
818 bdev
->bd_bdi
= &noop_backing_dev_info
;
822 static const struct super_operations bdev_sops
= {
823 .statfs
= simple_statfs
,
824 .alloc_inode
= bdev_alloc_inode
,
825 .destroy_inode
= bdev_destroy_inode
,
826 .drop_inode
= generic_delete_inode
,
827 .evict_inode
= bdev_evict_inode
,
830 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
831 int flags
, const char *dev_name
, void *data
)
834 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
836 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
840 static struct file_system_type bd_type
= {
843 .kill_sb
= kill_anon_super
,
846 struct super_block
*blockdev_superblock __read_mostly
;
847 EXPORT_SYMBOL_GPL(blockdev_superblock
);
849 void __init
bdev_cache_init(void)
852 static struct vfsmount
*bd_mnt
;
854 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
855 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
856 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
858 err
= register_filesystem(&bd_type
);
860 panic("Cannot register bdev pseudo-fs");
861 bd_mnt
= kern_mount(&bd_type
);
863 panic("Cannot create bdev pseudo-fs");
864 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
868 * Most likely _very_ bad one - but then it's hardly critical for small
869 * /dev and can be fixed when somebody will need really large one.
870 * Keep in mind that it will be fed through icache hash function too.
872 static inline unsigned long hash(dev_t dev
)
874 return MAJOR(dev
)+MINOR(dev
);
877 static int bdev_test(struct inode
*inode
, void *data
)
879 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
882 static int bdev_set(struct inode
*inode
, void *data
)
884 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
888 static LIST_HEAD(all_bdevs
);
891 * If there is a bdev inode for this device, unhash it so that it gets evicted
892 * as soon as last inode reference is dropped.
894 void bdev_unhash_inode(dev_t dev
)
898 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
900 remove_inode_hash(inode
);
905 struct block_device
*bdget(dev_t dev
)
907 struct block_device
*bdev
;
910 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
911 bdev_test
, bdev_set
, &dev
);
916 bdev
= &BDEV_I(inode
)->bdev
;
918 if (inode
->i_state
& I_NEW
) {
919 bdev
->bd_contains
= NULL
;
920 bdev
->bd_super
= NULL
;
921 bdev
->bd_inode
= inode
;
922 bdev
->bd_block_size
= i_blocksize(inode
);
923 bdev
->bd_part_count
= 0;
924 bdev
->bd_invalidated
= 0;
925 inode
->i_mode
= S_IFBLK
;
927 inode
->i_bdev
= bdev
;
928 inode
->i_data
.a_ops
= &def_blk_aops
;
929 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
930 spin_lock(&bdev_lock
);
931 list_add(&bdev
->bd_list
, &all_bdevs
);
932 spin_unlock(&bdev_lock
);
933 unlock_new_inode(inode
);
938 EXPORT_SYMBOL(bdget
);
941 * bdgrab -- Grab a reference to an already referenced block device
942 * @bdev: Block device to grab a reference to.
944 struct block_device
*bdgrab(struct block_device
*bdev
)
946 ihold(bdev
->bd_inode
);
949 EXPORT_SYMBOL(bdgrab
);
951 long nr_blockdev_pages(void)
953 struct block_device
*bdev
;
955 spin_lock(&bdev_lock
);
956 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
957 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
959 spin_unlock(&bdev_lock
);
963 void bdput(struct block_device
*bdev
)
965 iput(bdev
->bd_inode
);
968 EXPORT_SYMBOL(bdput
);
970 static struct block_device
*bd_acquire(struct inode
*inode
)
972 struct block_device
*bdev
;
974 spin_lock(&bdev_lock
);
975 bdev
= inode
->i_bdev
;
976 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
978 spin_unlock(&bdev_lock
);
981 spin_unlock(&bdev_lock
);
984 * i_bdev references block device inode that was already shut down
985 * (corresponding device got removed). Remove the reference and look
986 * up block device inode again just in case new device got
987 * reestablished under the same device number.
992 bdev
= bdget(inode
->i_rdev
);
994 spin_lock(&bdev_lock
);
995 if (!inode
->i_bdev
) {
997 * We take an additional reference to bd_inode,
998 * and it's released in clear_inode() of inode.
999 * So, we can access it via ->i_mapping always
1003 inode
->i_bdev
= bdev
;
1004 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
1006 spin_unlock(&bdev_lock
);
1011 /* Call when you free inode */
1013 void bd_forget(struct inode
*inode
)
1015 struct block_device
*bdev
= NULL
;
1017 spin_lock(&bdev_lock
);
1018 if (!sb_is_blkdev_sb(inode
->i_sb
))
1019 bdev
= inode
->i_bdev
;
1020 inode
->i_bdev
= NULL
;
1021 inode
->i_mapping
= &inode
->i_data
;
1022 spin_unlock(&bdev_lock
);
1029 * bd_may_claim - test whether a block device can be claimed
1030 * @bdev: block device of interest
1031 * @whole: whole block device containing @bdev, may equal @bdev
1032 * @holder: holder trying to claim @bdev
1034 * Test whether @bdev can be claimed by @holder.
1037 * spin_lock(&bdev_lock).
1040 * %true if @bdev can be claimed, %false otherwise.
1042 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1045 if (bdev
->bd_holder
== holder
)
1046 return true; /* already a holder */
1047 else if (bdev
->bd_holder
!= NULL
)
1048 return false; /* held by someone else */
1049 else if (whole
== bdev
)
1050 return true; /* is a whole device which isn't held */
1052 else if (whole
->bd_holder
== bd_may_claim
)
1053 return true; /* is a partition of a device that is being partitioned */
1054 else if (whole
->bd_holder
!= NULL
)
1055 return false; /* is a partition of a held device */
1057 return true; /* is a partition of an un-held device */
1061 * bd_prepare_to_claim - prepare to claim a block device
1062 * @bdev: block device of interest
1063 * @whole: the whole device containing @bdev, may equal @bdev
1064 * @holder: holder trying to claim @bdev
1066 * Prepare to claim @bdev. This function fails if @bdev is already
1067 * claimed by another holder and waits if another claiming is in
1068 * progress. This function doesn't actually claim. On successful
1069 * return, the caller has ownership of bd_claiming and bd_holder[s].
1072 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1073 * it multiple times.
1076 * 0 if @bdev can be claimed, -EBUSY otherwise.
1078 static int bd_prepare_to_claim(struct block_device
*bdev
,
1079 struct block_device
*whole
, void *holder
)
1082 /* if someone else claimed, fail */
1083 if (!bd_may_claim(bdev
, whole
, holder
))
1086 /* if claiming is already in progress, wait for it to finish */
1087 if (whole
->bd_claiming
) {
1088 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1091 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1092 spin_unlock(&bdev_lock
);
1094 finish_wait(wq
, &wait
);
1095 spin_lock(&bdev_lock
);
1103 static struct gendisk
*bdev_get_gendisk(struct block_device
*bdev
, int *partno
)
1105 struct gendisk
*disk
= get_gendisk(bdev
->bd_dev
, partno
);
1110 * Now that we hold gendisk reference we make sure bdev we looked up is
1111 * not stale. If it is, it means device got removed and created before
1112 * we looked up gendisk and we fail open in such case. Associating
1113 * unhashed bdev with newly created gendisk could lead to two bdevs
1114 * (and thus two independent caches) being associated with one device
1117 if (inode_unhashed(bdev
->bd_inode
)) {
1118 put_disk_and_module(disk
);
1125 * bd_start_claiming - start claiming a block device
1126 * @bdev: block device of interest
1127 * @holder: holder trying to claim @bdev
1129 * @bdev is about to be opened exclusively. Check @bdev can be opened
1130 * exclusively and mark that an exclusive open is in progress. Each
1131 * successful call to this function must be matched with a call to
1132 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1135 * This function is used to gain exclusive access to the block device
1136 * without actually causing other exclusive open attempts to fail. It
1137 * should be used when the open sequence itself requires exclusive
1138 * access but may subsequently fail.
1144 * Pointer to the block device containing @bdev on success, ERR_PTR()
1147 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1150 struct gendisk
*disk
;
1151 struct block_device
*whole
;
1157 * @bdev might not have been initialized properly yet, look up
1158 * and grab the outer block device the hard way.
1160 disk
= bdev_get_gendisk(bdev
, &partno
);
1162 return ERR_PTR(-ENXIO
);
1165 * Normally, @bdev should equal what's returned from bdget_disk()
1166 * if partno is 0; however, some drivers (floppy) use multiple
1167 * bdev's for the same physical device and @bdev may be one of the
1168 * aliases. Keep @bdev if partno is 0. This means claimer
1169 * tracking is broken for those devices but it has always been that
1173 whole
= bdget_disk(disk
, 0);
1175 whole
= bdgrab(bdev
);
1177 put_disk_and_module(disk
);
1179 return ERR_PTR(-ENOMEM
);
1181 /* prepare to claim, if successful, mark claiming in progress */
1182 spin_lock(&bdev_lock
);
1184 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1186 whole
->bd_claiming
= holder
;
1187 spin_unlock(&bdev_lock
);
1190 spin_unlock(&bdev_lock
);
1192 return ERR_PTR(err
);
1197 struct bd_holder_disk
{
1198 struct list_head list
;
1199 struct gendisk
*disk
;
1203 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1204 struct gendisk
*disk
)
1206 struct bd_holder_disk
*holder
;
1208 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1209 if (holder
->disk
== disk
)
1214 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1216 return sysfs_create_link(from
, to
, kobject_name(to
));
1219 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1221 sysfs_remove_link(from
, kobject_name(to
));
1225 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1226 * @bdev: the claimed slave bdev
1227 * @disk: the holding disk
1229 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1231 * This functions creates the following sysfs symlinks.
1233 * - from "slaves" directory of the holder @disk to the claimed @bdev
1234 * - from "holders" directory of the @bdev to the holder @disk
1236 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1237 * passed to bd_link_disk_holder(), then:
1239 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1240 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1242 * The caller must have claimed @bdev before calling this function and
1243 * ensure that both @bdev and @disk are valid during the creation and
1244 * lifetime of these symlinks.
1250 * 0 on success, -errno on failure.
1252 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1254 struct bd_holder_disk
*holder
;
1257 mutex_lock(&bdev
->bd_mutex
);
1259 WARN_ON_ONCE(!bdev
->bd_holder
);
1261 /* FIXME: remove the following once add_disk() handles errors */
1262 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1265 holder
= bd_find_holder_disk(bdev
, disk
);
1271 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1277 INIT_LIST_HEAD(&holder
->list
);
1278 holder
->disk
= disk
;
1281 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1285 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1289 * bdev could be deleted beneath us which would implicitly destroy
1290 * the holder directory. Hold on to it.
1292 kobject_get(bdev
->bd_part
->holder_dir
);
1294 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1298 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1302 mutex_unlock(&bdev
->bd_mutex
);
1305 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1308 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1309 * @bdev: the calimed slave bdev
1310 * @disk: the holding disk
1312 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1317 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1319 struct bd_holder_disk
*holder
;
1321 mutex_lock(&bdev
->bd_mutex
);
1323 holder
= bd_find_holder_disk(bdev
, disk
);
1325 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1326 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1327 del_symlink(bdev
->bd_part
->holder_dir
,
1328 &disk_to_dev(disk
)->kobj
);
1329 kobject_put(bdev
->bd_part
->holder_dir
);
1330 list_del_init(&holder
->list
);
1334 mutex_unlock(&bdev
->bd_mutex
);
1336 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1340 * flush_disk - invalidates all buffer-cache entries on a disk
1342 * @bdev: struct block device to be flushed
1343 * @kill_dirty: flag to guide handling of dirty inodes
1345 * Invalidates all buffer-cache entries on a disk. It should be called
1346 * when a disk has been changed -- either by a media change or online
1349 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1351 if (__invalidate_device(bdev
, kill_dirty
)) {
1352 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1353 "resized disk %s\n",
1354 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1359 if (disk_part_scan_enabled(bdev
->bd_disk
))
1360 bdev
->bd_invalidated
= 1;
1364 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1365 * @disk: struct gendisk to check
1366 * @bdev: struct bdev to adjust.
1367 * @verbose: if %true log a message about a size change if there is any
1369 * This routine checks to see if the bdev size does not match the disk size
1370 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1373 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
,
1376 loff_t disk_size
, bdev_size
;
1378 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1379 bdev_size
= i_size_read(bdev
->bd_inode
);
1380 if (disk_size
!= bdev_size
) {
1383 "%s: detected capacity change from %lld to %lld\n",
1384 disk
->disk_name
, bdev_size
, disk_size
);
1386 i_size_write(bdev
->bd_inode
, disk_size
);
1387 if (bdev_size
> disk_size
)
1388 flush_disk(bdev
, false);
1393 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1394 * @disk: struct gendisk to be revalidated
1396 * This routine is a wrapper for lower-level driver's revalidate_disk
1397 * call-backs. It is used to do common pre and post operations needed
1398 * for all revalidate_disk operations.
1400 int revalidate_disk(struct gendisk
*disk
)
1402 struct block_device
*bdev
;
1405 if (disk
->fops
->revalidate_disk
)
1406 ret
= disk
->fops
->revalidate_disk(disk
);
1407 bdev
= bdget_disk(disk
, 0);
1411 mutex_lock(&bdev
->bd_mutex
);
1412 check_disk_size_change(disk
, bdev
, ret
== 0);
1413 bdev
->bd_invalidated
= 0;
1414 mutex_unlock(&bdev
->bd_mutex
);
1418 EXPORT_SYMBOL(revalidate_disk
);
1421 * This routine checks whether a removable media has been changed,
1422 * and invalidates all buffer-cache-entries in that case. This
1423 * is a relatively slow routine, so we have to try to minimize using
1424 * it. Thus it is called only upon a 'mount' or 'open'. This
1425 * is the best way of combining speed and utility, I think.
1426 * People changing diskettes in the middle of an operation deserve
1429 int check_disk_change(struct block_device
*bdev
)
1431 struct gendisk
*disk
= bdev
->bd_disk
;
1432 const struct block_device_operations
*bdops
= disk
->fops
;
1433 unsigned int events
;
1435 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1436 DISK_EVENT_EJECT_REQUEST
);
1437 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1440 flush_disk(bdev
, true);
1441 if (bdops
->revalidate_disk
)
1442 bdops
->revalidate_disk(bdev
->bd_disk
);
1446 EXPORT_SYMBOL(check_disk_change
);
1448 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1450 inode_lock(bdev
->bd_inode
);
1451 i_size_write(bdev
->bd_inode
, size
);
1452 inode_unlock(bdev
->bd_inode
);
1454 EXPORT_SYMBOL(bd_set_size
);
1456 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1461 * mutex_lock(part->bd_mutex)
1462 * mutex_lock_nested(whole->bd_mutex, 1)
1465 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1467 struct gendisk
*disk
;
1471 bool first_open
= false;
1473 if (mode
& FMODE_READ
)
1475 if (mode
& FMODE_WRITE
)
1478 * hooks: /n/, see "layering violations".
1481 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1491 disk
= bdev_get_gendisk(bdev
, &partno
);
1495 disk_block_events(disk
);
1496 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1497 if (!bdev
->bd_openers
) {
1499 bdev
->bd_disk
= disk
;
1500 bdev
->bd_queue
= disk
->queue
;
1501 bdev
->bd_contains
= bdev
;
1502 bdev
->bd_partno
= partno
;
1506 bdev
->bd_part
= disk_get_part(disk
, partno
);
1511 if (disk
->fops
->open
) {
1512 ret
= disk
->fops
->open(bdev
, mode
);
1513 if (ret
== -ERESTARTSYS
) {
1514 /* Lost a race with 'disk' being
1515 * deleted, try again.
1518 disk_put_part(bdev
->bd_part
);
1519 bdev
->bd_part
= NULL
;
1520 bdev
->bd_disk
= NULL
;
1521 bdev
->bd_queue
= NULL
;
1522 mutex_unlock(&bdev
->bd_mutex
);
1523 disk_unblock_events(disk
);
1524 put_disk_and_module(disk
);
1530 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1531 set_init_blocksize(bdev
);
1535 * If the device is invalidated, rescan partition
1536 * if open succeeded or failed with -ENOMEDIUM.
1537 * The latter is necessary to prevent ghost
1538 * partitions on a removed medium.
1540 if (bdev
->bd_invalidated
) {
1542 rescan_partitions(disk
, bdev
);
1543 else if (ret
== -ENOMEDIUM
)
1544 invalidate_partitions(disk
, bdev
);
1550 struct block_device
*whole
;
1551 whole
= bdget_disk(disk
, 0);
1556 ret
= __blkdev_get(whole
, mode
, 1);
1559 bdev
->bd_contains
= whole
;
1560 bdev
->bd_part
= disk_get_part(disk
, partno
);
1561 if (!(disk
->flags
& GENHD_FL_UP
) ||
1562 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1566 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1567 set_init_blocksize(bdev
);
1570 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1571 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1573 if (bdev
->bd_contains
== bdev
) {
1575 if (bdev
->bd_disk
->fops
->open
)
1576 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1577 /* the same as first opener case, read comment there */
1578 if (bdev
->bd_invalidated
) {
1580 rescan_partitions(bdev
->bd_disk
, bdev
);
1581 else if (ret
== -ENOMEDIUM
)
1582 invalidate_partitions(bdev
->bd_disk
, bdev
);
1585 goto out_unlock_bdev
;
1590 bdev
->bd_part_count
++;
1591 mutex_unlock(&bdev
->bd_mutex
);
1592 disk_unblock_events(disk
);
1593 /* only one opener holds refs to the module and disk */
1595 put_disk_and_module(disk
);
1599 disk_put_part(bdev
->bd_part
);
1600 bdev
->bd_disk
= NULL
;
1601 bdev
->bd_part
= NULL
;
1602 bdev
->bd_queue
= NULL
;
1603 if (bdev
!= bdev
->bd_contains
)
1604 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1605 bdev
->bd_contains
= NULL
;
1607 mutex_unlock(&bdev
->bd_mutex
);
1608 disk_unblock_events(disk
);
1609 put_disk_and_module(disk
);
1617 * blkdev_get - open a block device
1618 * @bdev: block_device to open
1619 * @mode: FMODE_* mask
1620 * @holder: exclusive holder identifier
1622 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1623 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1624 * @holder is invalid. Exclusive opens may nest for the same @holder.
1626 * On success, the reference count of @bdev is unchanged. On failure,
1633 * 0 on success, -errno on failure.
1635 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1637 struct block_device
*whole
= NULL
;
1640 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1642 if ((mode
& FMODE_EXCL
) && holder
) {
1643 whole
= bd_start_claiming(bdev
, holder
);
1644 if (IS_ERR(whole
)) {
1646 return PTR_ERR(whole
);
1650 res
= __blkdev_get(bdev
, mode
, 0);
1653 struct gendisk
*disk
= whole
->bd_disk
;
1655 /* finish claiming */
1656 mutex_lock(&bdev
->bd_mutex
);
1657 spin_lock(&bdev_lock
);
1660 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1662 * Note that for a whole device bd_holders
1663 * will be incremented twice, and bd_holder
1664 * will be set to bd_may_claim before being
1667 whole
->bd_holders
++;
1668 whole
->bd_holder
= bd_may_claim
;
1670 bdev
->bd_holder
= holder
;
1673 /* tell others that we're done */
1674 BUG_ON(whole
->bd_claiming
!= holder
);
1675 whole
->bd_claiming
= NULL
;
1676 wake_up_bit(&whole
->bd_claiming
, 0);
1678 spin_unlock(&bdev_lock
);
1681 * Block event polling for write claims if requested. Any
1682 * write holder makes the write_holder state stick until
1683 * all are released. This is good enough and tracking
1684 * individual writeable reference is too fragile given the
1685 * way @mode is used in blkdev_get/put().
1687 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1688 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1689 bdev
->bd_write_holder
= true;
1690 disk_block_events(disk
);
1693 mutex_unlock(&bdev
->bd_mutex
);
1699 EXPORT_SYMBOL(blkdev_get
);
1702 * blkdev_get_by_path - open a block device by name
1703 * @path: path to the block device to open
1704 * @mode: FMODE_* mask
1705 * @holder: exclusive holder identifier
1707 * Open the blockdevice described by the device file at @path. @mode
1708 * and @holder are identical to blkdev_get().
1710 * On success, the returned block_device has reference count of one.
1716 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1718 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1721 struct block_device
*bdev
;
1724 bdev
= lookup_bdev(path
);
1728 err
= blkdev_get(bdev
, mode
, holder
);
1730 return ERR_PTR(err
);
1732 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1733 blkdev_put(bdev
, mode
);
1734 return ERR_PTR(-EACCES
);
1739 EXPORT_SYMBOL(blkdev_get_by_path
);
1742 * blkdev_get_by_dev - open a block device by device number
1743 * @dev: device number of block device to open
1744 * @mode: FMODE_* mask
1745 * @holder: exclusive holder identifier
1747 * Open the blockdevice described by device number @dev. @mode and
1748 * @holder are identical to blkdev_get().
1750 * Use it ONLY if you really do not have anything better - i.e. when
1751 * you are behind a truly sucky interface and all you are given is a
1752 * device number. _Never_ to be used for internal purposes. If you
1753 * ever need it - reconsider your API.
1755 * On success, the returned block_device has reference count of one.
1761 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1763 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1765 struct block_device
*bdev
;
1770 return ERR_PTR(-ENOMEM
);
1772 err
= blkdev_get(bdev
, mode
, holder
);
1774 return ERR_PTR(err
);
1778 EXPORT_SYMBOL(blkdev_get_by_dev
);
1780 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1782 struct block_device
*bdev
;
1785 * Preserve backwards compatibility and allow large file access
1786 * even if userspace doesn't ask for it explicitly. Some mkfs
1787 * binary needs it. We might want to drop this workaround
1788 * during an unstable branch.
1790 filp
->f_flags
|= O_LARGEFILE
;
1792 filp
->f_mode
|= FMODE_NOWAIT
;
1794 if (filp
->f_flags
& O_NDELAY
)
1795 filp
->f_mode
|= FMODE_NDELAY
;
1796 if (filp
->f_flags
& O_EXCL
)
1797 filp
->f_mode
|= FMODE_EXCL
;
1798 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1799 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1801 bdev
= bd_acquire(inode
);
1805 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1806 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
1808 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1811 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1813 struct gendisk
*disk
= bdev
->bd_disk
;
1814 struct block_device
*victim
= NULL
;
1816 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1818 bdev
->bd_part_count
--;
1820 if (!--bdev
->bd_openers
) {
1821 WARN_ON_ONCE(bdev
->bd_holders
);
1822 sync_blockdev(bdev
);
1825 bdev_write_inode(bdev
);
1827 if (bdev
->bd_contains
== bdev
) {
1828 if (disk
->fops
->release
)
1829 disk
->fops
->release(disk
, mode
);
1831 if (!bdev
->bd_openers
) {
1832 disk_put_part(bdev
->bd_part
);
1833 bdev
->bd_part
= NULL
;
1834 bdev
->bd_disk
= NULL
;
1835 if (bdev
!= bdev
->bd_contains
)
1836 victim
= bdev
->bd_contains
;
1837 bdev
->bd_contains
= NULL
;
1839 put_disk_and_module(disk
);
1841 mutex_unlock(&bdev
->bd_mutex
);
1844 __blkdev_put(victim
, mode
, 1);
1847 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1849 mutex_lock(&bdev
->bd_mutex
);
1851 if (mode
& FMODE_EXCL
) {
1855 * Release a claim on the device. The holder fields
1856 * are protected with bdev_lock. bd_mutex is to
1857 * synchronize disk_holder unlinking.
1859 spin_lock(&bdev_lock
);
1861 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1862 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1864 /* bd_contains might point to self, check in a separate step */
1865 if ((bdev_free
= !bdev
->bd_holders
))
1866 bdev
->bd_holder
= NULL
;
1867 if (!bdev
->bd_contains
->bd_holders
)
1868 bdev
->bd_contains
->bd_holder
= NULL
;
1870 spin_unlock(&bdev_lock
);
1873 * If this was the last claim, remove holder link and
1874 * unblock evpoll if it was a write holder.
1876 if (bdev_free
&& bdev
->bd_write_holder
) {
1877 disk_unblock_events(bdev
->bd_disk
);
1878 bdev
->bd_write_holder
= false;
1883 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1884 * event. This is to ensure detection of media removal commanded
1885 * from userland - e.g. eject(1).
1887 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1889 mutex_unlock(&bdev
->bd_mutex
);
1891 __blkdev_put(bdev
, mode
, 0);
1893 EXPORT_SYMBOL(blkdev_put
);
1895 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1897 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1898 blkdev_put(bdev
, filp
->f_mode
);
1902 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1904 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1905 fmode_t mode
= file
->f_mode
;
1908 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1909 * to updated it before every ioctl.
1911 if (file
->f_flags
& O_NDELAY
)
1912 mode
|= FMODE_NDELAY
;
1914 mode
&= ~FMODE_NDELAY
;
1916 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1920 * Write data to the block device. Only intended for the block device itself
1921 * and the raw driver which basically is a fake block device.
1923 * Does not take i_mutex for the write and thus is not for general purpose
1926 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1928 struct file
*file
= iocb
->ki_filp
;
1929 struct inode
*bd_inode
= bdev_file_inode(file
);
1930 loff_t size
= i_size_read(bd_inode
);
1931 struct blk_plug plug
;
1934 if (bdev_read_only(I_BDEV(bd_inode
)))
1937 if (!iov_iter_count(from
))
1940 if (iocb
->ki_pos
>= size
)
1943 if ((iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_DIRECT
)) == IOCB_NOWAIT
)
1946 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1948 blk_start_plug(&plug
);
1949 ret
= __generic_file_write_iter(iocb
, from
);
1951 ret
= generic_write_sync(iocb
, ret
);
1952 blk_finish_plug(&plug
);
1955 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1957 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1959 struct file
*file
= iocb
->ki_filp
;
1960 struct inode
*bd_inode
= bdev_file_inode(file
);
1961 loff_t size
= i_size_read(bd_inode
);
1962 loff_t pos
= iocb
->ki_pos
;
1968 iov_iter_truncate(to
, size
);
1969 return generic_file_read_iter(iocb
, to
);
1971 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1974 * Try to release a page associated with block device when the system
1975 * is under memory pressure.
1977 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1979 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1981 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1982 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1984 return try_to_free_buffers(page
);
1987 static int blkdev_writepages(struct address_space
*mapping
,
1988 struct writeback_control
*wbc
)
1990 return generic_writepages(mapping
, wbc
);
1993 static const struct address_space_operations def_blk_aops
= {
1994 .readpage
= blkdev_readpage
,
1995 .readpages
= blkdev_readpages
,
1996 .writepage
= blkdev_writepage
,
1997 .write_begin
= blkdev_write_begin
,
1998 .write_end
= blkdev_write_end
,
1999 .writepages
= blkdev_writepages
,
2000 .releasepage
= blkdev_releasepage
,
2001 .direct_IO
= blkdev_direct_IO
,
2002 .migratepage
= buffer_migrate_page_norefs
,
2003 .is_dirty_writeback
= buffer_check_dirty_writeback
,
2006 #define BLKDEV_FALLOC_FL_SUPPORTED \
2007 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2008 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2010 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
2013 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
2014 struct address_space
*mapping
;
2015 loff_t end
= start
+ len
- 1;
2019 /* Fail if we don't recognize the flags. */
2020 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
2023 /* Don't go off the end of the device. */
2024 isize
= i_size_read(bdev
->bd_inode
);
2028 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2029 len
= isize
- start
;
2030 end
= start
+ len
- 1;
2036 * Don't allow IO that isn't aligned to logical block size.
2038 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2041 /* Invalidate the page cache, including dirty pages. */
2042 mapping
= bdev
->bd_inode
->i_mapping
;
2043 truncate_inode_pages_range(mapping
, start
, end
);
2046 case FALLOC_FL_ZERO_RANGE
:
2047 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2048 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2049 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
2051 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2052 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2053 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
2055 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2056 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2066 * Invalidate again; if someone wandered in and dirtied a page,
2067 * the caller will be given -EBUSY. The third argument is
2068 * inclusive, so the rounding here is safe.
2070 return invalidate_inode_pages2_range(mapping
,
2071 start
>> PAGE_SHIFT
,
2075 const struct file_operations def_blk_fops
= {
2076 .open
= blkdev_open
,
2077 .release
= blkdev_close
,
2078 .llseek
= block_llseek
,
2079 .read_iter
= blkdev_read_iter
,
2080 .write_iter
= blkdev_write_iter
,
2081 .mmap
= generic_file_mmap
,
2082 .fsync
= blkdev_fsync
,
2083 .unlocked_ioctl
= block_ioctl
,
2084 #ifdef CONFIG_COMPAT
2085 .compat_ioctl
= compat_blkdev_ioctl
,
2087 .splice_read
= generic_file_splice_read
,
2088 .splice_write
= iter_file_splice_write
,
2089 .fallocate
= blkdev_fallocate
,
2092 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2095 mm_segment_t old_fs
= get_fs();
2097 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2102 EXPORT_SYMBOL(ioctl_by_bdev
);
2105 * lookup_bdev - lookup a struct block_device by name
2106 * @pathname: special file representing the block device
2108 * Get a reference to the blockdevice at @pathname in the current
2109 * namespace if possible and return it. Return ERR_PTR(error)
2112 struct block_device
*lookup_bdev(const char *pathname
)
2114 struct block_device
*bdev
;
2115 struct inode
*inode
;
2119 if (!pathname
|| !*pathname
)
2120 return ERR_PTR(-EINVAL
);
2122 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2124 return ERR_PTR(error
);
2126 inode
= d_backing_inode(path
.dentry
);
2128 if (!S_ISBLK(inode
->i_mode
))
2131 if (!may_open_dev(&path
))
2134 bdev
= bd_acquire(inode
);
2141 bdev
= ERR_PTR(error
);
2144 EXPORT_SYMBOL(lookup_bdev
);
2146 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2148 struct super_block
*sb
= get_super(bdev
);
2153 * no need to lock the super, get_super holds the
2154 * read mutex so the filesystem cannot go away
2155 * under us (->put_super runs with the write lock
2158 shrink_dcache_sb(sb
);
2159 res
= invalidate_inodes(sb
, kill_dirty
);
2162 invalidate_bdev(bdev
);
2165 EXPORT_SYMBOL(__invalidate_device
);
2167 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2169 struct inode
*inode
, *old_inode
= NULL
;
2171 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2172 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2173 struct address_space
*mapping
= inode
->i_mapping
;
2174 struct block_device
*bdev
;
2176 spin_lock(&inode
->i_lock
);
2177 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2178 mapping
->nrpages
== 0) {
2179 spin_unlock(&inode
->i_lock
);
2183 spin_unlock(&inode
->i_lock
);
2184 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2186 * We hold a reference to 'inode' so it couldn't have been
2187 * removed from s_inodes list while we dropped the
2188 * s_inode_list_lock We cannot iput the inode now as we can
2189 * be holding the last reference and we cannot iput it under
2190 * s_inode_list_lock. So we keep the reference and iput it
2195 bdev
= I_BDEV(inode
);
2197 mutex_lock(&bdev
->bd_mutex
);
2198 if (bdev
->bd_openers
)
2200 mutex_unlock(&bdev
->bd_mutex
);
2202 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2204 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);