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 int set_blocksize(struct block_device
*bdev
, int size
)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size
< bdev_logical_block_size(bdev
))
117 /* Don't change the size if it is same as current */
118 if (bdev
->bd_block_size
!= size
) {
120 bdev
->bd_block_size
= size
;
121 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
127 EXPORT_SYMBOL(set_blocksize
);
129 int sb_set_blocksize(struct super_block
*sb
, int size
)
131 if (set_blocksize(sb
->s_bdev
, size
))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb
->s_blocksize
= size
;
136 sb
->s_blocksize_bits
= blksize_bits(size
);
137 return sb
->s_blocksize
;
140 EXPORT_SYMBOL(sb_set_blocksize
);
142 int sb_min_blocksize(struct super_block
*sb
, int size
)
144 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
147 return sb_set_blocksize(sb
, size
);
150 EXPORT_SYMBOL(sb_min_blocksize
);
153 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
154 struct buffer_head
*bh
, int create
)
156 bh
->b_bdev
= I_BDEV(inode
);
157 bh
->b_blocknr
= iblock
;
158 set_buffer_mapped(bh
);
162 static struct inode
*bdev_file_inode(struct file
*file
)
164 return file
->f_mapping
->host
;
167 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
169 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
171 /* avoid the need for a I/O completion work item */
172 if (iocb
->ki_flags
& IOCB_DSYNC
)
177 #define DIO_INLINE_BIO_VECS 4
179 static void blkdev_bio_end_io_simple(struct bio
*bio
)
181 struct task_struct
*waiter
= bio
->bi_private
;
183 WRITE_ONCE(bio
->bi_private
, NULL
);
184 wake_up_process(waiter
);
188 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
191 struct file
*file
= iocb
->ki_filp
;
192 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
193 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
194 loff_t pos
= iocb
->ki_pos
;
195 bool should_dirty
= false;
201 if ((pos
| iov_iter_alignment(iter
)) &
202 (bdev_logical_block_size(bdev
) - 1))
205 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
208 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
213 bio_init(&bio
, vecs
, nr_pages
);
214 bio_set_dev(&bio
, bdev
);
215 bio
.bi_iter
.bi_sector
= pos
>> 9;
216 bio
.bi_write_hint
= iocb
->ki_hint
;
217 bio
.bi_private
= current
;
218 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
220 ret
= bio_iov_iter_get_pages(&bio
, iter
);
223 ret
= bio
.bi_iter
.bi_size
;
225 if (iov_iter_rw(iter
) == READ
) {
226 bio
.bi_opf
= REQ_OP_READ
;
227 if (iter_is_iovec(iter
))
230 bio
.bi_opf
= dio_bio_write_op(iocb
);
231 task_io_account_write(ret
);
234 qc
= submit_bio(&bio
);
236 set_current_state(TASK_UNINTERRUPTIBLE
);
237 if (!READ_ONCE(bio
.bi_private
))
239 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
240 !blk_poll(bdev_get_queue(bdev
), qc
))
243 __set_current_state(TASK_RUNNING
);
245 bio_for_each_segment_all(bvec
, &bio
, i
) {
246 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
247 set_page_dirty_lock(bvec
->bv_page
);
248 put_page(bvec
->bv_page
);
251 if (unlikely(bio
.bi_status
))
252 ret
= blk_status_to_errno(bio
.bi_status
);
255 if (vecs
!= inline_vecs
)
266 struct task_struct
*waiter
;
271 bool should_dirty
: 1;
276 static struct bio_set
*blkdev_dio_pool __read_mostly
;
278 static void blkdev_bio_end_io(struct bio
*bio
)
280 struct blkdev_dio
*dio
= bio
->bi_private
;
281 bool should_dirty
= dio
->should_dirty
;
283 if (dio
->multi_bio
&& !atomic_dec_and_test(&dio
->ref
)) {
284 if (bio
->bi_status
&& !dio
->bio
.bi_status
)
285 dio
->bio
.bi_status
= bio
->bi_status
;
288 struct kiocb
*iocb
= dio
->iocb
;
291 if (likely(!dio
->bio
.bi_status
)) {
295 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
298 dio
->iocb
->ki_complete(iocb
, ret
, 0);
301 struct task_struct
*waiter
= dio
->waiter
;
303 WRITE_ONCE(dio
->waiter
, NULL
);
304 wake_up_process(waiter
);
309 bio_check_pages_dirty(bio
);
311 struct bio_vec
*bvec
;
314 bio_for_each_segment_all(bvec
, bio
, i
)
315 put_page(bvec
->bv_page
);
321 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
323 struct file
*file
= iocb
->ki_filp
;
324 struct inode
*inode
= bdev_file_inode(file
);
325 struct block_device
*bdev
= I_BDEV(inode
);
326 struct blk_plug plug
;
327 struct blkdev_dio
*dio
;
329 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
330 loff_t pos
= iocb
->ki_pos
;
331 blk_qc_t qc
= BLK_QC_T_NONE
;
334 if ((pos
| iov_iter_alignment(iter
)) &
335 (bdev_logical_block_size(bdev
) - 1))
338 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, blkdev_dio_pool
);
339 bio_get(bio
); /* extra ref for the completion handler */
341 dio
= container_of(bio
, struct blkdev_dio
, bio
);
342 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
344 dio
->waiter
= current
;
349 dio
->multi_bio
= false;
350 dio
->should_dirty
= is_read
&& (iter
->type
== ITER_IOVEC
);
352 blk_start_plug(&plug
);
354 bio_set_dev(bio
, bdev
);
355 bio
->bi_iter
.bi_sector
= pos
>> 9;
356 bio
->bi_write_hint
= iocb
->ki_hint
;
357 bio
->bi_private
= dio
;
358 bio
->bi_end_io
= blkdev_bio_end_io
;
360 ret
= bio_iov_iter_get_pages(bio
, iter
);
362 bio
->bi_status
= BLK_STS_IOERR
;
368 bio
->bi_opf
= REQ_OP_READ
;
369 if (dio
->should_dirty
)
370 bio_set_pages_dirty(bio
);
372 bio
->bi_opf
= dio_bio_write_op(iocb
);
373 task_io_account_write(bio
->bi_iter
.bi_size
);
376 dio
->size
+= bio
->bi_iter
.bi_size
;
377 pos
+= bio
->bi_iter
.bi_size
;
379 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
381 qc
= submit_bio(bio
);
385 if (!dio
->multi_bio
) {
386 dio
->multi_bio
= true;
387 atomic_set(&dio
->ref
, 2);
389 atomic_inc(&dio
->ref
);
393 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
395 blk_finish_plug(&plug
);
401 set_current_state(TASK_UNINTERRUPTIBLE
);
402 if (!READ_ONCE(dio
->waiter
))
405 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
406 !blk_poll(bdev_get_queue(bdev
), qc
))
409 __set_current_state(TASK_RUNNING
);
412 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
421 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
425 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
428 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
429 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
431 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
434 static __init
int blkdev_init(void)
436 blkdev_dio_pool
= bioset_create(4, offsetof(struct blkdev_dio
, bio
), BIOSET_NEED_BVECS
);
437 if (!blkdev_dio_pool
)
441 module_init(blkdev_init
);
443 int __sync_blockdev(struct block_device
*bdev
, int wait
)
448 return filemap_flush(bdev
->bd_inode
->i_mapping
);
449 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
453 * Write out and wait upon all the dirty data associated with a block
454 * device via its mapping. Does not take the superblock lock.
456 int sync_blockdev(struct block_device
*bdev
)
458 return __sync_blockdev(bdev
, 1);
460 EXPORT_SYMBOL(sync_blockdev
);
463 * Write out and wait upon all dirty data associated with this
464 * device. Filesystem data as well as the underlying block
465 * device. Takes the superblock lock.
467 int fsync_bdev(struct block_device
*bdev
)
469 struct super_block
*sb
= get_super(bdev
);
471 int res
= sync_filesystem(sb
);
475 return sync_blockdev(bdev
);
477 EXPORT_SYMBOL(fsync_bdev
);
480 * freeze_bdev -- lock a filesystem and force it into a consistent state
481 * @bdev: blockdevice to lock
483 * If a superblock is found on this device, we take the s_umount semaphore
484 * on it to make sure nobody unmounts until the snapshot creation is done.
485 * The reference counter (bd_fsfreeze_count) guarantees that only the last
486 * unfreeze process can unfreeze the frozen filesystem actually when multiple
487 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
488 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
491 struct super_block
*freeze_bdev(struct block_device
*bdev
)
493 struct super_block
*sb
;
496 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
497 if (++bdev
->bd_fsfreeze_count
> 1) {
499 * We don't even need to grab a reference - the first call
500 * to freeze_bdev grab an active reference and only the last
501 * thaw_bdev drops it.
503 sb
= get_super(bdev
);
506 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
510 sb
= get_active_super(bdev
);
513 if (sb
->s_op
->freeze_super
)
514 error
= sb
->s_op
->freeze_super(sb
);
516 error
= freeze_super(sb
);
518 deactivate_super(sb
);
519 bdev
->bd_fsfreeze_count
--;
520 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
521 return ERR_PTR(error
);
523 deactivate_super(sb
);
526 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
527 return sb
; /* thaw_bdev releases s->s_umount */
529 EXPORT_SYMBOL(freeze_bdev
);
532 * thaw_bdev -- unlock filesystem
533 * @bdev: blockdevice to unlock
534 * @sb: associated superblock
536 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
538 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
542 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
543 if (!bdev
->bd_fsfreeze_count
)
547 if (--bdev
->bd_fsfreeze_count
> 0)
553 if (sb
->s_op
->thaw_super
)
554 error
= sb
->s_op
->thaw_super(sb
);
556 error
= thaw_super(sb
);
558 bdev
->bd_fsfreeze_count
++;
560 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
563 EXPORT_SYMBOL(thaw_bdev
);
565 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
567 return block_write_full_page(page
, blkdev_get_block
, wbc
);
570 static int blkdev_readpage(struct file
* file
, struct page
* page
)
572 return block_read_full_page(page
, blkdev_get_block
);
575 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
576 struct list_head
*pages
, unsigned nr_pages
)
578 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
581 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
582 loff_t pos
, unsigned len
, unsigned flags
,
583 struct page
**pagep
, void **fsdata
)
585 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
589 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
590 loff_t pos
, unsigned len
, unsigned copied
,
591 struct page
*page
, void *fsdata
)
594 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
604 * for a block special file file_inode(file)->i_size is zero
605 * so we compute the size by hand (just as in block_read/write above)
607 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
609 struct inode
*bd_inode
= bdev_file_inode(file
);
612 inode_lock(bd_inode
);
613 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
614 inode_unlock(bd_inode
);
618 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
620 struct inode
*bd_inode
= bdev_file_inode(filp
);
621 struct block_device
*bdev
= I_BDEV(bd_inode
);
624 error
= file_write_and_wait_range(filp
, start
, end
);
629 * There is no need to serialise calls to blkdev_issue_flush with
630 * i_mutex and doing so causes performance issues with concurrent
631 * O_SYNC writers to a block device.
633 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
634 if (error
== -EOPNOTSUPP
)
639 EXPORT_SYMBOL(blkdev_fsync
);
642 * bdev_read_page() - Start reading a page from a block device
643 * @bdev: The device to read the page from
644 * @sector: The offset on the device to read the page to (need not be aligned)
645 * @page: The page to read
647 * On entry, the page should be locked. It will be unlocked when the page
648 * has been read. If the block driver implements rw_page synchronously,
649 * that will be true on exit from this function, but it need not be.
651 * Errors returned by this function are usually "soft", eg out of memory, or
652 * queue full; callers should try a different route to read this page rather
653 * than propagate an error back up the stack.
655 * Return: negative errno if an error occurs, 0 if submission was successful.
657 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
660 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
661 int result
= -EOPNOTSUPP
;
663 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
666 result
= blk_queue_enter(bdev
->bd_queue
, 0);
669 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
670 blk_queue_exit(bdev
->bd_queue
);
673 EXPORT_SYMBOL_GPL(bdev_read_page
);
676 * bdev_write_page() - Start writing a page to a block device
677 * @bdev: The device to write the page to
678 * @sector: The offset on the device to write the page to (need not be aligned)
679 * @page: The page to write
680 * @wbc: The writeback_control for the write
682 * On entry, the page should be locked and not currently under writeback.
683 * On exit, if the write started successfully, the page will be unlocked and
684 * under writeback. If the write failed already (eg the driver failed to
685 * queue the page to the device), the page will still be locked. If the
686 * caller is a ->writepage implementation, it will need to unlock the page.
688 * Errors returned by this function are usually "soft", eg out of memory, or
689 * queue full; callers should try a different route to write 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_write_page(struct block_device
*bdev
, sector_t sector
,
695 struct page
*page
, struct writeback_control
*wbc
)
698 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
700 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
702 result
= blk_queue_enter(bdev
->bd_queue
, 0);
706 set_page_writeback(page
);
707 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
709 end_page_writeback(page
);
711 clean_page_buffers(page
);
714 blk_queue_exit(bdev
->bd_queue
);
717 EXPORT_SYMBOL_GPL(bdev_write_page
);
723 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
724 static struct kmem_cache
* bdev_cachep __read_mostly
;
726 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
728 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
731 return &ei
->vfs_inode
;
734 static void bdev_i_callback(struct rcu_head
*head
)
736 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
737 struct bdev_inode
*bdi
= BDEV_I(inode
);
739 kmem_cache_free(bdev_cachep
, bdi
);
742 static void bdev_destroy_inode(struct inode
*inode
)
744 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
747 static void init_once(void *foo
)
749 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
750 struct block_device
*bdev
= &ei
->bdev
;
752 memset(bdev
, 0, sizeof(*bdev
));
753 mutex_init(&bdev
->bd_mutex
);
754 INIT_LIST_HEAD(&bdev
->bd_list
);
756 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
758 bdev
->bd_bdi
= &noop_backing_dev_info
;
759 inode_init_once(&ei
->vfs_inode
);
760 /* Initialize mutex for freeze. */
761 mutex_init(&bdev
->bd_fsfreeze_mutex
);
764 static void bdev_evict_inode(struct inode
*inode
)
766 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
767 truncate_inode_pages_final(&inode
->i_data
);
768 invalidate_inode_buffers(inode
); /* is it needed here? */
770 spin_lock(&bdev_lock
);
771 list_del_init(&bdev
->bd_list
);
772 spin_unlock(&bdev_lock
);
773 /* Detach inode from wb early as bdi_put() may free bdi->wb */
774 inode_detach_wb(inode
);
775 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
776 bdi_put(bdev
->bd_bdi
);
777 bdev
->bd_bdi
= &noop_backing_dev_info
;
781 static const struct super_operations bdev_sops
= {
782 .statfs
= simple_statfs
,
783 .alloc_inode
= bdev_alloc_inode
,
784 .destroy_inode
= bdev_destroy_inode
,
785 .drop_inode
= generic_delete_inode
,
786 .evict_inode
= bdev_evict_inode
,
789 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
790 int flags
, const char *dev_name
, void *data
)
793 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
795 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
799 static struct file_system_type bd_type
= {
802 .kill_sb
= kill_anon_super
,
805 struct super_block
*blockdev_superblock __read_mostly
;
806 EXPORT_SYMBOL_GPL(blockdev_superblock
);
808 void __init
bdev_cache_init(void)
811 static struct vfsmount
*bd_mnt
;
813 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
814 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
815 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
817 err
= register_filesystem(&bd_type
);
819 panic("Cannot register bdev pseudo-fs");
820 bd_mnt
= kern_mount(&bd_type
);
822 panic("Cannot create bdev pseudo-fs");
823 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
827 * Most likely _very_ bad one - but then it's hardly critical for small
828 * /dev and can be fixed when somebody will need really large one.
829 * Keep in mind that it will be fed through icache hash function too.
831 static inline unsigned long hash(dev_t dev
)
833 return MAJOR(dev
)+MINOR(dev
);
836 static int bdev_test(struct inode
*inode
, void *data
)
838 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
841 static int bdev_set(struct inode
*inode
, void *data
)
843 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
847 static LIST_HEAD(all_bdevs
);
850 * If there is a bdev inode for this device, unhash it so that it gets evicted
851 * as soon as last inode reference is dropped.
853 void bdev_unhash_inode(dev_t dev
)
857 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
859 remove_inode_hash(inode
);
864 struct block_device
*bdget(dev_t dev
)
866 struct block_device
*bdev
;
869 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
870 bdev_test
, bdev_set
, &dev
);
875 bdev
= &BDEV_I(inode
)->bdev
;
877 if (inode
->i_state
& I_NEW
) {
878 bdev
->bd_contains
= NULL
;
879 bdev
->bd_super
= NULL
;
880 bdev
->bd_inode
= inode
;
881 bdev
->bd_block_size
= i_blocksize(inode
);
882 bdev
->bd_part_count
= 0;
883 bdev
->bd_invalidated
= 0;
884 inode
->i_mode
= S_IFBLK
;
886 inode
->i_bdev
= bdev
;
887 inode
->i_data
.a_ops
= &def_blk_aops
;
888 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
889 spin_lock(&bdev_lock
);
890 list_add(&bdev
->bd_list
, &all_bdevs
);
891 spin_unlock(&bdev_lock
);
892 unlock_new_inode(inode
);
897 EXPORT_SYMBOL(bdget
);
900 * bdgrab -- Grab a reference to an already referenced block device
901 * @bdev: Block device to grab a reference to.
903 struct block_device
*bdgrab(struct block_device
*bdev
)
905 ihold(bdev
->bd_inode
);
908 EXPORT_SYMBOL(bdgrab
);
910 long nr_blockdev_pages(void)
912 struct block_device
*bdev
;
914 spin_lock(&bdev_lock
);
915 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
916 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
918 spin_unlock(&bdev_lock
);
922 void bdput(struct block_device
*bdev
)
924 iput(bdev
->bd_inode
);
927 EXPORT_SYMBOL(bdput
);
929 static struct block_device
*bd_acquire(struct inode
*inode
)
931 struct block_device
*bdev
;
933 spin_lock(&bdev_lock
);
934 bdev
= inode
->i_bdev
;
935 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
937 spin_unlock(&bdev_lock
);
940 spin_unlock(&bdev_lock
);
943 * i_bdev references block device inode that was already shut down
944 * (corresponding device got removed). Remove the reference and look
945 * up block device inode again just in case new device got
946 * reestablished under the same device number.
951 bdev
= bdget(inode
->i_rdev
);
953 spin_lock(&bdev_lock
);
954 if (!inode
->i_bdev
) {
956 * We take an additional reference to bd_inode,
957 * and it's released in clear_inode() of inode.
958 * So, we can access it via ->i_mapping always
962 inode
->i_bdev
= bdev
;
963 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
965 spin_unlock(&bdev_lock
);
970 /* Call when you free inode */
972 void bd_forget(struct inode
*inode
)
974 struct block_device
*bdev
= NULL
;
976 spin_lock(&bdev_lock
);
977 if (!sb_is_blkdev_sb(inode
->i_sb
))
978 bdev
= inode
->i_bdev
;
979 inode
->i_bdev
= NULL
;
980 inode
->i_mapping
= &inode
->i_data
;
981 spin_unlock(&bdev_lock
);
988 * bd_may_claim - test whether a block device can be claimed
989 * @bdev: block device of interest
990 * @whole: whole block device containing @bdev, may equal @bdev
991 * @holder: holder trying to claim @bdev
993 * Test whether @bdev can be claimed by @holder.
996 * spin_lock(&bdev_lock).
999 * %true if @bdev can be claimed, %false otherwise.
1001 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1004 if (bdev
->bd_holder
== holder
)
1005 return true; /* already a holder */
1006 else if (bdev
->bd_holder
!= NULL
)
1007 return false; /* held by someone else */
1008 else if (whole
== bdev
)
1009 return true; /* is a whole device which isn't held */
1011 else if (whole
->bd_holder
== bd_may_claim
)
1012 return true; /* is a partition of a device that is being partitioned */
1013 else if (whole
->bd_holder
!= NULL
)
1014 return false; /* is a partition of a held device */
1016 return true; /* is a partition of an un-held device */
1020 * bd_prepare_to_claim - prepare to claim a block device
1021 * @bdev: block device of interest
1022 * @whole: the whole device containing @bdev, may equal @bdev
1023 * @holder: holder trying to claim @bdev
1025 * Prepare to claim @bdev. This function fails if @bdev is already
1026 * claimed by another holder and waits if another claiming is in
1027 * progress. This function doesn't actually claim. On successful
1028 * return, the caller has ownership of bd_claiming and bd_holder[s].
1031 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1032 * it multiple times.
1035 * 0 if @bdev can be claimed, -EBUSY otherwise.
1037 static int bd_prepare_to_claim(struct block_device
*bdev
,
1038 struct block_device
*whole
, void *holder
)
1041 /* if someone else claimed, fail */
1042 if (!bd_may_claim(bdev
, whole
, holder
))
1045 /* if claiming is already in progress, wait for it to finish */
1046 if (whole
->bd_claiming
) {
1047 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1050 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1051 spin_unlock(&bdev_lock
);
1053 finish_wait(wq
, &wait
);
1054 spin_lock(&bdev_lock
);
1063 * bd_start_claiming - start claiming a block device
1064 * @bdev: block device of interest
1065 * @holder: holder trying to claim @bdev
1067 * @bdev is about to be opened exclusively. Check @bdev can be opened
1068 * exclusively and mark that an exclusive open is in progress. Each
1069 * successful call to this function must be matched with a call to
1070 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1073 * This function is used to gain exclusive access to the block device
1074 * without actually causing other exclusive open attempts to fail. It
1075 * should be used when the open sequence itself requires exclusive
1076 * access but may subsequently fail.
1082 * Pointer to the block device containing @bdev on success, ERR_PTR()
1085 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1088 struct gendisk
*disk
;
1089 struct block_device
*whole
;
1095 * @bdev might not have been initialized properly yet, look up
1096 * and grab the outer block device the hard way.
1098 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1100 return ERR_PTR(-ENXIO
);
1103 * Normally, @bdev should equal what's returned from bdget_disk()
1104 * if partno is 0; however, some drivers (floppy) use multiple
1105 * bdev's for the same physical device and @bdev may be one of the
1106 * aliases. Keep @bdev if partno is 0. This means claimer
1107 * tracking is broken for those devices but it has always been that
1111 whole
= bdget_disk(disk
, 0);
1113 whole
= bdgrab(bdev
);
1115 module_put(disk
->fops
->owner
);
1118 return ERR_PTR(-ENOMEM
);
1120 /* prepare to claim, if successful, mark claiming in progress */
1121 spin_lock(&bdev_lock
);
1123 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1125 whole
->bd_claiming
= holder
;
1126 spin_unlock(&bdev_lock
);
1129 spin_unlock(&bdev_lock
);
1131 return ERR_PTR(err
);
1136 struct bd_holder_disk
{
1137 struct list_head list
;
1138 struct gendisk
*disk
;
1142 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1143 struct gendisk
*disk
)
1145 struct bd_holder_disk
*holder
;
1147 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1148 if (holder
->disk
== disk
)
1153 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1155 return sysfs_create_link(from
, to
, kobject_name(to
));
1158 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1160 sysfs_remove_link(from
, kobject_name(to
));
1164 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1165 * @bdev: the claimed slave bdev
1166 * @disk: the holding disk
1168 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1170 * This functions creates the following sysfs symlinks.
1172 * - from "slaves" directory of the holder @disk to the claimed @bdev
1173 * - from "holders" directory of the @bdev to the holder @disk
1175 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1176 * passed to bd_link_disk_holder(), then:
1178 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1179 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1181 * The caller must have claimed @bdev before calling this function and
1182 * ensure that both @bdev and @disk are valid during the creation and
1183 * lifetime of these symlinks.
1189 * 0 on success, -errno on failure.
1191 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1193 struct bd_holder_disk
*holder
;
1196 mutex_lock(&bdev
->bd_mutex
);
1198 WARN_ON_ONCE(!bdev
->bd_holder
);
1200 /* FIXME: remove the following once add_disk() handles errors */
1201 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1204 holder
= bd_find_holder_disk(bdev
, disk
);
1210 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1216 INIT_LIST_HEAD(&holder
->list
);
1217 holder
->disk
= disk
;
1220 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1224 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1228 * bdev could be deleted beneath us which would implicitly destroy
1229 * the holder directory. Hold on to it.
1231 kobject_get(bdev
->bd_part
->holder_dir
);
1233 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1237 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1241 mutex_unlock(&bdev
->bd_mutex
);
1244 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1247 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1248 * @bdev: the calimed slave bdev
1249 * @disk: the holding disk
1251 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1256 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1258 struct bd_holder_disk
*holder
;
1260 mutex_lock(&bdev
->bd_mutex
);
1262 holder
= bd_find_holder_disk(bdev
, disk
);
1264 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1265 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1266 del_symlink(bdev
->bd_part
->holder_dir
,
1267 &disk_to_dev(disk
)->kobj
);
1268 kobject_put(bdev
->bd_part
->holder_dir
);
1269 list_del_init(&holder
->list
);
1273 mutex_unlock(&bdev
->bd_mutex
);
1275 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1279 * flush_disk - invalidates all buffer-cache entries on a disk
1281 * @bdev: struct block device to be flushed
1282 * @kill_dirty: flag to guide handling of dirty inodes
1284 * Invalidates all buffer-cache entries on a disk. It should be called
1285 * when a disk has been changed -- either by a media change or online
1288 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1290 if (__invalidate_device(bdev
, kill_dirty
)) {
1291 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1292 "resized disk %s\n",
1293 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1298 if (disk_part_scan_enabled(bdev
->bd_disk
))
1299 bdev
->bd_invalidated
= 1;
1303 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1304 * @disk: struct gendisk to check
1305 * @bdev: struct bdev to adjust.
1307 * This routine checks to see if the bdev size does not match the disk size
1308 * and adjusts it if it differs.
1310 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1312 loff_t disk_size
, bdev_size
;
1314 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1315 bdev_size
= i_size_read(bdev
->bd_inode
);
1316 if (disk_size
!= bdev_size
) {
1318 "%s: detected capacity change from %lld to %lld\n",
1319 disk
->disk_name
, bdev_size
, disk_size
);
1320 i_size_write(bdev
->bd_inode
, disk_size
);
1321 flush_disk(bdev
, false);
1324 EXPORT_SYMBOL(check_disk_size_change
);
1327 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1328 * @disk: struct gendisk to be revalidated
1330 * This routine is a wrapper for lower-level driver's revalidate_disk
1331 * call-backs. It is used to do common pre and post operations needed
1332 * for all revalidate_disk operations.
1334 int revalidate_disk(struct gendisk
*disk
)
1336 struct block_device
*bdev
;
1339 if (disk
->fops
->revalidate_disk
)
1340 ret
= disk
->fops
->revalidate_disk(disk
);
1341 bdev
= bdget_disk(disk
, 0);
1345 mutex_lock(&bdev
->bd_mutex
);
1346 check_disk_size_change(disk
, bdev
);
1347 bdev
->bd_invalidated
= 0;
1348 mutex_unlock(&bdev
->bd_mutex
);
1352 EXPORT_SYMBOL(revalidate_disk
);
1355 * This routine checks whether a removable media has been changed,
1356 * and invalidates all buffer-cache-entries in that case. This
1357 * is a relatively slow routine, so we have to try to minimize using
1358 * it. Thus it is called only upon a 'mount' or 'open'. This
1359 * is the best way of combining speed and utility, I think.
1360 * People changing diskettes in the middle of an operation deserve
1363 int check_disk_change(struct block_device
*bdev
)
1365 struct gendisk
*disk
= bdev
->bd_disk
;
1366 const struct block_device_operations
*bdops
= disk
->fops
;
1367 unsigned int events
;
1369 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1370 DISK_EVENT_EJECT_REQUEST
);
1371 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1374 flush_disk(bdev
, true);
1375 if (bdops
->revalidate_disk
)
1376 bdops
->revalidate_disk(bdev
->bd_disk
);
1380 EXPORT_SYMBOL(check_disk_change
);
1382 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1384 unsigned bsize
= bdev_logical_block_size(bdev
);
1386 inode_lock(bdev
->bd_inode
);
1387 i_size_write(bdev
->bd_inode
, size
);
1388 inode_unlock(bdev
->bd_inode
);
1389 while (bsize
< PAGE_SIZE
) {
1394 bdev
->bd_block_size
= bsize
;
1395 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1397 EXPORT_SYMBOL(bd_set_size
);
1399 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1404 * mutex_lock(part->bd_mutex)
1405 * mutex_lock_nested(whole->bd_mutex, 1)
1408 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1410 struct gendisk
*disk
;
1411 struct module
*owner
;
1416 if (mode
& FMODE_READ
)
1418 if (mode
& FMODE_WRITE
)
1421 * hooks: /n/, see "layering violations".
1424 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1434 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1437 owner
= disk
->fops
->owner
;
1439 disk_block_events(disk
);
1440 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1441 if (!bdev
->bd_openers
) {
1442 bdev
->bd_disk
= disk
;
1443 bdev
->bd_queue
= disk
->queue
;
1444 bdev
->bd_contains
= bdev
;
1445 bdev
->bd_partno
= partno
;
1449 bdev
->bd_part
= disk_get_part(disk
, partno
);
1454 if (disk
->fops
->open
) {
1455 ret
= disk
->fops
->open(bdev
, mode
);
1456 if (ret
== -ERESTARTSYS
) {
1457 /* Lost a race with 'disk' being
1458 * deleted, try again.
1461 disk_put_part(bdev
->bd_part
);
1462 bdev
->bd_part
= NULL
;
1463 bdev
->bd_disk
= NULL
;
1464 bdev
->bd_queue
= NULL
;
1465 mutex_unlock(&bdev
->bd_mutex
);
1466 disk_unblock_events(disk
);
1474 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1477 * If the device is invalidated, rescan partition
1478 * if open succeeded or failed with -ENOMEDIUM.
1479 * The latter is necessary to prevent ghost
1480 * partitions on a removed medium.
1482 if (bdev
->bd_invalidated
) {
1484 rescan_partitions(disk
, bdev
);
1485 else if (ret
== -ENOMEDIUM
)
1486 invalidate_partitions(disk
, bdev
);
1492 struct block_device
*whole
;
1493 whole
= bdget_disk(disk
, 0);
1498 ret
= __blkdev_get(whole
, mode
, 1);
1501 bdev
->bd_contains
= whole
;
1502 bdev
->bd_part
= disk_get_part(disk
, partno
);
1503 if (!(disk
->flags
& GENHD_FL_UP
) ||
1504 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1508 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1511 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1512 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1514 if (bdev
->bd_contains
== bdev
) {
1516 if (bdev
->bd_disk
->fops
->open
)
1517 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1518 /* the same as first opener case, read comment there */
1519 if (bdev
->bd_invalidated
) {
1521 rescan_partitions(bdev
->bd_disk
, bdev
);
1522 else if (ret
== -ENOMEDIUM
)
1523 invalidate_partitions(bdev
->bd_disk
, bdev
);
1526 goto out_unlock_bdev
;
1528 /* only one opener holds refs to the module and disk */
1534 bdev
->bd_part_count
++;
1535 mutex_unlock(&bdev
->bd_mutex
);
1536 disk_unblock_events(disk
);
1540 disk_put_part(bdev
->bd_part
);
1541 bdev
->bd_disk
= NULL
;
1542 bdev
->bd_part
= NULL
;
1543 bdev
->bd_queue
= NULL
;
1544 if (bdev
!= bdev
->bd_contains
)
1545 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1546 bdev
->bd_contains
= NULL
;
1548 mutex_unlock(&bdev
->bd_mutex
);
1549 disk_unblock_events(disk
);
1559 * blkdev_get - open a block device
1560 * @bdev: block_device to open
1561 * @mode: FMODE_* mask
1562 * @holder: exclusive holder identifier
1564 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1565 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1566 * @holder is invalid. Exclusive opens may nest for the same @holder.
1568 * On success, the reference count of @bdev is unchanged. On failure,
1575 * 0 on success, -errno on failure.
1577 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1579 struct block_device
*whole
= NULL
;
1582 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1584 if ((mode
& FMODE_EXCL
) && holder
) {
1585 whole
= bd_start_claiming(bdev
, holder
);
1586 if (IS_ERR(whole
)) {
1588 return PTR_ERR(whole
);
1592 res
= __blkdev_get(bdev
, mode
, 0);
1595 struct gendisk
*disk
= whole
->bd_disk
;
1597 /* finish claiming */
1598 mutex_lock(&bdev
->bd_mutex
);
1599 spin_lock(&bdev_lock
);
1602 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1604 * Note that for a whole device bd_holders
1605 * will be incremented twice, and bd_holder
1606 * will be set to bd_may_claim before being
1609 whole
->bd_holders
++;
1610 whole
->bd_holder
= bd_may_claim
;
1612 bdev
->bd_holder
= holder
;
1615 /* tell others that we're done */
1616 BUG_ON(whole
->bd_claiming
!= holder
);
1617 whole
->bd_claiming
= NULL
;
1618 wake_up_bit(&whole
->bd_claiming
, 0);
1620 spin_unlock(&bdev_lock
);
1623 * Block event polling for write claims if requested. Any
1624 * write holder makes the write_holder state stick until
1625 * all are released. This is good enough and tracking
1626 * individual writeable reference is too fragile given the
1627 * way @mode is used in blkdev_get/put().
1629 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1630 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1631 bdev
->bd_write_holder
= true;
1632 disk_block_events(disk
);
1635 mutex_unlock(&bdev
->bd_mutex
);
1641 EXPORT_SYMBOL(blkdev_get
);
1644 * blkdev_get_by_path - open a block device by name
1645 * @path: path to the block device to open
1646 * @mode: FMODE_* mask
1647 * @holder: exclusive holder identifier
1649 * Open the blockdevice described by the device file at @path. @mode
1650 * and @holder are identical to blkdev_get().
1652 * On success, the returned block_device has reference count of one.
1658 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1660 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1663 struct block_device
*bdev
;
1667 if (mode
& FMODE_READ
)
1669 if (mode
& FMODE_WRITE
)
1671 bdev
= lookup_bdev(path
, perm
);
1675 err
= blkdev_get(bdev
, mode
, holder
);
1677 return ERR_PTR(err
);
1679 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1680 blkdev_put(bdev
, mode
);
1681 return ERR_PTR(-EACCES
);
1686 EXPORT_SYMBOL(blkdev_get_by_path
);
1689 * blkdev_get_by_dev - open a block device by device number
1690 * @dev: device number of block device to open
1691 * @mode: FMODE_* mask
1692 * @holder: exclusive holder identifier
1694 * Open the blockdevice described by device number @dev. @mode and
1695 * @holder are identical to blkdev_get().
1697 * Use it ONLY if you really do not have anything better - i.e. when
1698 * you are behind a truly sucky interface and all you are given is a
1699 * device number. _Never_ to be used for internal purposes. If you
1700 * ever need it - reconsider your API.
1702 * On success, the returned block_device has reference count of one.
1708 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1710 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1712 struct block_device
*bdev
;
1717 return ERR_PTR(-ENOMEM
);
1719 err
= blkdev_get(bdev
, mode
, holder
);
1721 return ERR_PTR(err
);
1725 EXPORT_SYMBOL(blkdev_get_by_dev
);
1727 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1729 struct block_device
*bdev
;
1732 * Preserve backwards compatibility and allow large file access
1733 * even if userspace doesn't ask for it explicitly. Some mkfs
1734 * binary needs it. We might want to drop this workaround
1735 * during an unstable branch.
1737 filp
->f_flags
|= O_LARGEFILE
;
1739 filp
->f_mode
|= FMODE_NOWAIT
;
1741 if (filp
->f_flags
& O_NDELAY
)
1742 filp
->f_mode
|= FMODE_NDELAY
;
1743 if (filp
->f_flags
& O_EXCL
)
1744 filp
->f_mode
|= FMODE_EXCL
;
1745 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1746 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1748 bdev
= bd_acquire(inode
);
1753 * A negative i_writecount for bdev->bd_inode means that the bdev
1754 * or one of its paritions is mounted in a user namespace. Deny
1755 * writing for non-root in this case, otherwise an unprivileged
1756 * user can attack the kernel by modifying the backing store of a
1757 * mounted filesystem.
1759 if ((filp
->f_mode
& FMODE_WRITE
) &&
1760 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
) &&
1761 !atomic_inc_unless_negative(&bdev
->bd_inode
->i_writecount
)) {
1766 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1767 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
1769 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1772 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1774 struct gendisk
*disk
= bdev
->bd_disk
;
1775 struct block_device
*victim
= NULL
;
1777 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1779 bdev
->bd_part_count
--;
1781 if (!--bdev
->bd_openers
) {
1782 WARN_ON_ONCE(bdev
->bd_holders
);
1783 sync_blockdev(bdev
);
1786 bdev_write_inode(bdev
);
1788 if (bdev
->bd_contains
== bdev
) {
1789 if (disk
->fops
->release
)
1790 disk
->fops
->release(disk
, mode
);
1792 if (!bdev
->bd_openers
) {
1793 struct module
*owner
= disk
->fops
->owner
;
1795 disk_put_part(bdev
->bd_part
);
1796 bdev
->bd_part
= NULL
;
1797 bdev
->bd_disk
= NULL
;
1798 if (bdev
!= bdev
->bd_contains
)
1799 victim
= bdev
->bd_contains
;
1800 bdev
->bd_contains
= NULL
;
1805 mutex_unlock(&bdev
->bd_mutex
);
1808 __blkdev_put(victim
, mode
, 1);
1811 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1813 mutex_lock(&bdev
->bd_mutex
);
1815 if (mode
& FMODE_EXCL
) {
1819 * Release a claim on the device. The holder fields
1820 * are protected with bdev_lock. bd_mutex is to
1821 * synchronize disk_holder unlinking.
1823 spin_lock(&bdev_lock
);
1825 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1826 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1828 /* bd_contains might point to self, check in a separate step */
1829 if ((bdev_free
= !bdev
->bd_holders
))
1830 bdev
->bd_holder
= NULL
;
1831 if (!bdev
->bd_contains
->bd_holders
)
1832 bdev
->bd_contains
->bd_holder
= NULL
;
1834 spin_unlock(&bdev_lock
);
1837 * If this was the last claim, remove holder link and
1838 * unblock evpoll if it was a write holder.
1840 if (bdev_free
&& bdev
->bd_write_holder
) {
1841 disk_unblock_events(bdev
->bd_disk
);
1842 bdev
->bd_write_holder
= false;
1847 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1848 * event. This is to ensure detection of media removal commanded
1849 * from userland - e.g. eject(1).
1851 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1853 mutex_unlock(&bdev
->bd_mutex
);
1855 __blkdev_put(bdev
, mode
, 0);
1857 EXPORT_SYMBOL(blkdev_put
);
1859 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1861 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1862 if (filp
->f_mode
& FMODE_WRITE
&&
1863 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
))
1864 atomic_dec(&bdev
->bd_inode
->i_writecount
);
1865 blkdev_put(bdev
, filp
->f_mode
);
1869 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1871 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1872 fmode_t mode
= file
->f_mode
;
1875 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1876 * to updated it before every ioctl.
1878 if (file
->f_flags
& O_NDELAY
)
1879 mode
|= FMODE_NDELAY
;
1881 mode
&= ~FMODE_NDELAY
;
1883 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1887 * Write data to the block device. Only intended for the block device itself
1888 * and the raw driver which basically is a fake block device.
1890 * Does not take i_mutex for the write and thus is not for general purpose
1893 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1895 struct file
*file
= iocb
->ki_filp
;
1896 struct inode
*bd_inode
= bdev_file_inode(file
);
1897 loff_t size
= i_size_read(bd_inode
);
1898 struct blk_plug plug
;
1901 if (bdev_read_only(I_BDEV(bd_inode
)))
1904 if (!iov_iter_count(from
))
1907 if (iocb
->ki_pos
>= size
)
1910 if ((iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_DIRECT
)) == IOCB_NOWAIT
)
1913 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1915 blk_start_plug(&plug
);
1916 ret
= __generic_file_write_iter(iocb
, from
);
1918 ret
= generic_write_sync(iocb
, ret
);
1919 blk_finish_plug(&plug
);
1922 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1924 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
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 loff_t pos
= iocb
->ki_pos
;
1935 iov_iter_truncate(to
, size
);
1936 return generic_file_read_iter(iocb
, to
);
1938 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1941 * Try to release a page associated with block device when the system
1942 * is under memory pressure.
1944 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1946 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1948 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1949 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1951 return try_to_free_buffers(page
);
1954 static int blkdev_writepages(struct address_space
*mapping
,
1955 struct writeback_control
*wbc
)
1957 if (dax_mapping(mapping
)) {
1958 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1960 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1962 return generic_writepages(mapping
, wbc
);
1965 static const struct address_space_operations def_blk_aops
= {
1966 .readpage
= blkdev_readpage
,
1967 .readpages
= blkdev_readpages
,
1968 .writepage
= blkdev_writepage
,
1969 .write_begin
= blkdev_write_begin
,
1970 .write_end
= blkdev_write_end
,
1971 .writepages
= blkdev_writepages
,
1972 .releasepage
= blkdev_releasepage
,
1973 .direct_IO
= blkdev_direct_IO
,
1974 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1977 #define BLKDEV_FALLOC_FL_SUPPORTED \
1978 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1979 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1981 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
1984 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1985 struct address_space
*mapping
;
1986 loff_t end
= start
+ len
- 1;
1990 /* Fail if we don't recognize the flags. */
1991 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
1994 /* Don't go off the end of the device. */
1995 isize
= i_size_read(bdev
->bd_inode
);
1999 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2000 len
= isize
- start
;
2001 end
= start
+ len
- 1;
2007 * Don't allow IO that isn't aligned to logical block size.
2009 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2012 /* Invalidate the page cache, including dirty pages. */
2013 mapping
= bdev
->bd_inode
->i_mapping
;
2014 truncate_inode_pages_range(mapping
, start
, end
);
2017 case FALLOC_FL_ZERO_RANGE
:
2018 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2019 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2020 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
2022 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2023 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2024 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
2026 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2027 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2037 * Invalidate again; if someone wandered in and dirtied a page,
2038 * the caller will be given -EBUSY. The third argument is
2039 * inclusive, so the rounding here is safe.
2041 return invalidate_inode_pages2_range(mapping
,
2042 start
>> PAGE_SHIFT
,
2046 const struct file_operations def_blk_fops
= {
2047 .open
= blkdev_open
,
2048 .release
= blkdev_close
,
2049 .llseek
= block_llseek
,
2050 .read_iter
= blkdev_read_iter
,
2051 .write_iter
= blkdev_write_iter
,
2052 .mmap
= generic_file_mmap
,
2053 .fsync
= blkdev_fsync
,
2054 .unlocked_ioctl
= block_ioctl
,
2055 #ifdef CONFIG_COMPAT
2056 .compat_ioctl
= compat_blkdev_ioctl
,
2058 .splice_read
= generic_file_splice_read
,
2059 .splice_write
= iter_file_splice_write
,
2060 .fallocate
= blkdev_fallocate
,
2063 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2066 mm_segment_t old_fs
= get_fs();
2068 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2073 EXPORT_SYMBOL(ioctl_by_bdev
);
2076 * lookup_bdev - lookup a struct block_device by name
2077 * @pathname: special file representing the block device
2078 * @mask: rights to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
2080 * Get a reference to the blockdevice at @pathname in the current
2081 * namespace if possible and return it. Return ERR_PTR(error)
2082 * otherwise. If @mask is non-zero, check for access rights to the
2083 * inode at @pathname.
2085 struct block_device
*lookup_bdev(const char *pathname
, int mask
)
2087 struct block_device
*bdev
;
2088 struct inode
*inode
;
2092 if (!pathname
|| !*pathname
)
2093 return ERR_PTR(-EINVAL
);
2095 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2097 return ERR_PTR(error
);
2099 inode
= d_backing_inode(path
.dentry
);
2100 if (mask
!= 0 && !capable(CAP_SYS_ADMIN
)) {
2101 error
= __inode_permission(inode
, mask
);
2106 if (!S_ISBLK(inode
->i_mode
))
2109 if (!may_open_dev(&path
))
2112 bdev
= bd_acquire(inode
);
2119 bdev
= ERR_PTR(error
);
2122 EXPORT_SYMBOL(lookup_bdev
);
2124 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2126 struct super_block
*sb
= get_super(bdev
);
2131 * no need to lock the super, get_super holds the
2132 * read mutex so the filesystem cannot go away
2133 * under us (->put_super runs with the write lock
2136 shrink_dcache_sb(sb
);
2137 res
= invalidate_inodes(sb
, kill_dirty
);
2140 invalidate_bdev(bdev
);
2143 EXPORT_SYMBOL(__invalidate_device
);
2145 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2147 struct inode
*inode
, *old_inode
= NULL
;
2149 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2150 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2151 struct address_space
*mapping
= inode
->i_mapping
;
2152 struct block_device
*bdev
;
2154 spin_lock(&inode
->i_lock
);
2155 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2156 mapping
->nrpages
== 0) {
2157 spin_unlock(&inode
->i_lock
);
2161 spin_unlock(&inode
->i_lock
);
2162 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2164 * We hold a reference to 'inode' so it couldn't have been
2165 * removed from s_inodes list while we dropped the
2166 * s_inode_list_lock We cannot iput the inode now as we can
2167 * be holding the last reference and we cannot iput it under
2168 * s_inode_list_lock. So we keep the reference and iput it
2173 bdev
= I_BDEV(inode
);
2175 mutex_lock(&bdev
->bd_mutex
);
2176 if (bdev
->bd_openers
)
2178 mutex_unlock(&bdev
->bd_mutex
);
2180 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2182 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);