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 wake_up_process(waiter
);
202 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
205 struct file
*file
= iocb
->ki_filp
;
206 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
207 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
208 loff_t pos
= iocb
->ki_pos
;
209 bool should_dirty
= false;
215 if ((pos
| iov_iter_alignment(iter
)) &
216 (bdev_logical_block_size(bdev
) - 1))
219 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
222 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
227 bio_init(&bio
, vecs
, nr_pages
);
228 bio_set_dev(&bio
, bdev
);
229 bio
.bi_iter
.bi_sector
= pos
>> 9;
230 bio
.bi_write_hint
= iocb
->ki_hint
;
231 bio
.bi_private
= current
;
232 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
234 ret
= bio_iov_iter_get_pages(&bio
, iter
);
237 ret
= bio
.bi_iter
.bi_size
;
239 if (iov_iter_rw(iter
) == READ
) {
240 bio
.bi_opf
= REQ_OP_READ
;
241 if (iter_is_iovec(iter
))
244 bio
.bi_opf
= dio_bio_write_op(iocb
);
245 task_io_account_write(ret
);
248 qc
= submit_bio(&bio
);
250 set_current_state(TASK_UNINTERRUPTIBLE
);
251 if (!READ_ONCE(bio
.bi_private
))
253 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
254 !blk_poll(bdev_get_queue(bdev
), qc
))
257 __set_current_state(TASK_RUNNING
);
259 bio_for_each_segment_all(bvec
, &bio
, i
) {
260 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
261 set_page_dirty_lock(bvec
->bv_page
);
262 put_page(bvec
->bv_page
);
265 if (unlikely(bio
.bi_status
))
266 ret
= blk_status_to_errno(bio
.bi_status
);
269 if (vecs
!= inline_vecs
)
280 struct task_struct
*waiter
;
285 bool should_dirty
: 1;
290 static struct bio_set
*blkdev_dio_pool __read_mostly
;
292 static void blkdev_bio_end_io(struct bio
*bio
)
294 struct blkdev_dio
*dio
= bio
->bi_private
;
295 bool should_dirty
= dio
->should_dirty
;
297 if (bio
->bi_status
&& !dio
->bio
.bi_status
)
298 dio
->bio
.bi_status
= bio
->bi_status
;
300 if (!dio
->multi_bio
|| atomic_dec_and_test(&dio
->ref
)) {
302 struct kiocb
*iocb
= dio
->iocb
;
305 if (likely(!dio
->bio
.bi_status
)) {
309 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
312 dio
->iocb
->ki_complete(iocb
, ret
, 0);
315 struct task_struct
*waiter
= dio
->waiter
;
317 WRITE_ONCE(dio
->waiter
, NULL
);
318 wake_up_process(waiter
);
323 bio_check_pages_dirty(bio
);
325 struct bio_vec
*bvec
;
328 bio_for_each_segment_all(bvec
, bio
, i
)
329 put_page(bvec
->bv_page
);
335 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
337 struct file
*file
= iocb
->ki_filp
;
338 struct inode
*inode
= bdev_file_inode(file
);
339 struct block_device
*bdev
= I_BDEV(inode
);
340 struct blk_plug plug
;
341 struct blkdev_dio
*dio
;
343 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
344 loff_t pos
= iocb
->ki_pos
;
345 blk_qc_t qc
= BLK_QC_T_NONE
;
348 if ((pos
| iov_iter_alignment(iter
)) &
349 (bdev_logical_block_size(bdev
) - 1))
352 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, blkdev_dio_pool
);
353 bio_get(bio
); /* extra ref for the completion handler */
355 dio
= container_of(bio
, struct blkdev_dio
, bio
);
356 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
358 dio
->waiter
= current
;
363 dio
->multi_bio
= false;
364 dio
->should_dirty
= is_read
&& (iter
->type
== ITER_IOVEC
);
366 blk_start_plug(&plug
);
368 bio_set_dev(bio
, bdev
);
369 bio
->bi_iter
.bi_sector
= pos
>> 9;
370 bio
->bi_write_hint
= iocb
->ki_hint
;
371 bio
->bi_private
= dio
;
372 bio
->bi_end_io
= blkdev_bio_end_io
;
374 ret
= bio_iov_iter_get_pages(bio
, iter
);
376 bio
->bi_status
= BLK_STS_IOERR
;
382 bio
->bi_opf
= REQ_OP_READ
;
383 if (dio
->should_dirty
)
384 bio_set_pages_dirty(bio
);
386 bio
->bi_opf
= dio_bio_write_op(iocb
);
387 task_io_account_write(bio
->bi_iter
.bi_size
);
390 dio
->size
+= bio
->bi_iter
.bi_size
;
391 pos
+= bio
->bi_iter
.bi_size
;
393 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
395 qc
= submit_bio(bio
);
399 if (!dio
->multi_bio
) {
400 dio
->multi_bio
= true;
401 atomic_set(&dio
->ref
, 2);
403 atomic_inc(&dio
->ref
);
407 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
409 blk_finish_plug(&plug
);
415 set_current_state(TASK_UNINTERRUPTIBLE
);
416 if (!READ_ONCE(dio
->waiter
))
419 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
420 !blk_poll(bdev_get_queue(bdev
), qc
))
423 __set_current_state(TASK_RUNNING
);
426 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
435 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
439 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
442 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
443 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
445 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
448 static __init
int blkdev_init(void)
450 blkdev_dio_pool
= bioset_create(4, offsetof(struct blkdev_dio
, bio
), BIOSET_NEED_BVECS
);
451 if (!blkdev_dio_pool
)
455 module_init(blkdev_init
);
457 int __sync_blockdev(struct block_device
*bdev
, int wait
)
462 return filemap_flush(bdev
->bd_inode
->i_mapping
);
463 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
467 * Write out and wait upon all the dirty data associated with a block
468 * device via its mapping. Does not take the superblock lock.
470 int sync_blockdev(struct block_device
*bdev
)
472 return __sync_blockdev(bdev
, 1);
474 EXPORT_SYMBOL(sync_blockdev
);
477 * Write out and wait upon all dirty data associated with this
478 * device. Filesystem data as well as the underlying block
479 * device. Takes the superblock lock.
481 int fsync_bdev(struct block_device
*bdev
)
483 struct super_block
*sb
= get_super(bdev
);
485 int res
= sync_filesystem(sb
);
489 return sync_blockdev(bdev
);
491 EXPORT_SYMBOL(fsync_bdev
);
494 * freeze_bdev -- lock a filesystem and force it into a consistent state
495 * @bdev: blockdevice to lock
497 * If a superblock is found on this device, we take the s_umount semaphore
498 * on it to make sure nobody unmounts until the snapshot creation is done.
499 * The reference counter (bd_fsfreeze_count) guarantees that only the last
500 * unfreeze process can unfreeze the frozen filesystem actually when multiple
501 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
502 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
505 struct super_block
*freeze_bdev(struct block_device
*bdev
)
507 struct super_block
*sb
;
510 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
511 if (++bdev
->bd_fsfreeze_count
> 1) {
513 * We don't even need to grab a reference - the first call
514 * to freeze_bdev grab an active reference and only the last
515 * thaw_bdev drops it.
517 sb
= get_super(bdev
);
520 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
524 sb
= get_active_super(bdev
);
527 if (sb
->s_op
->freeze_super
)
528 error
= sb
->s_op
->freeze_super(sb
);
530 error
= freeze_super(sb
);
532 deactivate_super(sb
);
533 bdev
->bd_fsfreeze_count
--;
534 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
535 return ERR_PTR(error
);
537 deactivate_super(sb
);
540 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
541 return sb
; /* thaw_bdev releases s->s_umount */
543 EXPORT_SYMBOL(freeze_bdev
);
546 * thaw_bdev -- unlock filesystem
547 * @bdev: blockdevice to unlock
548 * @sb: associated superblock
550 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
552 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
556 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
557 if (!bdev
->bd_fsfreeze_count
)
561 if (--bdev
->bd_fsfreeze_count
> 0)
567 if (sb
->s_op
->thaw_super
)
568 error
= sb
->s_op
->thaw_super(sb
);
570 error
= thaw_super(sb
);
572 bdev
->bd_fsfreeze_count
++;
574 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
577 EXPORT_SYMBOL(thaw_bdev
);
579 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
581 return block_write_full_page(page
, blkdev_get_block
, wbc
);
584 static int blkdev_readpage(struct file
* file
, struct page
* page
)
586 return block_read_full_page(page
, blkdev_get_block
);
589 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
590 struct list_head
*pages
, unsigned nr_pages
)
592 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
595 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
596 loff_t pos
, unsigned len
, unsigned flags
,
597 struct page
**pagep
, void **fsdata
)
599 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
603 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
604 loff_t pos
, unsigned len
, unsigned copied
,
605 struct page
*page
, void *fsdata
)
608 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
618 * for a block special file file_inode(file)->i_size is zero
619 * so we compute the size by hand (just as in block_read/write above)
621 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
623 struct inode
*bd_inode
= bdev_file_inode(file
);
626 inode_lock(bd_inode
);
627 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
628 inode_unlock(bd_inode
);
632 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
634 struct inode
*bd_inode
= bdev_file_inode(filp
);
635 struct block_device
*bdev
= I_BDEV(bd_inode
);
638 error
= file_write_and_wait_range(filp
, start
, end
);
643 * There is no need to serialise calls to blkdev_issue_flush with
644 * i_mutex and doing so causes performance issues with concurrent
645 * O_SYNC writers to a block device.
647 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
648 if (error
== -EOPNOTSUPP
)
653 EXPORT_SYMBOL(blkdev_fsync
);
656 * bdev_read_page() - Start reading a page from a block device
657 * @bdev: The device to read the page from
658 * @sector: The offset on the device to read the page to (need not be aligned)
659 * @page: The page to read
661 * On entry, the page should be locked. It will be unlocked when the page
662 * has been read. If the block driver implements rw_page synchronously,
663 * that will be true on exit from this function, but it need not be.
665 * Errors returned by this function are usually "soft", eg out of memory, or
666 * queue full; callers should try a different route to read this page rather
667 * than propagate an error back up the stack.
669 * Return: negative errno if an error occurs, 0 if submission was successful.
671 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
674 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
675 int result
= -EOPNOTSUPP
;
677 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
680 result
= blk_queue_enter(bdev
->bd_queue
, 0);
683 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
684 blk_queue_exit(bdev
->bd_queue
);
687 EXPORT_SYMBOL_GPL(bdev_read_page
);
690 * bdev_write_page() - Start writing a page to a block device
691 * @bdev: The device to write the page to
692 * @sector: The offset on the device to write the page to (need not be aligned)
693 * @page: The page to write
694 * @wbc: The writeback_control for the write
696 * On entry, the page should be locked and not currently under writeback.
697 * On exit, if the write started successfully, the page will be unlocked and
698 * under writeback. If the write failed already (eg the driver failed to
699 * queue the page to the device), the page will still be locked. If the
700 * caller is a ->writepage implementation, it will need to unlock the page.
702 * Errors returned by this function are usually "soft", eg out of memory, or
703 * queue full; callers should try a different route to write this page rather
704 * than propagate an error back up the stack.
706 * Return: negative errno if an error occurs, 0 if submission was successful.
708 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
709 struct page
*page
, struct writeback_control
*wbc
)
712 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
714 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
716 result
= blk_queue_enter(bdev
->bd_queue
, 0);
720 set_page_writeback(page
);
721 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
723 end_page_writeback(page
);
725 clean_page_buffers(page
);
728 blk_queue_exit(bdev
->bd_queue
);
731 EXPORT_SYMBOL_GPL(bdev_write_page
);
737 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
738 static struct kmem_cache
* bdev_cachep __read_mostly
;
740 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
742 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
745 return &ei
->vfs_inode
;
748 static void bdev_i_callback(struct rcu_head
*head
)
750 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
751 struct bdev_inode
*bdi
= BDEV_I(inode
);
753 kmem_cache_free(bdev_cachep
, bdi
);
756 static void bdev_destroy_inode(struct inode
*inode
)
758 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
761 static void init_once(void *foo
)
763 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
764 struct block_device
*bdev
= &ei
->bdev
;
766 memset(bdev
, 0, sizeof(*bdev
));
767 mutex_init(&bdev
->bd_mutex
);
768 INIT_LIST_HEAD(&bdev
->bd_list
);
770 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
772 bdev
->bd_bdi
= &noop_backing_dev_info
;
773 inode_init_once(&ei
->vfs_inode
);
774 /* Initialize mutex for freeze. */
775 mutex_init(&bdev
->bd_fsfreeze_mutex
);
778 static void bdev_evict_inode(struct inode
*inode
)
780 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
781 truncate_inode_pages_final(&inode
->i_data
);
782 invalidate_inode_buffers(inode
); /* is it needed here? */
784 spin_lock(&bdev_lock
);
785 list_del_init(&bdev
->bd_list
);
786 spin_unlock(&bdev_lock
);
787 /* Detach inode from wb early as bdi_put() may free bdi->wb */
788 inode_detach_wb(inode
);
789 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
790 bdi_put(bdev
->bd_bdi
);
791 bdev
->bd_bdi
= &noop_backing_dev_info
;
795 static const struct super_operations bdev_sops
= {
796 .statfs
= simple_statfs
,
797 .alloc_inode
= bdev_alloc_inode
,
798 .destroy_inode
= bdev_destroy_inode
,
799 .drop_inode
= generic_delete_inode
,
800 .evict_inode
= bdev_evict_inode
,
803 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
804 int flags
, const char *dev_name
, void *data
)
807 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
809 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
813 static struct file_system_type bd_type
= {
816 .kill_sb
= kill_anon_super
,
819 struct super_block
*blockdev_superblock __read_mostly
;
820 EXPORT_SYMBOL_GPL(blockdev_superblock
);
822 void __init
bdev_cache_init(void)
825 static struct vfsmount
*bd_mnt
;
827 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
828 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
829 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
831 err
= register_filesystem(&bd_type
);
833 panic("Cannot register bdev pseudo-fs");
834 bd_mnt
= kern_mount(&bd_type
);
836 panic("Cannot create bdev pseudo-fs");
837 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
841 * Most likely _very_ bad one - but then it's hardly critical for small
842 * /dev and can be fixed when somebody will need really large one.
843 * Keep in mind that it will be fed through icache hash function too.
845 static inline unsigned long hash(dev_t dev
)
847 return MAJOR(dev
)+MINOR(dev
);
850 static int bdev_test(struct inode
*inode
, void *data
)
852 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
855 static int bdev_set(struct inode
*inode
, void *data
)
857 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
861 static LIST_HEAD(all_bdevs
);
864 * If there is a bdev inode for this device, unhash it so that it gets evicted
865 * as soon as last inode reference is dropped.
867 void bdev_unhash_inode(dev_t dev
)
871 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
873 remove_inode_hash(inode
);
878 struct block_device
*bdget(dev_t dev
)
880 struct block_device
*bdev
;
883 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
884 bdev_test
, bdev_set
, &dev
);
889 bdev
= &BDEV_I(inode
)->bdev
;
891 if (inode
->i_state
& I_NEW
) {
892 bdev
->bd_contains
= NULL
;
893 bdev
->bd_super
= NULL
;
894 bdev
->bd_inode
= inode
;
895 bdev
->bd_block_size
= i_blocksize(inode
);
896 bdev
->bd_part_count
= 0;
897 bdev
->bd_invalidated
= 0;
898 inode
->i_mode
= S_IFBLK
;
900 inode
->i_bdev
= bdev
;
901 inode
->i_data
.a_ops
= &def_blk_aops
;
902 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
903 spin_lock(&bdev_lock
);
904 list_add(&bdev
->bd_list
, &all_bdevs
);
905 spin_unlock(&bdev_lock
);
906 unlock_new_inode(inode
);
911 EXPORT_SYMBOL(bdget
);
914 * bdgrab -- Grab a reference to an already referenced block device
915 * @bdev: Block device to grab a reference to.
917 struct block_device
*bdgrab(struct block_device
*bdev
)
919 ihold(bdev
->bd_inode
);
922 EXPORT_SYMBOL(bdgrab
);
924 long nr_blockdev_pages(void)
926 struct block_device
*bdev
;
928 spin_lock(&bdev_lock
);
929 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
930 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
932 spin_unlock(&bdev_lock
);
936 void bdput(struct block_device
*bdev
)
938 iput(bdev
->bd_inode
);
941 EXPORT_SYMBOL(bdput
);
943 static struct block_device
*bd_acquire(struct inode
*inode
)
945 struct block_device
*bdev
;
947 spin_lock(&bdev_lock
);
948 bdev
= inode
->i_bdev
;
949 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
951 spin_unlock(&bdev_lock
);
954 spin_unlock(&bdev_lock
);
957 * i_bdev references block device inode that was already shut down
958 * (corresponding device got removed). Remove the reference and look
959 * up block device inode again just in case new device got
960 * reestablished under the same device number.
965 bdev
= bdget(inode
->i_rdev
);
967 spin_lock(&bdev_lock
);
968 if (!inode
->i_bdev
) {
970 * We take an additional reference to bd_inode,
971 * and it's released in clear_inode() of inode.
972 * So, we can access it via ->i_mapping always
976 inode
->i_bdev
= bdev
;
977 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
979 spin_unlock(&bdev_lock
);
984 /* Call when you free inode */
986 void bd_forget(struct inode
*inode
)
988 struct block_device
*bdev
= NULL
;
990 spin_lock(&bdev_lock
);
991 if (!sb_is_blkdev_sb(inode
->i_sb
))
992 bdev
= inode
->i_bdev
;
993 inode
->i_bdev
= NULL
;
994 inode
->i_mapping
= &inode
->i_data
;
995 spin_unlock(&bdev_lock
);
1002 * bd_may_claim - test whether a block device can be claimed
1003 * @bdev: block device of interest
1004 * @whole: whole block device containing @bdev, may equal @bdev
1005 * @holder: holder trying to claim @bdev
1007 * Test whether @bdev can be claimed by @holder.
1010 * spin_lock(&bdev_lock).
1013 * %true if @bdev can be claimed, %false otherwise.
1015 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1018 if (bdev
->bd_holder
== holder
)
1019 return true; /* already a holder */
1020 else if (bdev
->bd_holder
!= NULL
)
1021 return false; /* held by someone else */
1022 else if (whole
== bdev
)
1023 return true; /* is a whole device which isn't held */
1025 else if (whole
->bd_holder
== bd_may_claim
)
1026 return true; /* is a partition of a device that is being partitioned */
1027 else if (whole
->bd_holder
!= NULL
)
1028 return false; /* is a partition of a held device */
1030 return true; /* is a partition of an un-held device */
1034 * bd_prepare_to_claim - prepare to claim a block device
1035 * @bdev: block device of interest
1036 * @whole: the whole device containing @bdev, may equal @bdev
1037 * @holder: holder trying to claim @bdev
1039 * Prepare to claim @bdev. This function fails if @bdev is already
1040 * claimed by another holder and waits if another claiming is in
1041 * progress. This function doesn't actually claim. On successful
1042 * return, the caller has ownership of bd_claiming and bd_holder[s].
1045 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1046 * it multiple times.
1049 * 0 if @bdev can be claimed, -EBUSY otherwise.
1051 static int bd_prepare_to_claim(struct block_device
*bdev
,
1052 struct block_device
*whole
, void *holder
)
1055 /* if someone else claimed, fail */
1056 if (!bd_may_claim(bdev
, whole
, holder
))
1059 /* if claiming is already in progress, wait for it to finish */
1060 if (whole
->bd_claiming
) {
1061 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1064 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1065 spin_unlock(&bdev_lock
);
1067 finish_wait(wq
, &wait
);
1068 spin_lock(&bdev_lock
);
1077 * bd_start_claiming - start claiming a block device
1078 * @bdev: block device of interest
1079 * @holder: holder trying to claim @bdev
1081 * @bdev is about to be opened exclusively. Check @bdev can be opened
1082 * exclusively and mark that an exclusive open is in progress. Each
1083 * successful call to this function must be matched with a call to
1084 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1087 * This function is used to gain exclusive access to the block device
1088 * without actually causing other exclusive open attempts to fail. It
1089 * should be used when the open sequence itself requires exclusive
1090 * access but may subsequently fail.
1096 * Pointer to the block device containing @bdev on success, ERR_PTR()
1099 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1102 struct gendisk
*disk
;
1103 struct block_device
*whole
;
1109 * @bdev might not have been initialized properly yet, look up
1110 * and grab the outer block device the hard way.
1112 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1114 return ERR_PTR(-ENXIO
);
1117 * Normally, @bdev should equal what's returned from bdget_disk()
1118 * if partno is 0; however, some drivers (floppy) use multiple
1119 * bdev's for the same physical device and @bdev may be one of the
1120 * aliases. Keep @bdev if partno is 0. This means claimer
1121 * tracking is broken for those devices but it has always been that
1125 whole
= bdget_disk(disk
, 0);
1127 whole
= bdgrab(bdev
);
1129 module_put(disk
->fops
->owner
);
1132 return ERR_PTR(-ENOMEM
);
1134 /* prepare to claim, if successful, mark claiming in progress */
1135 spin_lock(&bdev_lock
);
1137 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1139 whole
->bd_claiming
= holder
;
1140 spin_unlock(&bdev_lock
);
1143 spin_unlock(&bdev_lock
);
1145 return ERR_PTR(err
);
1150 struct bd_holder_disk
{
1151 struct list_head list
;
1152 struct gendisk
*disk
;
1156 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1157 struct gendisk
*disk
)
1159 struct bd_holder_disk
*holder
;
1161 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1162 if (holder
->disk
== disk
)
1167 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1169 return sysfs_create_link(from
, to
, kobject_name(to
));
1172 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1174 sysfs_remove_link(from
, kobject_name(to
));
1178 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1179 * @bdev: the claimed slave bdev
1180 * @disk: the holding disk
1182 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1184 * This functions creates the following sysfs symlinks.
1186 * - from "slaves" directory of the holder @disk to the claimed @bdev
1187 * - from "holders" directory of the @bdev to the holder @disk
1189 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1190 * passed to bd_link_disk_holder(), then:
1192 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1193 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1195 * The caller must have claimed @bdev before calling this function and
1196 * ensure that both @bdev and @disk are valid during the creation and
1197 * lifetime of these symlinks.
1203 * 0 on success, -errno on failure.
1205 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1207 struct bd_holder_disk
*holder
;
1210 mutex_lock(&bdev
->bd_mutex
);
1212 WARN_ON_ONCE(!bdev
->bd_holder
);
1214 /* FIXME: remove the following once add_disk() handles errors */
1215 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1218 holder
= bd_find_holder_disk(bdev
, disk
);
1224 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1230 INIT_LIST_HEAD(&holder
->list
);
1231 holder
->disk
= disk
;
1234 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1238 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1242 * bdev could be deleted beneath us which would implicitly destroy
1243 * the holder directory. Hold on to it.
1245 kobject_get(bdev
->bd_part
->holder_dir
);
1247 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1251 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1255 mutex_unlock(&bdev
->bd_mutex
);
1258 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1261 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1262 * @bdev: the calimed slave bdev
1263 * @disk: the holding disk
1265 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1270 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1272 struct bd_holder_disk
*holder
;
1274 mutex_lock(&bdev
->bd_mutex
);
1276 holder
= bd_find_holder_disk(bdev
, disk
);
1278 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1279 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1280 del_symlink(bdev
->bd_part
->holder_dir
,
1281 &disk_to_dev(disk
)->kobj
);
1282 kobject_put(bdev
->bd_part
->holder_dir
);
1283 list_del_init(&holder
->list
);
1287 mutex_unlock(&bdev
->bd_mutex
);
1289 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1293 * flush_disk - invalidates all buffer-cache entries on a disk
1295 * @bdev: struct block device to be flushed
1296 * @kill_dirty: flag to guide handling of dirty inodes
1298 * Invalidates all buffer-cache entries on a disk. It should be called
1299 * when a disk has been changed -- either by a media change or online
1302 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1304 if (__invalidate_device(bdev
, kill_dirty
)) {
1305 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1306 "resized disk %s\n",
1307 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1309 bdev
->bd_invalidated
= 1;
1313 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1314 * @disk: struct gendisk to check
1315 * @bdev: struct bdev to adjust.
1317 * This routine checks to see if the bdev size does not match the disk size
1318 * and adjusts it if it differs.
1320 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1322 loff_t disk_size
, bdev_size
;
1324 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1325 bdev_size
= i_size_read(bdev
->bd_inode
);
1326 if (disk_size
!= bdev_size
) {
1328 "%s: detected capacity change from %lld to %lld\n",
1329 disk
->disk_name
, bdev_size
, disk_size
);
1330 i_size_write(bdev
->bd_inode
, disk_size
);
1331 flush_disk(bdev
, false);
1334 EXPORT_SYMBOL(check_disk_size_change
);
1337 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1338 * @disk: struct gendisk to be revalidated
1340 * This routine is a wrapper for lower-level driver's revalidate_disk
1341 * call-backs. It is used to do common pre and post operations needed
1342 * for all revalidate_disk operations.
1344 int revalidate_disk(struct gendisk
*disk
)
1346 struct block_device
*bdev
;
1349 if (disk
->fops
->revalidate_disk
)
1350 ret
= disk
->fops
->revalidate_disk(disk
);
1351 bdev
= bdget_disk(disk
, 0);
1355 mutex_lock(&bdev
->bd_mutex
);
1356 check_disk_size_change(disk
, bdev
);
1357 bdev
->bd_invalidated
= 0;
1358 mutex_unlock(&bdev
->bd_mutex
);
1362 EXPORT_SYMBOL(revalidate_disk
);
1365 * This routine checks whether a removable media has been changed,
1366 * and invalidates all buffer-cache-entries in that case. This
1367 * is a relatively slow routine, so we have to try to minimize using
1368 * it. Thus it is called only upon a 'mount' or 'open'. This
1369 * is the best way of combining speed and utility, I think.
1370 * People changing diskettes in the middle of an operation deserve
1373 int check_disk_change(struct block_device
*bdev
)
1375 struct gendisk
*disk
= bdev
->bd_disk
;
1376 const struct block_device_operations
*bdops
= disk
->fops
;
1377 unsigned int events
;
1379 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1380 DISK_EVENT_EJECT_REQUEST
);
1381 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1384 flush_disk(bdev
, true);
1385 if (bdops
->revalidate_disk
)
1386 bdops
->revalidate_disk(bdev
->bd_disk
);
1390 EXPORT_SYMBOL(check_disk_change
);
1392 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1394 inode_lock(bdev
->bd_inode
);
1395 i_size_write(bdev
->bd_inode
, size
);
1396 inode_unlock(bdev
->bd_inode
);
1398 EXPORT_SYMBOL(bd_set_size
);
1400 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1402 static void bdev_disk_changed(struct block_device
*bdev
, bool invalidate
)
1404 if (disk_part_scan_enabled(bdev
->bd_disk
)) {
1406 invalidate_partitions(bdev
->bd_disk
, bdev
);
1408 rescan_partitions(bdev
->bd_disk
, bdev
);
1410 check_disk_size_change(bdev
->bd_disk
, bdev
);
1411 bdev
->bd_invalidated
= 0;
1418 * mutex_lock(part->bd_mutex)
1419 * mutex_lock_nested(whole->bd_mutex, 1)
1422 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1424 struct gendisk
*disk
;
1425 struct module
*owner
;
1430 if (mode
& FMODE_READ
)
1432 if (mode
& FMODE_WRITE
)
1435 * hooks: /n/, see "layering violations".
1438 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1448 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1451 owner
= disk
->fops
->owner
;
1453 disk_block_events(disk
);
1454 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1455 if (!bdev
->bd_openers
) {
1456 bdev
->bd_disk
= disk
;
1457 bdev
->bd_queue
= disk
->queue
;
1458 bdev
->bd_contains
= bdev
;
1459 bdev
->bd_partno
= partno
;
1463 bdev
->bd_part
= disk_get_part(disk
, partno
);
1468 if (disk
->fops
->open
) {
1469 ret
= disk
->fops
->open(bdev
, mode
);
1470 if (ret
== -ERESTARTSYS
) {
1471 /* Lost a race with 'disk' being
1472 * deleted, try again.
1475 disk_put_part(bdev
->bd_part
);
1476 bdev
->bd_part
= NULL
;
1477 bdev
->bd_disk
= NULL
;
1478 bdev
->bd_queue
= NULL
;
1479 mutex_unlock(&bdev
->bd_mutex
);
1480 disk_unblock_events(disk
);
1488 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1489 set_init_blocksize(bdev
);
1493 * If the device is invalidated, rescan partition
1494 * if open succeeded or failed with -ENOMEDIUM.
1495 * The latter is necessary to prevent ghost
1496 * partitions on a removed medium.
1498 if (bdev
->bd_invalidated
&&
1499 (!ret
|| ret
== -ENOMEDIUM
))
1500 bdev_disk_changed(bdev
, ret
== -ENOMEDIUM
);
1505 struct block_device
*whole
;
1506 whole
= bdget_disk(disk
, 0);
1511 ret
= __blkdev_get(whole
, mode
, 1);
1514 bdev
->bd_contains
= whole
;
1515 bdev
->bd_part
= disk_get_part(disk
, partno
);
1516 if (!(disk
->flags
& GENHD_FL_UP
) ||
1517 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1521 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1522 set_init_blocksize(bdev
);
1525 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1526 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1528 if (bdev
->bd_contains
== bdev
) {
1530 if (bdev
->bd_disk
->fops
->open
)
1531 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1532 /* the same as first opener case, read comment there */
1533 if (bdev
->bd_invalidated
&&
1534 (!ret
|| ret
== -ENOMEDIUM
))
1535 bdev_disk_changed(bdev
, ret
== -ENOMEDIUM
);
1537 goto out_unlock_bdev
;
1539 /* only one opener holds refs to the module and disk */
1545 bdev
->bd_part_count
++;
1546 mutex_unlock(&bdev
->bd_mutex
);
1547 disk_unblock_events(disk
);
1551 disk_put_part(bdev
->bd_part
);
1552 bdev
->bd_disk
= NULL
;
1553 bdev
->bd_part
= NULL
;
1554 bdev
->bd_queue
= NULL
;
1555 if (bdev
!= bdev
->bd_contains
)
1556 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1557 bdev
->bd_contains
= NULL
;
1559 mutex_unlock(&bdev
->bd_mutex
);
1560 disk_unblock_events(disk
);
1570 * blkdev_get - open a block device
1571 * @bdev: block_device to open
1572 * @mode: FMODE_* mask
1573 * @holder: exclusive holder identifier
1575 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1576 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1577 * @holder is invalid. Exclusive opens may nest for the same @holder.
1579 * On success, the reference count of @bdev is unchanged. On failure,
1586 * 0 on success, -errno on failure.
1588 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1590 struct block_device
*whole
= NULL
;
1593 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1595 if ((mode
& FMODE_EXCL
) && holder
) {
1596 whole
= bd_start_claiming(bdev
, holder
);
1597 if (IS_ERR(whole
)) {
1599 return PTR_ERR(whole
);
1603 res
= __blkdev_get(bdev
, mode
, 0);
1606 struct gendisk
*disk
= whole
->bd_disk
;
1608 /* finish claiming */
1609 mutex_lock(&bdev
->bd_mutex
);
1610 spin_lock(&bdev_lock
);
1613 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1615 * Note that for a whole device bd_holders
1616 * will be incremented twice, and bd_holder
1617 * will be set to bd_may_claim before being
1620 whole
->bd_holders
++;
1621 whole
->bd_holder
= bd_may_claim
;
1623 bdev
->bd_holder
= holder
;
1626 /* tell others that we're done */
1627 BUG_ON(whole
->bd_claiming
!= holder
);
1628 whole
->bd_claiming
= NULL
;
1629 wake_up_bit(&whole
->bd_claiming
, 0);
1631 spin_unlock(&bdev_lock
);
1634 * Block event polling for write claims if requested. Any
1635 * write holder makes the write_holder state stick until
1636 * all are released. This is good enough and tracking
1637 * individual writeable reference is too fragile given the
1638 * way @mode is used in blkdev_get/put().
1640 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1641 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1642 bdev
->bd_write_holder
= true;
1643 disk_block_events(disk
);
1646 mutex_unlock(&bdev
->bd_mutex
);
1652 EXPORT_SYMBOL(blkdev_get
);
1655 * blkdev_get_by_path - open a block device by name
1656 * @path: path to the block device to open
1657 * @mode: FMODE_* mask
1658 * @holder: exclusive holder identifier
1660 * Open the blockdevice described by the device file at @path. @mode
1661 * and @holder are identical to blkdev_get().
1663 * On success, the returned block_device has reference count of one.
1669 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1671 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1674 struct block_device
*bdev
;
1678 if (mode
& FMODE_READ
)
1680 if (mode
& FMODE_WRITE
)
1682 bdev
= lookup_bdev(path
, perm
);
1686 err
= blkdev_get(bdev
, mode
, holder
);
1688 return ERR_PTR(err
);
1690 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1691 blkdev_put(bdev
, mode
);
1692 return ERR_PTR(-EACCES
);
1697 EXPORT_SYMBOL(blkdev_get_by_path
);
1700 * blkdev_get_by_dev - open a block device by device number
1701 * @dev: device number of block device to open
1702 * @mode: FMODE_* mask
1703 * @holder: exclusive holder identifier
1705 * Open the blockdevice described by device number @dev. @mode and
1706 * @holder are identical to blkdev_get().
1708 * Use it ONLY if you really do not have anything better - i.e. when
1709 * you are behind a truly sucky interface and all you are given is a
1710 * device number. _Never_ to be used for internal purposes. If you
1711 * ever need it - reconsider your API.
1713 * On success, the returned block_device has reference count of one.
1719 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1721 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1723 struct block_device
*bdev
;
1728 return ERR_PTR(-ENOMEM
);
1730 err
= blkdev_get(bdev
, mode
, holder
);
1732 return ERR_PTR(err
);
1736 EXPORT_SYMBOL(blkdev_get_by_dev
);
1738 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1740 struct block_device
*bdev
;
1743 * Preserve backwards compatibility and allow large file access
1744 * even if userspace doesn't ask for it explicitly. Some mkfs
1745 * binary needs it. We might want to drop this workaround
1746 * during an unstable branch.
1748 filp
->f_flags
|= O_LARGEFILE
;
1750 filp
->f_mode
|= FMODE_NOWAIT
;
1752 if (filp
->f_flags
& O_NDELAY
)
1753 filp
->f_mode
|= FMODE_NDELAY
;
1754 if (filp
->f_flags
& O_EXCL
)
1755 filp
->f_mode
|= FMODE_EXCL
;
1756 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1757 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1759 bdev
= bd_acquire(inode
);
1764 * A negative i_writecount for bdev->bd_inode means that the bdev
1765 * or one of its paritions is mounted in a user namespace. Deny
1766 * writing for non-root in this case, otherwise an unprivileged
1767 * user can attack the kernel by modifying the backing store of a
1768 * mounted filesystem.
1770 if ((filp
->f_mode
& FMODE_WRITE
) &&
1771 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
) &&
1772 !atomic_inc_unless_negative(&bdev
->bd_inode
->i_writecount
)) {
1777 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1778 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
1780 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1783 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1785 struct gendisk
*disk
= bdev
->bd_disk
;
1786 struct block_device
*victim
= NULL
;
1788 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1790 bdev
->bd_part_count
--;
1792 if (!--bdev
->bd_openers
) {
1793 WARN_ON_ONCE(bdev
->bd_holders
);
1794 sync_blockdev(bdev
);
1797 bdev_write_inode(bdev
);
1799 if (bdev
->bd_contains
== bdev
) {
1800 if (disk
->fops
->release
)
1801 disk
->fops
->release(disk
, mode
);
1803 if (!bdev
->bd_openers
) {
1804 struct module
*owner
= disk
->fops
->owner
;
1806 disk_put_part(bdev
->bd_part
);
1807 bdev
->bd_part
= NULL
;
1808 bdev
->bd_disk
= NULL
;
1809 if (bdev
!= bdev
->bd_contains
)
1810 victim
= bdev
->bd_contains
;
1811 bdev
->bd_contains
= NULL
;
1816 mutex_unlock(&bdev
->bd_mutex
);
1819 __blkdev_put(victim
, mode
, 1);
1822 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1824 mutex_lock(&bdev
->bd_mutex
);
1826 if (mode
& FMODE_EXCL
) {
1830 * Release a claim on the device. The holder fields
1831 * are protected with bdev_lock. bd_mutex is to
1832 * synchronize disk_holder unlinking.
1834 spin_lock(&bdev_lock
);
1836 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1837 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1839 /* bd_contains might point to self, check in a separate step */
1840 if ((bdev_free
= !bdev
->bd_holders
))
1841 bdev
->bd_holder
= NULL
;
1842 if (!bdev
->bd_contains
->bd_holders
)
1843 bdev
->bd_contains
->bd_holder
= NULL
;
1845 spin_unlock(&bdev_lock
);
1848 * If this was the last claim, remove holder link and
1849 * unblock evpoll if it was a write holder.
1851 if (bdev_free
&& bdev
->bd_write_holder
) {
1852 disk_unblock_events(bdev
->bd_disk
);
1853 bdev
->bd_write_holder
= false;
1858 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1859 * event. This is to ensure detection of media removal commanded
1860 * from userland - e.g. eject(1).
1862 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1864 mutex_unlock(&bdev
->bd_mutex
);
1866 __blkdev_put(bdev
, mode
, 0);
1868 EXPORT_SYMBOL(blkdev_put
);
1870 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1872 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1873 if (filp
->f_mode
& FMODE_WRITE
&&
1874 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
))
1875 atomic_dec(&bdev
->bd_inode
->i_writecount
);
1876 blkdev_put(bdev
, filp
->f_mode
);
1880 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1882 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1883 fmode_t mode
= file
->f_mode
;
1886 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1887 * to updated it before every ioctl.
1889 if (file
->f_flags
& O_NDELAY
)
1890 mode
|= FMODE_NDELAY
;
1892 mode
&= ~FMODE_NDELAY
;
1894 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1898 * Write data to the block device. Only intended for the block device itself
1899 * and the raw driver which basically is a fake block device.
1901 * Does not take i_mutex for the write and thus is not for general purpose
1904 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1906 struct file
*file
= iocb
->ki_filp
;
1907 struct inode
*bd_inode
= bdev_file_inode(file
);
1908 loff_t size
= i_size_read(bd_inode
);
1909 struct blk_plug plug
;
1912 if (bdev_read_only(I_BDEV(bd_inode
)))
1915 if (!iov_iter_count(from
))
1918 if (iocb
->ki_pos
>= size
)
1921 if ((iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_DIRECT
)) == IOCB_NOWAIT
)
1924 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1926 blk_start_plug(&plug
);
1927 ret
= __generic_file_write_iter(iocb
, from
);
1929 ret
= generic_write_sync(iocb
, ret
);
1930 blk_finish_plug(&plug
);
1933 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1935 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1937 struct file
*file
= iocb
->ki_filp
;
1938 struct inode
*bd_inode
= bdev_file_inode(file
);
1939 loff_t size
= i_size_read(bd_inode
);
1940 loff_t pos
= iocb
->ki_pos
;
1946 iov_iter_truncate(to
, size
);
1947 return generic_file_read_iter(iocb
, to
);
1949 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1952 * Try to release a page associated with block device when the system
1953 * is under memory pressure.
1955 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1957 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1959 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1960 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1962 return try_to_free_buffers(page
);
1965 static int blkdev_writepages(struct address_space
*mapping
,
1966 struct writeback_control
*wbc
)
1968 if (dax_mapping(mapping
)) {
1969 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1971 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1973 return generic_writepages(mapping
, wbc
);
1976 static const struct address_space_operations def_blk_aops
= {
1977 .readpage
= blkdev_readpage
,
1978 .readpages
= blkdev_readpages
,
1979 .writepage
= blkdev_writepage
,
1980 .write_begin
= blkdev_write_begin
,
1981 .write_end
= blkdev_write_end
,
1982 .writepages
= blkdev_writepages
,
1983 .releasepage
= blkdev_releasepage
,
1984 .direct_IO
= blkdev_direct_IO
,
1985 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1988 #define BLKDEV_FALLOC_FL_SUPPORTED \
1989 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1990 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1992 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
1995 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1996 struct address_space
*mapping
;
1997 loff_t end
= start
+ len
- 1;
2001 /* Fail if we don't recognize the flags. */
2002 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
2005 /* Don't go off the end of the device. */
2006 isize
= i_size_read(bdev
->bd_inode
);
2010 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2011 len
= isize
- start
;
2012 end
= start
+ len
- 1;
2018 * Don't allow IO that isn't aligned to logical block size.
2020 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2023 /* Invalidate the page cache, including dirty pages. */
2024 mapping
= bdev
->bd_inode
->i_mapping
;
2025 truncate_inode_pages_range(mapping
, start
, end
);
2028 case FALLOC_FL_ZERO_RANGE
:
2029 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2030 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2031 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
2033 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2034 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2035 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
2037 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2038 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2048 * Invalidate again; if someone wandered in and dirtied a page,
2049 * the caller will be given -EBUSY. The third argument is
2050 * inclusive, so the rounding here is safe.
2052 return invalidate_inode_pages2_range(mapping
,
2053 start
>> PAGE_SHIFT
,
2057 const struct file_operations def_blk_fops
= {
2058 .open
= blkdev_open
,
2059 .release
= blkdev_close
,
2060 .llseek
= block_llseek
,
2061 .read_iter
= blkdev_read_iter
,
2062 .write_iter
= blkdev_write_iter
,
2063 .mmap
= generic_file_mmap
,
2064 .fsync
= blkdev_fsync
,
2065 .unlocked_ioctl
= block_ioctl
,
2066 #ifdef CONFIG_COMPAT
2067 .compat_ioctl
= compat_blkdev_ioctl
,
2069 .splice_read
= generic_file_splice_read
,
2070 .splice_write
= iter_file_splice_write
,
2071 .fallocate
= blkdev_fallocate
,
2074 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2077 mm_segment_t old_fs
= get_fs();
2079 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2084 EXPORT_SYMBOL(ioctl_by_bdev
);
2087 * lookup_bdev - lookup a struct block_device by name
2088 * @pathname: special file representing the block device
2089 * @mask: rights to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
2091 * Get a reference to the blockdevice at @pathname in the current
2092 * namespace if possible and return it. Return ERR_PTR(error)
2093 * otherwise. If @mask is non-zero, check for access rights to the
2094 * inode at @pathname.
2096 struct block_device
*lookup_bdev(const char *pathname
, int mask
)
2098 struct block_device
*bdev
;
2099 struct inode
*inode
;
2103 if (!pathname
|| !*pathname
)
2104 return ERR_PTR(-EINVAL
);
2106 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2108 return ERR_PTR(error
);
2110 inode
= d_backing_inode(path
.dentry
);
2111 if (mask
!= 0 && !capable(CAP_SYS_ADMIN
)) {
2112 error
= __inode_permission(inode
, mask
);
2117 if (!S_ISBLK(inode
->i_mode
))
2120 if (!may_open_dev(&path
))
2123 bdev
= bd_acquire(inode
);
2130 bdev
= ERR_PTR(error
);
2133 EXPORT_SYMBOL(lookup_bdev
);
2135 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2137 struct super_block
*sb
= get_super(bdev
);
2142 * no need to lock the super, get_super holds the
2143 * read mutex so the filesystem cannot go away
2144 * under us (->put_super runs with the write lock
2147 shrink_dcache_sb(sb
);
2148 res
= invalidate_inodes(sb
, kill_dirty
);
2151 invalidate_bdev(bdev
);
2154 EXPORT_SYMBOL(__invalidate_device
);
2156 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2158 struct inode
*inode
, *old_inode
= NULL
;
2160 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2161 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2162 struct address_space
*mapping
= inode
->i_mapping
;
2163 struct block_device
*bdev
;
2165 spin_lock(&inode
->i_lock
);
2166 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2167 mapping
->nrpages
== 0) {
2168 spin_unlock(&inode
->i_lock
);
2172 spin_unlock(&inode
->i_lock
);
2173 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2175 * We hold a reference to 'inode' so it couldn't have been
2176 * removed from s_inodes list while we dropped the
2177 * s_inode_list_lock We cannot iput the inode now as we can
2178 * be holding the last reference and we cannot iput it under
2179 * s_inode_list_lock. So we keep the reference and iput it
2184 bdev
= I_BDEV(inode
);
2186 mutex_lock(&bdev
->bd_mutex
);
2187 if (bdev
->bd_openers
)
2189 mutex_unlock(&bdev
->bd_mutex
);
2191 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2193 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);