1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/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/pseudo_fs.h>
29 #include <linux/uio.h>
30 #include <linux/namei.h>
31 #include <linux/log2.h>
32 #include <linux/cleancache.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
40 struct block_device bdev
;
41 struct inode vfs_inode
;
44 static const struct address_space_operations def_blk_aops
;
46 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
48 return container_of(inode
, struct bdev_inode
, vfs_inode
);
51 struct block_device
*I_BDEV(struct inode
*inode
)
53 return &BDEV_I(inode
)->bdev
;
55 EXPORT_SYMBOL(I_BDEV
);
57 static void bdev_write_inode(struct block_device
*bdev
)
59 struct inode
*inode
= bdev
->bd_inode
;
62 spin_lock(&inode
->i_lock
);
63 while (inode
->i_state
& I_DIRTY
) {
64 spin_unlock(&inode
->i_lock
);
65 ret
= write_inode_now(inode
, true);
67 char name
[BDEVNAME_SIZE
];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev
, name
), ret
);
72 spin_lock(&inode
->i_lock
);
74 spin_unlock(&inode
->i_lock
);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 void kill_bdev(struct block_device
*bdev
)
80 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
82 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
86 truncate_inode_pages(mapping
, 0);
88 EXPORT_SYMBOL(kill_bdev
);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device
*bdev
)
93 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
95 if (mapping
->nrpages
) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping
, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping
);
105 EXPORT_SYMBOL(invalidate_bdev
);
107 static void set_init_blocksize(struct block_device
*bdev
)
109 unsigned bsize
= bdev_logical_block_size(bdev
);
110 loff_t size
= i_size_read(bdev
->bd_inode
);
112 while (bsize
< PAGE_SIZE
) {
117 bdev
->bd_block_size
= bsize
;
118 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
121 int set_blocksize(struct block_device
*bdev
, int size
)
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
124 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
127 /* Size cannot be smaller than the size supported by the device */
128 if (size
< bdev_logical_block_size(bdev
))
131 /* Don't change the size if it is same as current */
132 if (bdev
->bd_block_size
!= size
) {
134 bdev
->bd_block_size
= size
;
135 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
141 EXPORT_SYMBOL(set_blocksize
);
143 int sb_set_blocksize(struct super_block
*sb
, int size
)
145 if (set_blocksize(sb
->s_bdev
, size
))
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb
->s_blocksize
= size
;
150 sb
->s_blocksize_bits
= blksize_bits(size
);
151 return sb
->s_blocksize
;
154 EXPORT_SYMBOL(sb_set_blocksize
);
156 int sb_min_blocksize(struct super_block
*sb
, int size
)
158 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
161 return sb_set_blocksize(sb
, size
);
164 EXPORT_SYMBOL(sb_min_blocksize
);
167 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
168 struct buffer_head
*bh
, int create
)
170 bh
->b_bdev
= I_BDEV(inode
);
171 bh
->b_blocknr
= iblock
;
172 set_buffer_mapped(bh
);
176 static struct inode
*bdev_file_inode(struct file
*file
)
178 return file
->f_mapping
->host
;
181 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
183 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
185 /* avoid the need for a I/O completion work item */
186 if (iocb
->ki_flags
& IOCB_DSYNC
)
191 #define DIO_INLINE_BIO_VECS 4
193 static void blkdev_bio_end_io_simple(struct bio
*bio
)
195 struct task_struct
*waiter
= bio
->bi_private
;
197 WRITE_ONCE(bio
->bi_private
, NULL
);
198 blk_wake_io_task(waiter
);
202 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
205 struct file
*file
= iocb
->ki_filp
;
206 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
207 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
;
208 loff_t pos
= iocb
->ki_pos
;
209 bool should_dirty
= false;
214 if ((pos
| iov_iter_alignment(iter
)) &
215 (bdev_logical_block_size(bdev
) - 1))
218 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
221 vecs
= kmalloc_array(nr_pages
, sizeof(struct bio_vec
),
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
;
233 bio
.bi_ioprio
= iocb
->ki_ioprio
;
235 ret
= bio_iov_iter_get_pages(&bio
, iter
);
238 ret
= bio
.bi_iter
.bi_size
;
240 if (iov_iter_rw(iter
) == READ
) {
241 bio
.bi_opf
= REQ_OP_READ
;
242 if (iter_is_iovec(iter
))
245 bio
.bi_opf
= dio_bio_write_op(iocb
);
246 task_io_account_write(ret
);
248 if (iocb
->ki_flags
& IOCB_HIPRI
)
249 bio_set_polled(&bio
, iocb
);
251 qc
= submit_bio(&bio
);
253 set_current_state(TASK_UNINTERRUPTIBLE
);
254 if (!READ_ONCE(bio
.bi_private
))
256 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
257 !blk_poll(bdev_get_queue(bdev
), qc
, true))
260 __set_current_state(TASK_RUNNING
);
262 bio_release_pages(&bio
, should_dirty
);
263 if (unlikely(bio
.bi_status
))
264 ret
= blk_status_to_errno(bio
.bi_status
);
267 if (vecs
!= inline_vecs
)
278 struct task_struct
*waiter
;
283 bool should_dirty
: 1;
288 static struct bio_set blkdev_dio_pool
;
290 static int blkdev_iopoll(struct kiocb
*kiocb
, bool wait
)
292 struct block_device
*bdev
= I_BDEV(kiocb
->ki_filp
->f_mapping
->host
);
293 struct request_queue
*q
= bdev_get_queue(bdev
);
295 return blk_poll(q
, READ_ONCE(kiocb
->ki_cookie
), wait
);
298 static void blkdev_bio_end_io(struct bio
*bio
)
300 struct blkdev_dio
*dio
= bio
->bi_private
;
301 bool should_dirty
= dio
->should_dirty
;
303 if (bio
->bi_status
&& !dio
->bio
.bi_status
)
304 dio
->bio
.bi_status
= bio
->bi_status
;
306 if (!dio
->multi_bio
|| atomic_dec_and_test(&dio
->ref
)) {
308 struct kiocb
*iocb
= dio
->iocb
;
311 if (likely(!dio
->bio
.bi_status
)) {
315 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
318 dio
->iocb
->ki_complete(iocb
, ret
, 0);
322 struct task_struct
*waiter
= dio
->waiter
;
324 WRITE_ONCE(dio
->waiter
, NULL
);
325 blk_wake_io_task(waiter
);
330 bio_check_pages_dirty(bio
);
332 bio_release_pages(bio
, false);
338 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
340 struct file
*file
= iocb
->ki_filp
;
341 struct inode
*inode
= bdev_file_inode(file
);
342 struct block_device
*bdev
= I_BDEV(inode
);
343 struct blk_plug plug
;
344 struct blkdev_dio
*dio
;
346 bool is_poll
= (iocb
->ki_flags
& IOCB_HIPRI
) != 0;
347 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
348 loff_t pos
= iocb
->ki_pos
;
349 blk_qc_t qc
= BLK_QC_T_NONE
;
352 if ((pos
| iov_iter_alignment(iter
)) &
353 (bdev_logical_block_size(bdev
) - 1))
356 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, &blkdev_dio_pool
);
358 dio
= container_of(bio
, struct blkdev_dio
, bio
);
359 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
361 dio
->waiter
= current
;
368 dio
->multi_bio
= false;
369 dio
->should_dirty
= is_read
&& iter_is_iovec(iter
);
372 * Don't plug for HIPRI/polled IO, as those should go straight
376 blk_start_plug(&plug
);
379 bio_set_dev(bio
, bdev
);
380 bio
->bi_iter
.bi_sector
= pos
>> 9;
381 bio
->bi_write_hint
= iocb
->ki_hint
;
382 bio
->bi_private
= dio
;
383 bio
->bi_end_io
= blkdev_bio_end_io
;
384 bio
->bi_ioprio
= iocb
->ki_ioprio
;
386 ret
= bio_iov_iter_get_pages(bio
, iter
);
388 bio
->bi_status
= BLK_STS_IOERR
;
394 bio
->bi_opf
= REQ_OP_READ
;
395 if (dio
->should_dirty
)
396 bio_set_pages_dirty(bio
);
398 bio
->bi_opf
= dio_bio_write_op(iocb
);
399 task_io_account_write(bio
->bi_iter
.bi_size
);
402 dio
->size
+= bio
->bi_iter
.bi_size
;
403 pos
+= bio
->bi_iter
.bi_size
;
405 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
409 if (iocb
->ki_flags
& IOCB_HIPRI
) {
410 bio_set_polled(bio
, iocb
);
414 qc
= submit_bio(bio
);
417 WRITE_ONCE(iocb
->ki_cookie
, qc
);
421 if (!dio
->multi_bio
) {
423 * AIO needs an extra reference to ensure the dio
424 * structure which is embedded into the first bio
429 dio
->multi_bio
= true;
430 atomic_set(&dio
->ref
, 2);
432 atomic_inc(&dio
->ref
);
436 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
440 blk_finish_plug(&plug
);
446 set_current_state(TASK_UNINTERRUPTIBLE
);
447 if (!READ_ONCE(dio
->waiter
))
450 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
451 !blk_poll(bdev_get_queue(bdev
), qc
, true))
454 __set_current_state(TASK_RUNNING
);
457 ret
= blk_status_to_errno(dio
->bio
.bi_status
);
466 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
470 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
473 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
474 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
476 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
479 static __init
int blkdev_init(void)
481 return bioset_init(&blkdev_dio_pool
, 4, offsetof(struct blkdev_dio
, bio
), BIOSET_NEED_BVECS
);
483 module_init(blkdev_init
);
485 int __sync_blockdev(struct block_device
*bdev
, int wait
)
490 return filemap_flush(bdev
->bd_inode
->i_mapping
);
491 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
495 * Write out and wait upon all the dirty data associated with a block
496 * device via its mapping. Does not take the superblock lock.
498 int sync_blockdev(struct block_device
*bdev
)
500 return __sync_blockdev(bdev
, 1);
502 EXPORT_SYMBOL(sync_blockdev
);
505 * Write out and wait upon all dirty data associated with this
506 * device. Filesystem data as well as the underlying block
507 * device. Takes the superblock lock.
509 int fsync_bdev(struct block_device
*bdev
)
511 struct super_block
*sb
= get_super(bdev
);
513 int res
= sync_filesystem(sb
);
517 return sync_blockdev(bdev
);
519 EXPORT_SYMBOL(fsync_bdev
);
522 * freeze_bdev -- lock a filesystem and force it into a consistent state
523 * @bdev: blockdevice to lock
525 * If a superblock is found on this device, we take the s_umount semaphore
526 * on it to make sure nobody unmounts until the snapshot creation is done.
527 * The reference counter (bd_fsfreeze_count) guarantees that only the last
528 * unfreeze process can unfreeze the frozen filesystem actually when multiple
529 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
530 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
533 struct super_block
*freeze_bdev(struct block_device
*bdev
)
535 struct super_block
*sb
;
538 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
539 if (++bdev
->bd_fsfreeze_count
> 1) {
541 * We don't even need to grab a reference - the first call
542 * to freeze_bdev grab an active reference and only the last
543 * thaw_bdev drops it.
545 sb
= get_super(bdev
);
548 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
552 sb
= get_active_super(bdev
);
555 if (sb
->s_op
->freeze_super
)
556 error
= sb
->s_op
->freeze_super(sb
);
558 error
= freeze_super(sb
);
560 deactivate_super(sb
);
561 bdev
->bd_fsfreeze_count
--;
562 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
563 return ERR_PTR(error
);
565 deactivate_super(sb
);
568 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
569 return sb
; /* thaw_bdev releases s->s_umount */
571 EXPORT_SYMBOL(freeze_bdev
);
574 * thaw_bdev -- unlock filesystem
575 * @bdev: blockdevice to unlock
576 * @sb: associated superblock
578 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
580 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
584 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
585 if (!bdev
->bd_fsfreeze_count
)
589 if (--bdev
->bd_fsfreeze_count
> 0)
595 if (sb
->s_op
->thaw_super
)
596 error
= sb
->s_op
->thaw_super(sb
);
598 error
= thaw_super(sb
);
600 bdev
->bd_fsfreeze_count
++;
602 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
605 EXPORT_SYMBOL(thaw_bdev
);
607 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
609 return block_write_full_page(page
, blkdev_get_block
, wbc
);
612 static int blkdev_readpage(struct file
* file
, struct page
* page
)
614 return block_read_full_page(page
, blkdev_get_block
);
617 static void blkdev_readahead(struct readahead_control
*rac
)
619 mpage_readahead(rac
, blkdev_get_block
);
622 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
623 loff_t pos
, unsigned len
, unsigned flags
,
624 struct page
**pagep
, void **fsdata
)
626 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
630 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
631 loff_t pos
, unsigned len
, unsigned copied
,
632 struct page
*page
, void *fsdata
)
635 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
645 * for a block special file file_inode(file)->i_size is zero
646 * so we compute the size by hand (just as in block_read/write above)
648 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
650 struct inode
*bd_inode
= bdev_file_inode(file
);
653 inode_lock(bd_inode
);
654 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
655 inode_unlock(bd_inode
);
659 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
661 struct inode
*bd_inode
= bdev_file_inode(filp
);
662 struct block_device
*bdev
= I_BDEV(bd_inode
);
665 error
= file_write_and_wait_range(filp
, start
, end
);
670 * There is no need to serialise calls to blkdev_issue_flush with
671 * i_mutex and doing so causes performance issues with concurrent
672 * O_SYNC writers to a block device.
674 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
675 if (error
== -EOPNOTSUPP
)
680 EXPORT_SYMBOL(blkdev_fsync
);
683 * bdev_read_page() - Start reading a page from a block device
684 * @bdev: The device to read the page from
685 * @sector: The offset on the device to read the page to (need not be aligned)
686 * @page: The page to read
688 * On entry, the page should be locked. It will be unlocked when the page
689 * has been read. If the block driver implements rw_page synchronously,
690 * that will be true on exit from this function, but it need not be.
692 * Errors returned by this function are usually "soft", eg out of memory, or
693 * queue full; callers should try a different route to read this page rather
694 * than propagate an error back up the stack.
696 * Return: negative errno if an error occurs, 0 if submission was successful.
698 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
701 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
702 int result
= -EOPNOTSUPP
;
704 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
707 result
= blk_queue_enter(bdev
->bd_queue
, 0);
710 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
712 blk_queue_exit(bdev
->bd_queue
);
715 EXPORT_SYMBOL_GPL(bdev_read_page
);
718 * bdev_write_page() - Start writing a page to a block device
719 * @bdev: The device to write the page to
720 * @sector: The offset on the device to write the page to (need not be aligned)
721 * @page: The page to write
722 * @wbc: The writeback_control for the write
724 * On entry, the page should be locked and not currently under writeback.
725 * On exit, if the write started successfully, the page will be unlocked and
726 * under writeback. If the write failed already (eg the driver failed to
727 * queue the page to the device), the page will still be locked. If the
728 * caller is a ->writepage implementation, it will need to unlock the page.
730 * Errors returned by this function are usually "soft", eg out of memory, or
731 * queue full; callers should try a different route to write this page rather
732 * than propagate an error back up the stack.
734 * Return: negative errno if an error occurs, 0 if submission was successful.
736 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
737 struct page
*page
, struct writeback_control
*wbc
)
740 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
742 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
744 result
= blk_queue_enter(bdev
->bd_queue
, 0);
748 set_page_writeback(page
);
749 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
752 end_page_writeback(page
);
754 clean_page_buffers(page
);
757 blk_queue_exit(bdev
->bd_queue
);
760 EXPORT_SYMBOL_GPL(bdev_write_page
);
766 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
767 static struct kmem_cache
* bdev_cachep __read_mostly
;
769 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
771 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
774 return &ei
->vfs_inode
;
777 static void bdev_free_inode(struct inode
*inode
)
779 kmem_cache_free(bdev_cachep
, BDEV_I(inode
));
782 static void init_once(void *foo
)
784 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
785 struct block_device
*bdev
= &ei
->bdev
;
787 memset(bdev
, 0, sizeof(*bdev
));
788 mutex_init(&bdev
->bd_mutex
);
789 INIT_LIST_HEAD(&bdev
->bd_list
);
791 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
793 bdev
->bd_bdi
= &noop_backing_dev_info
;
794 inode_init_once(&ei
->vfs_inode
);
795 /* Initialize mutex for freeze. */
796 mutex_init(&bdev
->bd_fsfreeze_mutex
);
799 static void bdev_evict_inode(struct inode
*inode
)
801 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
802 truncate_inode_pages_final(&inode
->i_data
);
803 invalidate_inode_buffers(inode
); /* is it needed here? */
805 spin_lock(&bdev_lock
);
806 list_del_init(&bdev
->bd_list
);
807 spin_unlock(&bdev_lock
);
808 /* Detach inode from wb early as bdi_put() may free bdi->wb */
809 inode_detach_wb(inode
);
810 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
811 bdi_put(bdev
->bd_bdi
);
812 bdev
->bd_bdi
= &noop_backing_dev_info
;
816 static const struct super_operations bdev_sops
= {
817 .statfs
= simple_statfs
,
818 .alloc_inode
= bdev_alloc_inode
,
819 .free_inode
= bdev_free_inode
,
820 .drop_inode
= generic_delete_inode
,
821 .evict_inode
= bdev_evict_inode
,
824 static int bd_init_fs_context(struct fs_context
*fc
)
826 struct pseudo_fs_context
*ctx
= init_pseudo(fc
, BDEVFS_MAGIC
);
829 fc
->s_iflags
|= SB_I_CGROUPWB
;
830 ctx
->ops
= &bdev_sops
;
834 static struct file_system_type bd_type
= {
836 .init_fs_context
= bd_init_fs_context
,
837 .kill_sb
= kill_anon_super
,
840 struct super_block
*blockdev_superblock __read_mostly
;
841 EXPORT_SYMBOL_GPL(blockdev_superblock
);
843 void __init
bdev_cache_init(void)
846 static struct vfsmount
*bd_mnt
;
848 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
849 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
850 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
852 err
= register_filesystem(&bd_type
);
854 panic("Cannot register bdev pseudo-fs");
855 bd_mnt
= kern_mount(&bd_type
);
857 panic("Cannot create bdev pseudo-fs");
858 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
862 * Most likely _very_ bad one - but then it's hardly critical for small
863 * /dev and can be fixed when somebody will need really large one.
864 * Keep in mind that it will be fed through icache hash function too.
866 static inline unsigned long hash(dev_t dev
)
868 return MAJOR(dev
)+MINOR(dev
);
871 static int bdev_test(struct inode
*inode
, void *data
)
873 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
876 static int bdev_set(struct inode
*inode
, void *data
)
878 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
882 static LIST_HEAD(all_bdevs
);
885 * If there is a bdev inode for this device, unhash it so that it gets evicted
886 * as soon as last inode reference is dropped.
888 void bdev_unhash_inode(dev_t dev
)
892 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
894 remove_inode_hash(inode
);
899 struct block_device
*bdget(dev_t dev
)
901 struct block_device
*bdev
;
904 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
905 bdev_test
, bdev_set
, &dev
);
910 bdev
= &BDEV_I(inode
)->bdev
;
912 if (inode
->i_state
& I_NEW
) {
913 bdev
->bd_contains
= NULL
;
914 bdev
->bd_super
= NULL
;
915 bdev
->bd_inode
= inode
;
916 bdev
->bd_block_size
= i_blocksize(inode
);
917 bdev
->bd_part_count
= 0;
918 bdev
->bd_invalidated
= 0;
919 inode
->i_mode
= S_IFBLK
;
921 inode
->i_bdev
= bdev
;
922 inode
->i_data
.a_ops
= &def_blk_aops
;
923 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
924 spin_lock(&bdev_lock
);
925 list_add(&bdev
->bd_list
, &all_bdevs
);
926 spin_unlock(&bdev_lock
);
927 unlock_new_inode(inode
);
932 EXPORT_SYMBOL(bdget
);
935 * bdgrab -- Grab a reference to an already referenced block device
936 * @bdev: Block device to grab a reference to.
938 struct block_device
*bdgrab(struct block_device
*bdev
)
940 ihold(bdev
->bd_inode
);
943 EXPORT_SYMBOL(bdgrab
);
945 long nr_blockdev_pages(void)
947 struct block_device
*bdev
;
949 spin_lock(&bdev_lock
);
950 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
951 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
953 spin_unlock(&bdev_lock
);
957 void bdput(struct block_device
*bdev
)
959 iput(bdev
->bd_inode
);
962 EXPORT_SYMBOL(bdput
);
964 static struct block_device
*bd_acquire(struct inode
*inode
)
966 struct block_device
*bdev
;
968 spin_lock(&bdev_lock
);
969 bdev
= inode
->i_bdev
;
970 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
972 spin_unlock(&bdev_lock
);
975 spin_unlock(&bdev_lock
);
978 * i_bdev references block device inode that was already shut down
979 * (corresponding device got removed). Remove the reference and look
980 * up block device inode again just in case new device got
981 * reestablished under the same device number.
986 bdev
= bdget(inode
->i_rdev
);
988 spin_lock(&bdev_lock
);
989 if (!inode
->i_bdev
) {
991 * We take an additional reference to bd_inode,
992 * and it's released in clear_inode() of inode.
993 * So, we can access it via ->i_mapping always
997 inode
->i_bdev
= bdev
;
998 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
1000 spin_unlock(&bdev_lock
);
1005 /* Call when you free inode */
1007 void bd_forget(struct inode
*inode
)
1009 struct block_device
*bdev
= NULL
;
1011 spin_lock(&bdev_lock
);
1012 if (!sb_is_blkdev_sb(inode
->i_sb
))
1013 bdev
= inode
->i_bdev
;
1014 inode
->i_bdev
= NULL
;
1015 inode
->i_mapping
= &inode
->i_data
;
1016 spin_unlock(&bdev_lock
);
1023 * bd_may_claim - test whether a block device can be claimed
1024 * @bdev: block device of interest
1025 * @whole: whole block device containing @bdev, may equal @bdev
1026 * @holder: holder trying to claim @bdev
1028 * Test whether @bdev can be claimed by @holder.
1031 * spin_lock(&bdev_lock).
1034 * %true if @bdev can be claimed, %false otherwise.
1036 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1039 if (bdev
->bd_holder
== holder
)
1040 return true; /* already a holder */
1041 else if (bdev
->bd_holder
!= NULL
)
1042 return false; /* held by someone else */
1043 else if (whole
== bdev
)
1044 return true; /* is a whole device which isn't held */
1046 else if (whole
->bd_holder
== bd_may_claim
)
1047 return true; /* is a partition of a device that is being partitioned */
1048 else if (whole
->bd_holder
!= NULL
)
1049 return false; /* is a partition of a held device */
1051 return true; /* is a partition of an un-held device */
1055 * bd_prepare_to_claim - prepare to claim a block device
1056 * @bdev: block device of interest
1057 * @whole: the whole device containing @bdev, may equal @bdev
1058 * @holder: holder trying to claim @bdev
1060 * Prepare to claim @bdev. This function fails if @bdev is already
1061 * claimed by another holder and waits if another claiming is in
1062 * progress. This function doesn't actually claim. On successful
1063 * return, the caller has ownership of bd_claiming and bd_holder[s].
1066 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1067 * it multiple times.
1070 * 0 if @bdev can be claimed, -EBUSY otherwise.
1072 static int bd_prepare_to_claim(struct block_device
*bdev
,
1073 struct block_device
*whole
, void *holder
)
1076 /* if someone else claimed, fail */
1077 if (!bd_may_claim(bdev
, whole
, holder
))
1080 /* if claiming is already in progress, wait for it to finish */
1081 if (whole
->bd_claiming
) {
1082 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1085 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1086 spin_unlock(&bdev_lock
);
1088 finish_wait(wq
, &wait
);
1089 spin_lock(&bdev_lock
);
1097 static struct gendisk
*bdev_get_gendisk(struct block_device
*bdev
, int *partno
)
1099 struct gendisk
*disk
= get_gendisk(bdev
->bd_dev
, partno
);
1104 * Now that we hold gendisk reference we make sure bdev we looked up is
1105 * not stale. If it is, it means device got removed and created before
1106 * we looked up gendisk and we fail open in such case. Associating
1107 * unhashed bdev with newly created gendisk could lead to two bdevs
1108 * (and thus two independent caches) being associated with one device
1111 if (inode_unhashed(bdev
->bd_inode
)) {
1112 put_disk_and_module(disk
);
1119 * bd_start_claiming - start claiming a block device
1120 * @bdev: block device of interest
1121 * @holder: holder trying to claim @bdev
1123 * @bdev is about to be opened exclusively. Check @bdev can be opened
1124 * exclusively and mark that an exclusive open is in progress. Each
1125 * successful call to this function must be matched with a call to
1126 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1129 * This function is used to gain exclusive access to the block device
1130 * without actually causing other exclusive open attempts to fail. It
1131 * should be used when the open sequence itself requires exclusive
1132 * access but may subsequently fail.
1138 * Pointer to the block device containing @bdev on success, ERR_PTR()
1141 struct block_device
*bd_start_claiming(struct block_device
*bdev
, void *holder
)
1143 struct gendisk
*disk
;
1144 struct block_device
*whole
;
1150 * @bdev might not have been initialized properly yet, look up
1151 * and grab the outer block device the hard way.
1153 disk
= bdev_get_gendisk(bdev
, &partno
);
1155 return ERR_PTR(-ENXIO
);
1158 * Normally, @bdev should equal what's returned from bdget_disk()
1159 * if partno is 0; however, some drivers (floppy) use multiple
1160 * bdev's for the same physical device and @bdev may be one of the
1161 * aliases. Keep @bdev if partno is 0. This means claimer
1162 * tracking is broken for those devices but it has always been that
1166 whole
= bdget_disk(disk
, 0);
1168 whole
= bdgrab(bdev
);
1170 put_disk_and_module(disk
);
1172 return ERR_PTR(-ENOMEM
);
1174 /* prepare to claim, if successful, mark claiming in progress */
1175 spin_lock(&bdev_lock
);
1177 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1179 whole
->bd_claiming
= holder
;
1180 spin_unlock(&bdev_lock
);
1183 spin_unlock(&bdev_lock
);
1185 return ERR_PTR(err
);
1188 EXPORT_SYMBOL(bd_start_claiming
);
1190 static void bd_clear_claiming(struct block_device
*whole
, void *holder
)
1192 lockdep_assert_held(&bdev_lock
);
1193 /* tell others that we're done */
1194 BUG_ON(whole
->bd_claiming
!= holder
);
1195 whole
->bd_claiming
= NULL
;
1196 wake_up_bit(&whole
->bd_claiming
, 0);
1200 * bd_finish_claiming - finish claiming of a block device
1201 * @bdev: block device of interest
1202 * @whole: whole block device (returned from bd_start_claiming())
1203 * @holder: holder that has claimed @bdev
1205 * Finish exclusive open of a block device. Mark the device as exlusively
1206 * open by the holder and wake up all waiters for exclusive open to finish.
1208 void bd_finish_claiming(struct block_device
*bdev
, struct block_device
*whole
,
1211 spin_lock(&bdev_lock
);
1212 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1214 * Note that for a whole device bd_holders will be incremented twice,
1215 * and bd_holder will be set to bd_may_claim before being set to holder
1217 whole
->bd_holders
++;
1218 whole
->bd_holder
= bd_may_claim
;
1220 bdev
->bd_holder
= holder
;
1221 bd_clear_claiming(whole
, holder
);
1222 spin_unlock(&bdev_lock
);
1224 EXPORT_SYMBOL(bd_finish_claiming
);
1227 * bd_abort_claiming - abort claiming of a block device
1228 * @bdev: block device of interest
1229 * @whole: whole block device (returned from bd_start_claiming())
1230 * @holder: holder that has claimed @bdev
1232 * Abort claiming of a block device when the exclusive open failed. This can be
1233 * also used when exclusive open is not actually desired and we just needed
1234 * to block other exclusive openers for a while.
1236 void bd_abort_claiming(struct block_device
*bdev
, struct block_device
*whole
,
1239 spin_lock(&bdev_lock
);
1240 bd_clear_claiming(whole
, holder
);
1241 spin_unlock(&bdev_lock
);
1243 EXPORT_SYMBOL(bd_abort_claiming
);
1246 struct bd_holder_disk
{
1247 struct list_head list
;
1248 struct gendisk
*disk
;
1252 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1253 struct gendisk
*disk
)
1255 struct bd_holder_disk
*holder
;
1257 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1258 if (holder
->disk
== disk
)
1263 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1265 return sysfs_create_link(from
, to
, kobject_name(to
));
1268 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1270 sysfs_remove_link(from
, kobject_name(to
));
1274 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1275 * @bdev: the claimed slave bdev
1276 * @disk: the holding disk
1278 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1280 * This functions creates the following sysfs symlinks.
1282 * - from "slaves" directory of the holder @disk to the claimed @bdev
1283 * - from "holders" directory of the @bdev to the holder @disk
1285 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1286 * passed to bd_link_disk_holder(), then:
1288 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1289 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1291 * The caller must have claimed @bdev before calling this function and
1292 * ensure that both @bdev and @disk are valid during the creation and
1293 * lifetime of these symlinks.
1299 * 0 on success, -errno on failure.
1301 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1303 struct bd_holder_disk
*holder
;
1306 mutex_lock(&bdev
->bd_mutex
);
1308 WARN_ON_ONCE(!bdev
->bd_holder
);
1310 /* FIXME: remove the following once add_disk() handles errors */
1311 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1314 holder
= bd_find_holder_disk(bdev
, disk
);
1320 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1326 INIT_LIST_HEAD(&holder
->list
);
1327 holder
->disk
= disk
;
1330 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1334 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1338 * bdev could be deleted beneath us which would implicitly destroy
1339 * the holder directory. Hold on to it.
1341 kobject_get(bdev
->bd_part
->holder_dir
);
1343 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1347 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1351 mutex_unlock(&bdev
->bd_mutex
);
1354 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1357 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1358 * @bdev: the calimed slave bdev
1359 * @disk: the holding disk
1361 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1366 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1368 struct bd_holder_disk
*holder
;
1370 mutex_lock(&bdev
->bd_mutex
);
1372 holder
= bd_find_holder_disk(bdev
, disk
);
1374 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1375 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1376 del_symlink(bdev
->bd_part
->holder_dir
,
1377 &disk_to_dev(disk
)->kobj
);
1378 kobject_put(bdev
->bd_part
->holder_dir
);
1379 list_del_init(&holder
->list
);
1383 mutex_unlock(&bdev
->bd_mutex
);
1385 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1389 * flush_disk - invalidates all buffer-cache entries on a disk
1391 * @bdev: struct block device to be flushed
1392 * @kill_dirty: flag to guide handling of dirty inodes
1394 * Invalidates all buffer-cache entries on a disk. It should be called
1395 * when a disk has been changed -- either by a media change or online
1398 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1400 if (__invalidate_device(bdev
, kill_dirty
)) {
1401 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1402 "resized disk %s\n",
1403 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1405 bdev
->bd_invalidated
= 1;
1409 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1410 * @disk: struct gendisk to check
1411 * @bdev: struct bdev to adjust.
1412 * @verbose: if %true log a message about a size change if there is any
1414 * This routine checks to see if the bdev size does not match the disk size
1415 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1418 static void check_disk_size_change(struct gendisk
*disk
,
1419 struct block_device
*bdev
, bool verbose
)
1421 loff_t disk_size
, bdev_size
;
1423 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1424 bdev_size
= i_size_read(bdev
->bd_inode
);
1425 if (disk_size
!= bdev_size
) {
1428 "%s: detected capacity change from %lld to %lld\n",
1429 disk
->disk_name
, bdev_size
, disk_size
);
1431 i_size_write(bdev
->bd_inode
, disk_size
);
1432 if (bdev_size
> disk_size
)
1433 flush_disk(bdev
, false);
1435 bdev
->bd_invalidated
= 0;
1439 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1440 * @disk: struct gendisk to be revalidated
1442 * This routine is a wrapper for lower-level driver's revalidate_disk
1443 * call-backs. It is used to do common pre and post operations needed
1444 * for all revalidate_disk operations.
1446 int revalidate_disk(struct gendisk
*disk
)
1450 if (disk
->fops
->revalidate_disk
)
1451 ret
= disk
->fops
->revalidate_disk(disk
);
1454 * Hidden disks don't have associated bdev so there's no point in
1457 if (!(disk
->flags
& GENHD_FL_HIDDEN
)) {
1458 struct block_device
*bdev
= bdget_disk(disk
, 0);
1463 mutex_lock(&bdev
->bd_mutex
);
1464 check_disk_size_change(disk
, bdev
, ret
== 0);
1465 mutex_unlock(&bdev
->bd_mutex
);
1470 EXPORT_SYMBOL(revalidate_disk
);
1473 * This routine checks whether a removable media has been changed,
1474 * and invalidates all buffer-cache-entries in that case. This
1475 * is a relatively slow routine, so we have to try to minimize using
1476 * it. Thus it is called only upon a 'mount' or 'open'. This
1477 * is the best way of combining speed and utility, I think.
1478 * People changing diskettes in the middle of an operation deserve
1481 int check_disk_change(struct block_device
*bdev
)
1483 struct gendisk
*disk
= bdev
->bd_disk
;
1484 const struct block_device_operations
*bdops
= disk
->fops
;
1485 unsigned int events
;
1487 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1488 DISK_EVENT_EJECT_REQUEST
);
1489 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1492 flush_disk(bdev
, true);
1493 if (bdops
->revalidate_disk
)
1494 bdops
->revalidate_disk(bdev
->bd_disk
);
1498 EXPORT_SYMBOL(check_disk_change
);
1500 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1502 inode_lock(bdev
->bd_inode
);
1503 i_size_write(bdev
->bd_inode
, size
);
1504 inode_unlock(bdev
->bd_inode
);
1506 EXPORT_SYMBOL(bd_set_size
);
1508 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1510 int bdev_disk_changed(struct block_device
*bdev
, bool invalidate
)
1512 struct gendisk
*disk
= bdev
->bd_disk
;
1515 lockdep_assert_held(&bdev
->bd_mutex
);
1518 ret
= blk_drop_partitions(disk
, bdev
);
1523 * Historically we only set the capacity to zero for devices that
1524 * support partitions (independ of actually having partitions created).
1525 * Doing that is rather inconsistent, but changing it broke legacy
1526 * udisks polling for legacy ide-cdrom devices. Use the crude check
1527 * below to get the sane behavior for most device while not breaking
1528 * userspace for this particular setup.
1531 if (disk_part_scan_enabled(disk
) ||
1532 !(disk
->flags
& GENHD_FL_REMOVABLE
))
1533 set_capacity(disk
, 0);
1535 if (disk
->fops
->revalidate_disk
)
1536 disk
->fops
->revalidate_disk(disk
);
1539 check_disk_size_change(disk
, bdev
, !invalidate
);
1541 if (get_capacity(disk
)) {
1542 ret
= blk_add_partitions(disk
, bdev
);
1545 } else if (invalidate
) {
1547 * Tell userspace that the media / partition table may have
1550 kobject_uevent(&disk_to_dev(disk
)->kobj
, KOBJ_CHANGE
);
1556 * Only exported for for loop and dasd for historic reasons. Don't use in new
1559 EXPORT_SYMBOL_GPL(bdev_disk_changed
);
1564 * mutex_lock(part->bd_mutex)
1565 * mutex_lock_nested(whole->bd_mutex, 1)
1568 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1570 struct gendisk
*disk
;
1574 bool first_open
= false;
1576 if (mode
& FMODE_READ
)
1578 if (mode
& FMODE_WRITE
)
1581 * hooks: /n/, see "layering violations".
1584 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1594 disk
= bdev_get_gendisk(bdev
, &partno
);
1598 disk_block_events(disk
);
1599 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1600 if (!bdev
->bd_openers
) {
1602 bdev
->bd_disk
= disk
;
1603 bdev
->bd_queue
= disk
->queue
;
1604 bdev
->bd_contains
= bdev
;
1605 bdev
->bd_partno
= partno
;
1609 bdev
->bd_part
= disk_get_part(disk
, partno
);
1614 if (disk
->fops
->open
) {
1615 ret
= disk
->fops
->open(bdev
, mode
);
1616 if (ret
== -ERESTARTSYS
) {
1617 /* Lost a race with 'disk' being
1618 * deleted, try again.
1621 disk_put_part(bdev
->bd_part
);
1622 bdev
->bd_part
= NULL
;
1623 bdev
->bd_disk
= NULL
;
1624 bdev
->bd_queue
= NULL
;
1625 mutex_unlock(&bdev
->bd_mutex
);
1626 disk_unblock_events(disk
);
1627 put_disk_and_module(disk
);
1633 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1634 set_init_blocksize(bdev
);
1638 * If the device is invalidated, rescan partition
1639 * if open succeeded or failed with -ENOMEDIUM.
1640 * The latter is necessary to prevent ghost
1641 * partitions on a removed medium.
1643 if (bdev
->bd_invalidated
&&
1644 (!ret
|| ret
== -ENOMEDIUM
))
1645 bdev_disk_changed(bdev
, ret
== -ENOMEDIUM
);
1650 struct block_device
*whole
;
1651 whole
= bdget_disk(disk
, 0);
1656 ret
= __blkdev_get(whole
, mode
, 1);
1659 bdev
->bd_contains
= whole
;
1660 bdev
->bd_part
= disk_get_part(disk
, partno
);
1661 if (!(disk
->flags
& GENHD_FL_UP
) ||
1662 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1666 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1667 set_init_blocksize(bdev
);
1670 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1671 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1673 if (bdev
->bd_contains
== bdev
) {
1675 if (bdev
->bd_disk
->fops
->open
)
1676 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1677 /* the same as first opener case, read comment there */
1678 if (bdev
->bd_invalidated
&&
1679 (!ret
|| ret
== -ENOMEDIUM
))
1680 bdev_disk_changed(bdev
, ret
== -ENOMEDIUM
);
1682 goto out_unlock_bdev
;
1687 bdev
->bd_part_count
++;
1688 mutex_unlock(&bdev
->bd_mutex
);
1689 disk_unblock_events(disk
);
1690 /* only one opener holds refs to the module and disk */
1692 put_disk_and_module(disk
);
1696 disk_put_part(bdev
->bd_part
);
1697 bdev
->bd_disk
= NULL
;
1698 bdev
->bd_part
= NULL
;
1699 bdev
->bd_queue
= NULL
;
1700 if (bdev
!= bdev
->bd_contains
)
1701 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1702 bdev
->bd_contains
= NULL
;
1704 mutex_unlock(&bdev
->bd_mutex
);
1705 disk_unblock_events(disk
);
1706 put_disk_and_module(disk
);
1714 * blkdev_get - open a block device
1715 * @bdev: block_device to open
1716 * @mode: FMODE_* mask
1717 * @holder: exclusive holder identifier
1719 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1720 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1721 * @holder is invalid. Exclusive opens may nest for the same @holder.
1723 * On success, the reference count of @bdev is unchanged. On failure,
1730 * 0 on success, -errno on failure.
1732 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1734 struct block_device
*whole
= NULL
;
1737 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1739 if ((mode
& FMODE_EXCL
) && holder
) {
1740 whole
= bd_start_claiming(bdev
, holder
);
1741 if (IS_ERR(whole
)) {
1743 return PTR_ERR(whole
);
1747 res
= __blkdev_get(bdev
, mode
, 0);
1750 struct gendisk
*disk
= whole
->bd_disk
;
1752 /* finish claiming */
1753 mutex_lock(&bdev
->bd_mutex
);
1755 bd_finish_claiming(bdev
, whole
, holder
);
1757 bd_abort_claiming(bdev
, whole
, holder
);
1759 * Block event polling for write claims if requested. Any
1760 * write holder makes the write_holder state stick until
1761 * all are released. This is good enough and tracking
1762 * individual writeable reference is too fragile given the
1763 * way @mode is used in blkdev_get/put().
1765 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1766 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1767 bdev
->bd_write_holder
= true;
1768 disk_block_events(disk
);
1771 mutex_unlock(&bdev
->bd_mutex
);
1777 EXPORT_SYMBOL(blkdev_get
);
1780 * blkdev_get_by_path - open a block device by name
1781 * @path: path to the block device to open
1782 * @mode: FMODE_* mask
1783 * @holder: exclusive holder identifier
1785 * Open the blockdevice described by the device file at @path. @mode
1786 * and @holder are identical to blkdev_get().
1788 * On success, the returned block_device has reference count of one.
1794 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1796 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1799 struct block_device
*bdev
;
1802 bdev
= lookup_bdev(path
);
1806 err
= blkdev_get(bdev
, mode
, holder
);
1808 return ERR_PTR(err
);
1810 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1811 blkdev_put(bdev
, mode
);
1812 return ERR_PTR(-EACCES
);
1817 EXPORT_SYMBOL(blkdev_get_by_path
);
1820 * blkdev_get_by_dev - open a block device by device number
1821 * @dev: device number of block device to open
1822 * @mode: FMODE_* mask
1823 * @holder: exclusive holder identifier
1825 * Open the blockdevice described by device number @dev. @mode and
1826 * @holder are identical to blkdev_get().
1828 * Use it ONLY if you really do not have anything better - i.e. when
1829 * you are behind a truly sucky interface and all you are given is a
1830 * device number. _Never_ to be used for internal purposes. If you
1831 * ever need it - reconsider your API.
1833 * On success, the returned block_device has reference count of one.
1839 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1841 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1843 struct block_device
*bdev
;
1848 return ERR_PTR(-ENOMEM
);
1850 err
= blkdev_get(bdev
, mode
, holder
);
1852 return ERR_PTR(err
);
1856 EXPORT_SYMBOL(blkdev_get_by_dev
);
1858 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1860 struct block_device
*bdev
;
1863 * Preserve backwards compatibility and allow large file access
1864 * even if userspace doesn't ask for it explicitly. Some mkfs
1865 * binary needs it. We might want to drop this workaround
1866 * during an unstable branch.
1868 filp
->f_flags
|= O_LARGEFILE
;
1870 filp
->f_mode
|= FMODE_NOWAIT
;
1872 if (filp
->f_flags
& O_NDELAY
)
1873 filp
->f_mode
|= FMODE_NDELAY
;
1874 if (filp
->f_flags
& O_EXCL
)
1875 filp
->f_mode
|= FMODE_EXCL
;
1876 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1877 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1879 bdev
= bd_acquire(inode
);
1883 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1884 filp
->f_wb_err
= filemap_sample_wb_err(filp
->f_mapping
);
1886 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1889 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1891 struct gendisk
*disk
= bdev
->bd_disk
;
1892 struct block_device
*victim
= NULL
;
1895 * Sync early if it looks like we're the last one. If someone else
1896 * opens the block device between now and the decrement of bd_openers
1897 * then we did a sync that we didn't need to, but that's not the end
1898 * of the world and we want to avoid long (could be several minute)
1899 * syncs while holding the mutex.
1901 if (bdev
->bd_openers
== 1)
1902 sync_blockdev(bdev
);
1904 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1906 bdev
->bd_part_count
--;
1908 if (!--bdev
->bd_openers
) {
1909 WARN_ON_ONCE(bdev
->bd_holders
);
1910 sync_blockdev(bdev
);
1913 bdev_write_inode(bdev
);
1915 if (bdev
->bd_contains
== bdev
) {
1916 if (disk
->fops
->release
)
1917 disk
->fops
->release(disk
, mode
);
1919 if (!bdev
->bd_openers
) {
1920 disk_put_part(bdev
->bd_part
);
1921 bdev
->bd_part
= NULL
;
1922 bdev
->bd_disk
= NULL
;
1923 if (bdev
!= bdev
->bd_contains
)
1924 victim
= bdev
->bd_contains
;
1925 bdev
->bd_contains
= NULL
;
1927 put_disk_and_module(disk
);
1929 mutex_unlock(&bdev
->bd_mutex
);
1932 __blkdev_put(victim
, mode
, 1);
1935 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1937 mutex_lock(&bdev
->bd_mutex
);
1939 if (mode
& FMODE_EXCL
) {
1943 * Release a claim on the device. The holder fields
1944 * are protected with bdev_lock. bd_mutex is to
1945 * synchronize disk_holder unlinking.
1947 spin_lock(&bdev_lock
);
1949 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1950 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1952 /* bd_contains might point to self, check in a separate step */
1953 if ((bdev_free
= !bdev
->bd_holders
))
1954 bdev
->bd_holder
= NULL
;
1955 if (!bdev
->bd_contains
->bd_holders
)
1956 bdev
->bd_contains
->bd_holder
= NULL
;
1958 spin_unlock(&bdev_lock
);
1961 * If this was the last claim, remove holder link and
1962 * unblock evpoll if it was a write holder.
1964 if (bdev_free
&& bdev
->bd_write_holder
) {
1965 disk_unblock_events(bdev
->bd_disk
);
1966 bdev
->bd_write_holder
= false;
1971 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1972 * event. This is to ensure detection of media removal commanded
1973 * from userland - e.g. eject(1).
1975 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1977 mutex_unlock(&bdev
->bd_mutex
);
1979 __blkdev_put(bdev
, mode
, 0);
1981 EXPORT_SYMBOL(blkdev_put
);
1983 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1985 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1986 blkdev_put(bdev
, filp
->f_mode
);
1990 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1992 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1993 fmode_t mode
= file
->f_mode
;
1996 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1997 * to updated it before every ioctl.
1999 if (file
->f_flags
& O_NDELAY
)
2000 mode
|= FMODE_NDELAY
;
2002 mode
&= ~FMODE_NDELAY
;
2004 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
2008 * Write data to the block device. Only intended for the block device itself
2009 * and the raw driver which basically is a fake block device.
2011 * Does not take i_mutex for the write and thus is not for general purpose
2014 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
2016 struct file
*file
= iocb
->ki_filp
;
2017 struct inode
*bd_inode
= bdev_file_inode(file
);
2018 loff_t size
= i_size_read(bd_inode
);
2019 struct blk_plug plug
;
2022 if (bdev_read_only(I_BDEV(bd_inode
)))
2025 /* uswsusp needs write permission to the swap */
2026 if (IS_SWAPFILE(bd_inode
) && !hibernation_available())
2029 if (!iov_iter_count(from
))
2032 if (iocb
->ki_pos
>= size
)
2035 if ((iocb
->ki_flags
& (IOCB_NOWAIT
| IOCB_DIRECT
)) == IOCB_NOWAIT
)
2038 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
2040 blk_start_plug(&plug
);
2041 ret
= __generic_file_write_iter(iocb
, from
);
2043 ret
= generic_write_sync(iocb
, ret
);
2044 blk_finish_plug(&plug
);
2047 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
2049 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2051 struct file
*file
= iocb
->ki_filp
;
2052 struct inode
*bd_inode
= bdev_file_inode(file
);
2053 loff_t size
= i_size_read(bd_inode
);
2054 loff_t pos
= iocb
->ki_pos
;
2060 iov_iter_truncate(to
, size
);
2061 return generic_file_read_iter(iocb
, to
);
2063 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
2066 * Try to release a page associated with block device when the system
2067 * is under memory pressure.
2069 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
2071 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
2073 if (super
&& super
->s_op
->bdev_try_to_free_page
)
2074 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
2076 return try_to_free_buffers(page
);
2079 static int blkdev_writepages(struct address_space
*mapping
,
2080 struct writeback_control
*wbc
)
2082 return generic_writepages(mapping
, wbc
);
2085 static const struct address_space_operations def_blk_aops
= {
2086 .readpage
= blkdev_readpage
,
2087 .readahead
= blkdev_readahead
,
2088 .writepage
= blkdev_writepage
,
2089 .write_begin
= blkdev_write_begin
,
2090 .write_end
= blkdev_write_end
,
2091 .writepages
= blkdev_writepages
,
2092 .releasepage
= blkdev_releasepage
,
2093 .direct_IO
= blkdev_direct_IO
,
2094 .migratepage
= buffer_migrate_page_norefs
,
2095 .is_dirty_writeback
= buffer_check_dirty_writeback
,
2098 #define BLKDEV_FALLOC_FL_SUPPORTED \
2099 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2100 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2102 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
2105 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
2106 struct address_space
*mapping
;
2107 loff_t end
= start
+ len
- 1;
2111 /* Fail if we don't recognize the flags. */
2112 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
2115 /* Don't go off the end of the device. */
2116 isize
= i_size_read(bdev
->bd_inode
);
2120 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2121 len
= isize
- start
;
2122 end
= start
+ len
- 1;
2128 * Don't allow IO that isn't aligned to logical block size.
2130 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2133 /* Invalidate the page cache, including dirty pages. */
2134 mapping
= bdev
->bd_inode
->i_mapping
;
2135 truncate_inode_pages_range(mapping
, start
, end
);
2138 case FALLOC_FL_ZERO_RANGE
:
2139 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2140 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2141 GFP_KERNEL
, BLKDEV_ZERO_NOUNMAP
);
2143 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2144 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2145 GFP_KERNEL
, BLKDEV_ZERO_NOFALLBACK
);
2147 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2148 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2158 * Invalidate again; if someone wandered in and dirtied a page,
2159 * the caller will be given -EBUSY. The third argument is
2160 * inclusive, so the rounding here is safe.
2162 return invalidate_inode_pages2_range(mapping
,
2163 start
>> PAGE_SHIFT
,
2167 const struct file_operations def_blk_fops
= {
2168 .open
= blkdev_open
,
2169 .release
= blkdev_close
,
2170 .llseek
= block_llseek
,
2171 .read_iter
= blkdev_read_iter
,
2172 .write_iter
= blkdev_write_iter
,
2173 .iopoll
= blkdev_iopoll
,
2174 .mmap
= generic_file_mmap
,
2175 .fsync
= blkdev_fsync
,
2176 .unlocked_ioctl
= block_ioctl
,
2177 #ifdef CONFIG_COMPAT
2178 .compat_ioctl
= compat_blkdev_ioctl
,
2180 .splice_read
= generic_file_splice_read
,
2181 .splice_write
= iter_file_splice_write
,
2182 .fallocate
= blkdev_fallocate
,
2185 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2188 mm_segment_t old_fs
= get_fs();
2190 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2195 EXPORT_SYMBOL(ioctl_by_bdev
);
2198 * lookup_bdev - lookup a struct block_device by name
2199 * @pathname: special file representing the block device
2201 * Get a reference to the blockdevice at @pathname in the current
2202 * namespace if possible and return it. Return ERR_PTR(error)
2205 struct block_device
*lookup_bdev(const char *pathname
)
2207 struct block_device
*bdev
;
2208 struct inode
*inode
;
2212 if (!pathname
|| !*pathname
)
2213 return ERR_PTR(-EINVAL
);
2215 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2217 return ERR_PTR(error
);
2219 inode
= d_backing_inode(path
.dentry
);
2221 if (!S_ISBLK(inode
->i_mode
))
2224 if (!may_open_dev(&path
))
2227 bdev
= bd_acquire(inode
);
2234 bdev
= ERR_PTR(error
);
2237 EXPORT_SYMBOL(lookup_bdev
);
2239 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2241 struct super_block
*sb
= get_super(bdev
);
2246 * no need to lock the super, get_super holds the
2247 * read mutex so the filesystem cannot go away
2248 * under us (->put_super runs with the write lock
2251 shrink_dcache_sb(sb
);
2252 res
= invalidate_inodes(sb
, kill_dirty
);
2255 invalidate_bdev(bdev
);
2258 EXPORT_SYMBOL(__invalidate_device
);
2260 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2262 struct inode
*inode
, *old_inode
= NULL
;
2264 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2265 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2266 struct address_space
*mapping
= inode
->i_mapping
;
2267 struct block_device
*bdev
;
2269 spin_lock(&inode
->i_lock
);
2270 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2271 mapping
->nrpages
== 0) {
2272 spin_unlock(&inode
->i_lock
);
2276 spin_unlock(&inode
->i_lock
);
2277 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2279 * We hold a reference to 'inode' so it couldn't have been
2280 * removed from s_inodes list while we dropped the
2281 * s_inode_list_lock We cannot iput the inode now as we can
2282 * be holding the last reference and we cannot iput it under
2283 * s_inode_list_lock. So we keep the reference and iput it
2288 bdev
= I_BDEV(inode
);
2290 mutex_lock(&bdev
->bd_mutex
);
2291 if (bdev
->bd_openers
)
2293 mutex_unlock(&bdev
->bd_mutex
);
2295 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2297 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);