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/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <asm/uaccess.h>
36 struct block_device bdev
;
37 struct inode vfs_inode
;
40 static const struct address_space_operations def_blk_aops
;
42 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
44 return container_of(inode
, struct bdev_inode
, vfs_inode
);
47 struct block_device
*I_BDEV(struct inode
*inode
)
49 return &BDEV_I(inode
)->bdev
;
51 EXPORT_SYMBOL(I_BDEV
);
53 static void bdev_write_inode(struct block_device
*bdev
)
55 struct inode
*inode
= bdev
->bd_inode
;
58 spin_lock(&inode
->i_lock
);
59 while (inode
->i_state
& I_DIRTY
) {
60 spin_unlock(&inode
->i_lock
);
61 ret
= write_inode_now(inode
, true);
63 char name
[BDEVNAME_SIZE
];
64 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
65 "for block device %s (err=%d).\n",
66 bdevname(bdev
, name
), ret
);
68 spin_lock(&inode
->i_lock
);
70 spin_unlock(&inode
->i_lock
);
73 /* Kill _all_ buffers and pagecache , dirty or not.. */
74 void kill_bdev(struct block_device
*bdev
)
76 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
78 if (mapping
->nrpages
== 0 && mapping
->nrshadows
== 0)
82 truncate_inode_pages(mapping
, 0);
84 EXPORT_SYMBOL(kill_bdev
);
86 /* Invalidate clean unused buffers and pagecache. */
87 void invalidate_bdev(struct block_device
*bdev
)
89 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
91 if (mapping
->nrpages
== 0)
95 lru_add_drain_all(); /* make sure all lru add caches are flushed */
96 invalidate_mapping_pages(mapping
, 0, -1);
97 /* 99% of the time, we don't need to flush the cleancache on the bdev.
98 * But, for the strange corners, lets be cautious
100 cleancache_invalidate_inode(mapping
);
102 EXPORT_SYMBOL(invalidate_bdev
);
104 int set_blocksize(struct block_device
*bdev
, int size
)
106 /* Size must be a power of two, and between 512 and PAGE_SIZE */
107 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
110 /* Size cannot be smaller than the size supported by the device */
111 if (size
< bdev_logical_block_size(bdev
))
114 /* Don't change the size if it is same as current */
115 if (bdev
->bd_block_size
!= size
) {
117 bdev
->bd_block_size
= size
;
118 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
124 EXPORT_SYMBOL(set_blocksize
);
126 int sb_set_blocksize(struct super_block
*sb
, int size
)
128 if (set_blocksize(sb
->s_bdev
, size
))
130 /* If we get here, we know size is power of two
131 * and it's value is between 512 and PAGE_SIZE */
132 sb
->s_blocksize
= size
;
133 sb
->s_blocksize_bits
= blksize_bits(size
);
134 return sb
->s_blocksize
;
137 EXPORT_SYMBOL(sb_set_blocksize
);
139 int sb_min_blocksize(struct super_block
*sb
, int size
)
141 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
144 return sb_set_blocksize(sb
, size
);
147 EXPORT_SYMBOL(sb_min_blocksize
);
150 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
151 struct buffer_head
*bh
, int create
)
153 bh
->b_bdev
= I_BDEV(inode
);
154 bh
->b_blocknr
= iblock
;
155 set_buffer_mapped(bh
);
160 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
162 struct file
*file
= iocb
->ki_filp
;
163 struct inode
*inode
= file
->f_mapping
->host
;
166 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
167 NULL
, DIO_SKIP_DIO_COUNT
);
168 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
169 blkdev_get_block
, NULL
, NULL
,
173 int __sync_blockdev(struct block_device
*bdev
, int wait
)
178 return filemap_flush(bdev
->bd_inode
->i_mapping
);
179 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
183 * Write out and wait upon all the dirty data associated with a block
184 * device via its mapping. Does not take the superblock lock.
186 int sync_blockdev(struct block_device
*bdev
)
188 return __sync_blockdev(bdev
, 1);
190 EXPORT_SYMBOL(sync_blockdev
);
193 * Write out and wait upon all dirty data associated with this
194 * device. Filesystem data as well as the underlying block
195 * device. Takes the superblock lock.
197 int fsync_bdev(struct block_device
*bdev
)
199 struct super_block
*sb
= get_super(bdev
);
201 int res
= sync_filesystem(sb
);
205 return sync_blockdev(bdev
);
207 EXPORT_SYMBOL(fsync_bdev
);
210 * freeze_bdev -- lock a filesystem and force it into a consistent state
211 * @bdev: blockdevice to lock
213 * If a superblock is found on this device, we take the s_umount semaphore
214 * on it to make sure nobody unmounts until the snapshot creation is done.
215 * The reference counter (bd_fsfreeze_count) guarantees that only the last
216 * unfreeze process can unfreeze the frozen filesystem actually when multiple
217 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
218 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
221 struct super_block
*freeze_bdev(struct block_device
*bdev
)
223 struct super_block
*sb
;
226 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
227 if (++bdev
->bd_fsfreeze_count
> 1) {
229 * We don't even need to grab a reference - the first call
230 * to freeze_bdev grab an active reference and only the last
231 * thaw_bdev drops it.
233 sb
= get_super(bdev
);
235 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
239 sb
= get_active_super(bdev
);
242 if (sb
->s_op
->freeze_super
)
243 error
= sb
->s_op
->freeze_super(sb
);
245 error
= freeze_super(sb
);
247 deactivate_super(sb
);
248 bdev
->bd_fsfreeze_count
--;
249 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
250 return ERR_PTR(error
);
252 deactivate_super(sb
);
255 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
256 return sb
; /* thaw_bdev releases s->s_umount */
258 EXPORT_SYMBOL(freeze_bdev
);
261 * thaw_bdev -- unlock filesystem
262 * @bdev: blockdevice to unlock
263 * @sb: associated superblock
265 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
267 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
271 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
272 if (!bdev
->bd_fsfreeze_count
)
276 if (--bdev
->bd_fsfreeze_count
> 0)
282 if (sb
->s_op
->thaw_super
)
283 error
= sb
->s_op
->thaw_super(sb
);
285 error
= thaw_super(sb
);
287 bdev
->bd_fsfreeze_count
++;
288 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
292 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
295 EXPORT_SYMBOL(thaw_bdev
);
297 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
299 return block_write_full_page(page
, blkdev_get_block
, wbc
);
302 static int blkdev_readpage(struct file
* file
, struct page
* page
)
304 return block_read_full_page(page
, blkdev_get_block
);
307 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
308 struct list_head
*pages
, unsigned nr_pages
)
310 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
313 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
314 loff_t pos
, unsigned len
, unsigned flags
,
315 struct page
**pagep
, void **fsdata
)
317 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
321 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
322 loff_t pos
, unsigned len
, unsigned copied
,
323 struct page
*page
, void *fsdata
)
326 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
329 page_cache_release(page
);
336 * for a block special file file_inode(file)->i_size is zero
337 * so we compute the size by hand (just as in block_read/write above)
339 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
341 struct inode
*bd_inode
= file
->f_mapping
->host
;
344 mutex_lock(&bd_inode
->i_mutex
);
345 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
346 mutex_unlock(&bd_inode
->i_mutex
);
350 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
352 struct inode
*bd_inode
= filp
->f_mapping
->host
;
353 struct block_device
*bdev
= I_BDEV(bd_inode
);
356 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
361 * There is no need to serialise calls to blkdev_issue_flush with
362 * i_mutex and doing so causes performance issues with concurrent
363 * O_SYNC writers to a block device.
365 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
366 if (error
== -EOPNOTSUPP
)
371 EXPORT_SYMBOL(blkdev_fsync
);
374 * bdev_read_page() - Start reading a page from a block device
375 * @bdev: The device to read the page from
376 * @sector: The offset on the device to read the page to (need not be aligned)
377 * @page: The page to read
379 * On entry, the page should be locked. It will be unlocked when the page
380 * has been read. If the block driver implements rw_page synchronously,
381 * that will be true on exit from this function, but it need not be.
383 * Errors returned by this function are usually "soft", eg out of memory, or
384 * queue full; callers should try a different route to read this page rather
385 * than propagate an error back up the stack.
387 * Return: negative errno if an error occurs, 0 if submission was successful.
389 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
392 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
393 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
395 return ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
397 EXPORT_SYMBOL_GPL(bdev_read_page
);
400 * bdev_write_page() - Start writing a page to a block device
401 * @bdev: The device to write the page to
402 * @sector: The offset on the device to write the page to (need not be aligned)
403 * @page: The page to write
404 * @wbc: The writeback_control for the write
406 * On entry, the page should be locked and not currently under writeback.
407 * On exit, if the write started successfully, the page will be unlocked and
408 * under writeback. If the write failed already (eg the driver failed to
409 * queue the page to the device), the page will still be locked. If the
410 * caller is a ->writepage implementation, it will need to unlock the page.
412 * Errors returned by this function are usually "soft", eg out of memory, or
413 * queue full; callers should try a different route to write this page rather
414 * than propagate an error back up the stack.
416 * Return: negative errno if an error occurs, 0 if submission was successful.
418 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
419 struct page
*page
, struct writeback_control
*wbc
)
422 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
423 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
424 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
426 set_page_writeback(page
);
427 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
429 end_page_writeback(page
);
434 EXPORT_SYMBOL_GPL(bdev_write_page
);
437 * bdev_direct_access() - Get the address for directly-accessibly memory
438 * @bdev: The device containing the memory
439 * @sector: The offset within the device
440 * @addr: Where to put the address of the memory
441 * @pfn: The Page Frame Number for the memory
442 * @size: The number of bytes requested
444 * If a block device is made up of directly addressable memory, this function
445 * will tell the caller the PFN and the address of the memory. The address
446 * may be directly dereferenced within the kernel without the need to call
447 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
450 * Return: negative errno if an error occurs, otherwise the number of bytes
451 * accessible at this address.
453 long bdev_direct_access(struct block_device
*bdev
, sector_t sector
,
454 void __pmem
**addr
, unsigned long *pfn
, long size
)
457 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
460 * The device driver is allowed to sleep, in order to make the
461 * memory directly accessible.
467 if (!ops
->direct_access
)
469 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
470 part_nr_sects_read(bdev
->bd_part
))
472 sector
+= get_start_sect(bdev
);
473 if (sector
% (PAGE_SIZE
/ 512))
475 avail
= ops
->direct_access(bdev
, sector
, addr
, pfn
);
478 return min(avail
, size
);
480 EXPORT_SYMBOL_GPL(bdev_direct_access
);
486 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
487 static struct kmem_cache
* bdev_cachep __read_mostly
;
489 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
491 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
494 return &ei
->vfs_inode
;
497 static void bdev_i_callback(struct rcu_head
*head
)
499 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
500 struct bdev_inode
*bdi
= BDEV_I(inode
);
502 kmem_cache_free(bdev_cachep
, bdi
);
505 static void bdev_destroy_inode(struct inode
*inode
)
507 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
510 static void init_once(void *foo
)
512 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
513 struct block_device
*bdev
= &ei
->bdev
;
515 memset(bdev
, 0, sizeof(*bdev
));
516 mutex_init(&bdev
->bd_mutex
);
517 INIT_LIST_HEAD(&bdev
->bd_inodes
);
518 INIT_LIST_HEAD(&bdev
->bd_list
);
520 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
522 inode_init_once(&ei
->vfs_inode
);
523 /* Initialize mutex for freeze. */
524 mutex_init(&bdev
->bd_fsfreeze_mutex
);
527 static inline void __bd_forget(struct inode
*inode
)
529 list_del_init(&inode
->i_devices
);
530 inode
->i_bdev
= NULL
;
531 inode
->i_mapping
= &inode
->i_data
;
534 static void bdev_evict_inode(struct inode
*inode
)
536 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
538 truncate_inode_pages_final(&inode
->i_data
);
539 invalidate_inode_buffers(inode
); /* is it needed here? */
541 spin_lock(&bdev_lock
);
542 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
543 __bd_forget(list_entry(p
, struct inode
, i_devices
));
545 list_del_init(&bdev
->bd_list
);
546 spin_unlock(&bdev_lock
);
549 static const struct super_operations bdev_sops
= {
550 .statfs
= simple_statfs
,
551 .alloc_inode
= bdev_alloc_inode
,
552 .destroy_inode
= bdev_destroy_inode
,
553 .drop_inode
= generic_delete_inode
,
554 .evict_inode
= bdev_evict_inode
,
557 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
558 int flags
, const char *dev_name
, void *data
)
560 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
563 static struct file_system_type bd_type
= {
566 .kill_sb
= kill_anon_super
,
569 struct super_block
*blockdev_superblock __read_mostly
;
570 EXPORT_SYMBOL_GPL(blockdev_superblock
);
572 void __init
bdev_cache_init(void)
575 static struct vfsmount
*bd_mnt
;
577 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
578 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
579 SLAB_MEM_SPREAD
|SLAB_PANIC
),
581 err
= register_filesystem(&bd_type
);
583 panic("Cannot register bdev pseudo-fs");
584 bd_mnt
= kern_mount(&bd_type
);
586 panic("Cannot create bdev pseudo-fs");
587 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
591 * Most likely _very_ bad one - but then it's hardly critical for small
592 * /dev and can be fixed when somebody will need really large one.
593 * Keep in mind that it will be fed through icache hash function too.
595 static inline unsigned long hash(dev_t dev
)
597 return MAJOR(dev
)+MINOR(dev
);
600 static int bdev_test(struct inode
*inode
, void *data
)
602 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
605 static int bdev_set(struct inode
*inode
, void *data
)
607 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
611 static LIST_HEAD(all_bdevs
);
613 struct block_device
*bdget(dev_t dev
)
615 struct block_device
*bdev
;
618 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
619 bdev_test
, bdev_set
, &dev
);
624 bdev
= &BDEV_I(inode
)->bdev
;
626 if (inode
->i_state
& I_NEW
) {
627 bdev
->bd_contains
= NULL
;
628 bdev
->bd_super
= NULL
;
629 bdev
->bd_inode
= inode
;
630 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
631 bdev
->bd_part_count
= 0;
632 bdev
->bd_invalidated
= 0;
633 inode
->i_mode
= S_IFBLK
;
635 inode
->i_bdev
= bdev
;
636 inode
->i_data
.a_ops
= &def_blk_aops
;
637 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
638 spin_lock(&bdev_lock
);
639 list_add(&bdev
->bd_list
, &all_bdevs
);
640 spin_unlock(&bdev_lock
);
641 unlock_new_inode(inode
);
646 EXPORT_SYMBOL(bdget
);
649 * bdgrab -- Grab a reference to an already referenced block device
650 * @bdev: Block device to grab a reference to.
652 struct block_device
*bdgrab(struct block_device
*bdev
)
654 ihold(bdev
->bd_inode
);
657 EXPORT_SYMBOL(bdgrab
);
659 long nr_blockdev_pages(void)
661 struct block_device
*bdev
;
663 spin_lock(&bdev_lock
);
664 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
665 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
667 spin_unlock(&bdev_lock
);
671 void bdput(struct block_device
*bdev
)
673 iput(bdev
->bd_inode
);
676 EXPORT_SYMBOL(bdput
);
678 static struct block_device
*bd_acquire(struct inode
*inode
)
680 struct block_device
*bdev
;
682 spin_lock(&bdev_lock
);
683 bdev
= inode
->i_bdev
;
685 ihold(bdev
->bd_inode
);
686 spin_unlock(&bdev_lock
);
689 spin_unlock(&bdev_lock
);
691 bdev
= bdget(inode
->i_rdev
);
693 spin_lock(&bdev_lock
);
694 if (!inode
->i_bdev
) {
696 * We take an additional reference to bd_inode,
697 * and it's released in clear_inode() of inode.
698 * So, we can access it via ->i_mapping always
701 ihold(bdev
->bd_inode
);
702 inode
->i_bdev
= bdev
;
703 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
704 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
706 spin_unlock(&bdev_lock
);
711 /* Call when you free inode */
713 void bd_forget(struct inode
*inode
)
715 struct block_device
*bdev
= NULL
;
717 spin_lock(&bdev_lock
);
718 if (!sb_is_blkdev_sb(inode
->i_sb
))
719 bdev
= inode
->i_bdev
;
721 spin_unlock(&bdev_lock
);
724 iput(bdev
->bd_inode
);
728 * bd_may_claim - test whether a block device can be claimed
729 * @bdev: block device of interest
730 * @whole: whole block device containing @bdev, may equal @bdev
731 * @holder: holder trying to claim @bdev
733 * Test whether @bdev can be claimed by @holder.
736 * spin_lock(&bdev_lock).
739 * %true if @bdev can be claimed, %false otherwise.
741 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
744 if (bdev
->bd_holder
== holder
)
745 return true; /* already a holder */
746 else if (bdev
->bd_holder
!= NULL
)
747 return false; /* held by someone else */
748 else if (bdev
->bd_contains
== bdev
)
749 return true; /* is a whole device which isn't held */
751 else if (whole
->bd_holder
== bd_may_claim
)
752 return true; /* is a partition of a device that is being partitioned */
753 else if (whole
->bd_holder
!= NULL
)
754 return false; /* is a partition of a held device */
756 return true; /* is a partition of an un-held device */
760 * bd_prepare_to_claim - prepare to claim a block device
761 * @bdev: block device of interest
762 * @whole: the whole device containing @bdev, may equal @bdev
763 * @holder: holder trying to claim @bdev
765 * Prepare to claim @bdev. This function fails if @bdev is already
766 * claimed by another holder and waits if another claiming is in
767 * progress. This function doesn't actually claim. On successful
768 * return, the caller has ownership of bd_claiming and bd_holder[s].
771 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
775 * 0 if @bdev can be claimed, -EBUSY otherwise.
777 static int bd_prepare_to_claim(struct block_device
*bdev
,
778 struct block_device
*whole
, void *holder
)
781 /* if someone else claimed, fail */
782 if (!bd_may_claim(bdev
, whole
, holder
))
785 /* if claiming is already in progress, wait for it to finish */
786 if (whole
->bd_claiming
) {
787 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
790 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
791 spin_unlock(&bdev_lock
);
793 finish_wait(wq
, &wait
);
794 spin_lock(&bdev_lock
);
803 * bd_start_claiming - start claiming a block device
804 * @bdev: block device of interest
805 * @holder: holder trying to claim @bdev
807 * @bdev is about to be opened exclusively. Check @bdev can be opened
808 * exclusively and mark that an exclusive open is in progress. Each
809 * successful call to this function must be matched with a call to
810 * either bd_finish_claiming() or bd_abort_claiming() (which do not
813 * This function is used to gain exclusive access to the block device
814 * without actually causing other exclusive open attempts to fail. It
815 * should be used when the open sequence itself requires exclusive
816 * access but may subsequently fail.
822 * Pointer to the block device containing @bdev on success, ERR_PTR()
825 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
828 struct gendisk
*disk
;
829 struct block_device
*whole
;
835 * @bdev might not have been initialized properly yet, look up
836 * and grab the outer block device the hard way.
838 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
840 return ERR_PTR(-ENXIO
);
843 * Normally, @bdev should equal what's returned from bdget_disk()
844 * if partno is 0; however, some drivers (floppy) use multiple
845 * bdev's for the same physical device and @bdev may be one of the
846 * aliases. Keep @bdev if partno is 0. This means claimer
847 * tracking is broken for those devices but it has always been that
851 whole
= bdget_disk(disk
, 0);
853 whole
= bdgrab(bdev
);
855 module_put(disk
->fops
->owner
);
858 return ERR_PTR(-ENOMEM
);
860 /* prepare to claim, if successful, mark claiming in progress */
861 spin_lock(&bdev_lock
);
863 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
865 whole
->bd_claiming
= holder
;
866 spin_unlock(&bdev_lock
);
869 spin_unlock(&bdev_lock
);
876 struct bd_holder_disk
{
877 struct list_head list
;
878 struct gendisk
*disk
;
882 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
883 struct gendisk
*disk
)
885 struct bd_holder_disk
*holder
;
887 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
888 if (holder
->disk
== disk
)
893 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
895 return sysfs_create_link(from
, to
, kobject_name(to
));
898 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
900 sysfs_remove_link(from
, kobject_name(to
));
904 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
905 * @bdev: the claimed slave bdev
906 * @disk: the holding disk
908 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
910 * This functions creates the following sysfs symlinks.
912 * - from "slaves" directory of the holder @disk to the claimed @bdev
913 * - from "holders" directory of the @bdev to the holder @disk
915 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
916 * passed to bd_link_disk_holder(), then:
918 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
919 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
921 * The caller must have claimed @bdev before calling this function and
922 * ensure that both @bdev and @disk are valid during the creation and
923 * lifetime of these symlinks.
929 * 0 on success, -errno on failure.
931 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
933 struct bd_holder_disk
*holder
;
936 mutex_lock(&bdev
->bd_mutex
);
938 WARN_ON_ONCE(!bdev
->bd_holder
);
940 /* FIXME: remove the following once add_disk() handles errors */
941 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
944 holder
= bd_find_holder_disk(bdev
, disk
);
950 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
956 INIT_LIST_HEAD(&holder
->list
);
960 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
964 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
968 * bdev could be deleted beneath us which would implicitly destroy
969 * the holder directory. Hold on to it.
971 kobject_get(bdev
->bd_part
->holder_dir
);
973 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
977 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
981 mutex_unlock(&bdev
->bd_mutex
);
984 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
987 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
988 * @bdev: the calimed slave bdev
989 * @disk: the holding disk
991 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
996 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
998 struct bd_holder_disk
*holder
;
1000 mutex_lock(&bdev
->bd_mutex
);
1002 holder
= bd_find_holder_disk(bdev
, disk
);
1004 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1005 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1006 del_symlink(bdev
->bd_part
->holder_dir
,
1007 &disk_to_dev(disk
)->kobj
);
1008 kobject_put(bdev
->bd_part
->holder_dir
);
1009 list_del_init(&holder
->list
);
1013 mutex_unlock(&bdev
->bd_mutex
);
1015 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1019 * flush_disk - invalidates all buffer-cache entries on a disk
1021 * @bdev: struct block device to be flushed
1022 * @kill_dirty: flag to guide handling of dirty inodes
1024 * Invalidates all buffer-cache entries on a disk. It should be called
1025 * when a disk has been changed -- either by a media change or online
1028 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1030 if (__invalidate_device(bdev
, kill_dirty
)) {
1031 char name
[BDEVNAME_SIZE
] = "";
1034 disk_name(bdev
->bd_disk
, 0, name
);
1035 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1036 "resized disk %s\n", name
);
1041 if (disk_part_scan_enabled(bdev
->bd_disk
))
1042 bdev
->bd_invalidated
= 1;
1046 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1047 * @disk: struct gendisk to check
1048 * @bdev: struct bdev to adjust.
1050 * This routine checks to see if the bdev size does not match the disk size
1051 * and adjusts it if it differs.
1053 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1055 loff_t disk_size
, bdev_size
;
1057 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1058 bdev_size
= i_size_read(bdev
->bd_inode
);
1059 if (disk_size
!= bdev_size
) {
1060 char name
[BDEVNAME_SIZE
];
1062 disk_name(disk
, 0, name
);
1064 "%s: detected capacity change from %lld to %lld\n",
1065 name
, bdev_size
, disk_size
);
1066 i_size_write(bdev
->bd_inode
, disk_size
);
1067 flush_disk(bdev
, false);
1070 EXPORT_SYMBOL(check_disk_size_change
);
1073 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1074 * @disk: struct gendisk to be revalidated
1076 * This routine is a wrapper for lower-level driver's revalidate_disk
1077 * call-backs. It is used to do common pre and post operations needed
1078 * for all revalidate_disk operations.
1080 int revalidate_disk(struct gendisk
*disk
)
1082 struct block_device
*bdev
;
1085 if (disk
->fops
->revalidate_disk
)
1086 ret
= disk
->fops
->revalidate_disk(disk
);
1087 blk_integrity_revalidate(disk
);
1088 bdev
= bdget_disk(disk
, 0);
1092 mutex_lock(&bdev
->bd_mutex
);
1093 check_disk_size_change(disk
, bdev
);
1094 bdev
->bd_invalidated
= 0;
1095 mutex_unlock(&bdev
->bd_mutex
);
1099 EXPORT_SYMBOL(revalidate_disk
);
1102 * This routine checks whether a removable media has been changed,
1103 * and invalidates all buffer-cache-entries in that case. This
1104 * is a relatively slow routine, so we have to try to minimize using
1105 * it. Thus it is called only upon a 'mount' or 'open'. This
1106 * is the best way of combining speed and utility, I think.
1107 * People changing diskettes in the middle of an operation deserve
1110 int check_disk_change(struct block_device
*bdev
)
1112 struct gendisk
*disk
= bdev
->bd_disk
;
1113 const struct block_device_operations
*bdops
= disk
->fops
;
1114 unsigned int events
;
1116 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1117 DISK_EVENT_EJECT_REQUEST
);
1118 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1121 flush_disk(bdev
, true);
1122 if (bdops
->revalidate_disk
)
1123 bdops
->revalidate_disk(bdev
->bd_disk
);
1127 EXPORT_SYMBOL(check_disk_change
);
1129 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1131 unsigned bsize
= bdev_logical_block_size(bdev
);
1133 mutex_lock(&bdev
->bd_inode
->i_mutex
);
1134 i_size_write(bdev
->bd_inode
, size
);
1135 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
1136 while (bsize
< PAGE_CACHE_SIZE
) {
1141 bdev
->bd_block_size
= bsize
;
1142 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1144 EXPORT_SYMBOL(bd_set_size
);
1146 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1151 * mutex_lock(part->bd_mutex)
1152 * mutex_lock_nested(whole->bd_mutex, 1)
1155 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1157 struct gendisk
*disk
;
1158 struct module
*owner
;
1163 if (mode
& FMODE_READ
)
1165 if (mode
& FMODE_WRITE
)
1168 * hooks: /n/, see "layering violations".
1171 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1181 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1184 owner
= disk
->fops
->owner
;
1186 disk_block_events(disk
);
1187 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1188 if (!bdev
->bd_openers
) {
1189 bdev
->bd_disk
= disk
;
1190 bdev
->bd_queue
= disk
->queue
;
1191 bdev
->bd_contains
= bdev
;
1192 bdev
->bd_inode
->i_flags
= disk
->fops
->direct_access
? S_DAX
: 0;
1195 bdev
->bd_part
= disk_get_part(disk
, partno
);
1200 if (disk
->fops
->open
) {
1201 ret
= disk
->fops
->open(bdev
, mode
);
1202 if (ret
== -ERESTARTSYS
) {
1203 /* Lost a race with 'disk' being
1204 * deleted, try again.
1207 disk_put_part(bdev
->bd_part
);
1208 bdev
->bd_part
= NULL
;
1209 bdev
->bd_disk
= NULL
;
1210 bdev
->bd_queue
= NULL
;
1211 mutex_unlock(&bdev
->bd_mutex
);
1212 disk_unblock_events(disk
);
1220 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1223 * If the device is invalidated, rescan partition
1224 * if open succeeded or failed with -ENOMEDIUM.
1225 * The latter is necessary to prevent ghost
1226 * partitions on a removed medium.
1228 if (bdev
->bd_invalidated
) {
1230 rescan_partitions(disk
, bdev
);
1231 else if (ret
== -ENOMEDIUM
)
1232 invalidate_partitions(disk
, bdev
);
1237 struct block_device
*whole
;
1238 whole
= bdget_disk(disk
, 0);
1243 ret
= __blkdev_get(whole
, mode
, 1);
1246 bdev
->bd_contains
= whole
;
1247 bdev
->bd_part
= disk_get_part(disk
, partno
);
1248 if (!(disk
->flags
& GENHD_FL_UP
) ||
1249 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1253 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1255 * If the partition is not aligned on a page
1256 * boundary, we can't do dax I/O to it.
1258 if ((bdev
->bd_part
->start_sect
% (PAGE_SIZE
/ 512)) ||
1259 (bdev
->bd_part
->nr_sects
% (PAGE_SIZE
/ 512)))
1260 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1263 if (bdev
->bd_contains
== bdev
) {
1265 if (bdev
->bd_disk
->fops
->open
)
1266 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1267 /* the same as first opener case, read comment there */
1268 if (bdev
->bd_invalidated
) {
1270 rescan_partitions(bdev
->bd_disk
, bdev
);
1271 else if (ret
== -ENOMEDIUM
)
1272 invalidate_partitions(bdev
->bd_disk
, bdev
);
1275 goto out_unlock_bdev
;
1277 /* only one opener holds refs to the module and disk */
1283 bdev
->bd_part_count
++;
1284 mutex_unlock(&bdev
->bd_mutex
);
1285 disk_unblock_events(disk
);
1289 disk_put_part(bdev
->bd_part
);
1290 bdev
->bd_disk
= NULL
;
1291 bdev
->bd_part
= NULL
;
1292 bdev
->bd_queue
= NULL
;
1293 if (bdev
!= bdev
->bd_contains
)
1294 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1295 bdev
->bd_contains
= NULL
;
1297 mutex_unlock(&bdev
->bd_mutex
);
1298 disk_unblock_events(disk
);
1308 * blkdev_get - open a block device
1309 * @bdev: block_device to open
1310 * @mode: FMODE_* mask
1311 * @holder: exclusive holder identifier
1313 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1314 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1315 * @holder is invalid. Exclusive opens may nest for the same @holder.
1317 * On success, the reference count of @bdev is unchanged. On failure,
1324 * 0 on success, -errno on failure.
1326 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1328 struct block_device
*whole
= NULL
;
1331 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1333 if ((mode
& FMODE_EXCL
) && holder
) {
1334 whole
= bd_start_claiming(bdev
, holder
);
1335 if (IS_ERR(whole
)) {
1337 return PTR_ERR(whole
);
1341 res
= __blkdev_get(bdev
, mode
, 0);
1344 struct gendisk
*disk
= whole
->bd_disk
;
1346 /* finish claiming */
1347 mutex_lock(&bdev
->bd_mutex
);
1348 spin_lock(&bdev_lock
);
1351 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1353 * Note that for a whole device bd_holders
1354 * will be incremented twice, and bd_holder
1355 * will be set to bd_may_claim before being
1358 whole
->bd_holders
++;
1359 whole
->bd_holder
= bd_may_claim
;
1361 bdev
->bd_holder
= holder
;
1364 /* tell others that we're done */
1365 BUG_ON(whole
->bd_claiming
!= holder
);
1366 whole
->bd_claiming
= NULL
;
1367 wake_up_bit(&whole
->bd_claiming
, 0);
1369 spin_unlock(&bdev_lock
);
1372 * Block event polling for write claims if requested. Any
1373 * write holder makes the write_holder state stick until
1374 * all are released. This is good enough and tracking
1375 * individual writeable reference is too fragile given the
1376 * way @mode is used in blkdev_get/put().
1378 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1379 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1380 bdev
->bd_write_holder
= true;
1381 disk_block_events(disk
);
1384 mutex_unlock(&bdev
->bd_mutex
);
1390 EXPORT_SYMBOL(blkdev_get
);
1393 * blkdev_get_by_path - open a block device by name
1394 * @path: path to the block device to open
1395 * @mode: FMODE_* mask
1396 * @holder: exclusive holder identifier
1398 * Open the blockdevice described by the device file at @path. @mode
1399 * and @holder are identical to blkdev_get().
1401 * On success, the returned block_device has reference count of one.
1407 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1409 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1412 struct block_device
*bdev
;
1415 bdev
= lookup_bdev(path
);
1419 err
= blkdev_get(bdev
, mode
, holder
);
1421 return ERR_PTR(err
);
1423 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1424 blkdev_put(bdev
, mode
);
1425 return ERR_PTR(-EACCES
);
1430 EXPORT_SYMBOL(blkdev_get_by_path
);
1433 * blkdev_get_by_dev - open a block device by device number
1434 * @dev: device number of block device to open
1435 * @mode: FMODE_* mask
1436 * @holder: exclusive holder identifier
1438 * Open the blockdevice described by device number @dev. @mode and
1439 * @holder are identical to blkdev_get().
1441 * Use it ONLY if you really do not have anything better - i.e. when
1442 * you are behind a truly sucky interface and all you are given is a
1443 * device number. _Never_ to be used for internal purposes. If you
1444 * ever need it - reconsider your API.
1446 * On success, the returned block_device has reference count of one.
1452 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1454 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1456 struct block_device
*bdev
;
1461 return ERR_PTR(-ENOMEM
);
1463 err
= blkdev_get(bdev
, mode
, holder
);
1465 return ERR_PTR(err
);
1469 EXPORT_SYMBOL(blkdev_get_by_dev
);
1471 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1473 struct block_device
*bdev
;
1476 * Preserve backwards compatibility and allow large file access
1477 * even if userspace doesn't ask for it explicitly. Some mkfs
1478 * binary needs it. We might want to drop this workaround
1479 * during an unstable branch.
1481 filp
->f_flags
|= O_LARGEFILE
;
1483 if (filp
->f_flags
& O_NDELAY
)
1484 filp
->f_mode
|= FMODE_NDELAY
;
1485 if (filp
->f_flags
& O_EXCL
)
1486 filp
->f_mode
|= FMODE_EXCL
;
1487 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1488 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1490 bdev
= bd_acquire(inode
);
1494 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1496 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1499 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1501 struct gendisk
*disk
= bdev
->bd_disk
;
1502 struct block_device
*victim
= NULL
;
1504 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1506 bdev
->bd_part_count
--;
1508 if (!--bdev
->bd_openers
) {
1509 WARN_ON_ONCE(bdev
->bd_holders
);
1510 sync_blockdev(bdev
);
1513 * ->release can cause the queue to disappear, so flush all
1514 * dirty data before.
1516 bdev_write_inode(bdev
);
1518 if (bdev
->bd_contains
== bdev
) {
1519 if (disk
->fops
->release
)
1520 disk
->fops
->release(disk
, mode
);
1522 if (!bdev
->bd_openers
) {
1523 struct module
*owner
= disk
->fops
->owner
;
1525 disk_put_part(bdev
->bd_part
);
1526 bdev
->bd_part
= NULL
;
1527 bdev
->bd_disk
= NULL
;
1528 if (bdev
!= bdev
->bd_contains
)
1529 victim
= bdev
->bd_contains
;
1530 bdev
->bd_contains
= NULL
;
1535 mutex_unlock(&bdev
->bd_mutex
);
1538 __blkdev_put(victim
, mode
, 1);
1541 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1543 mutex_lock(&bdev
->bd_mutex
);
1545 if (mode
& FMODE_EXCL
) {
1549 * Release a claim on the device. The holder fields
1550 * are protected with bdev_lock. bd_mutex is to
1551 * synchronize disk_holder unlinking.
1553 spin_lock(&bdev_lock
);
1555 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1556 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1558 /* bd_contains might point to self, check in a separate step */
1559 if ((bdev_free
= !bdev
->bd_holders
))
1560 bdev
->bd_holder
= NULL
;
1561 if (!bdev
->bd_contains
->bd_holders
)
1562 bdev
->bd_contains
->bd_holder
= NULL
;
1564 spin_unlock(&bdev_lock
);
1567 * If this was the last claim, remove holder link and
1568 * unblock evpoll if it was a write holder.
1570 if (bdev_free
&& bdev
->bd_write_holder
) {
1571 disk_unblock_events(bdev
->bd_disk
);
1572 bdev
->bd_write_holder
= false;
1577 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1578 * event. This is to ensure detection of media removal commanded
1579 * from userland - e.g. eject(1).
1581 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1583 mutex_unlock(&bdev
->bd_mutex
);
1585 __blkdev_put(bdev
, mode
, 0);
1587 EXPORT_SYMBOL(blkdev_put
);
1589 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1591 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1592 blkdev_put(bdev
, filp
->f_mode
);
1596 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1598 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1599 fmode_t mode
= file
->f_mode
;
1602 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1603 * to updated it before every ioctl.
1605 if (file
->f_flags
& O_NDELAY
)
1606 mode
|= FMODE_NDELAY
;
1608 mode
&= ~FMODE_NDELAY
;
1610 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1614 * Write data to the block device. Only intended for the block device itself
1615 * and the raw driver which basically is a fake block device.
1617 * Does not take i_mutex for the write and thus is not for general purpose
1620 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1622 struct file
*file
= iocb
->ki_filp
;
1623 struct inode
*bd_inode
= file
->f_mapping
->host
;
1624 loff_t size
= i_size_read(bd_inode
);
1625 struct blk_plug plug
;
1628 if (bdev_read_only(I_BDEV(bd_inode
)))
1631 if (!iov_iter_count(from
))
1634 if (iocb
->ki_pos
>= size
)
1637 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1639 blk_start_plug(&plug
);
1640 ret
= __generic_file_write_iter(iocb
, from
);
1643 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1647 blk_finish_plug(&plug
);
1650 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1652 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1654 struct file
*file
= iocb
->ki_filp
;
1655 struct inode
*bd_inode
= file
->f_mapping
->host
;
1656 loff_t size
= i_size_read(bd_inode
);
1657 loff_t pos
= iocb
->ki_pos
;
1663 iov_iter_truncate(to
, size
);
1664 return generic_file_read_iter(iocb
, to
);
1666 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1669 * Try to release a page associated with block device when the system
1670 * is under memory pressure.
1672 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1674 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1676 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1677 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1679 return try_to_free_buffers(page
);
1682 static const struct address_space_operations def_blk_aops
= {
1683 .readpage
= blkdev_readpage
,
1684 .readpages
= blkdev_readpages
,
1685 .writepage
= blkdev_writepage
,
1686 .write_begin
= blkdev_write_begin
,
1687 .write_end
= blkdev_write_end
,
1688 .writepages
= generic_writepages
,
1689 .releasepage
= blkdev_releasepage
,
1690 .direct_IO
= blkdev_direct_IO
,
1691 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1694 const struct file_operations def_blk_fops
= {
1695 .open
= blkdev_open
,
1696 .release
= blkdev_close
,
1697 .llseek
= block_llseek
,
1698 .read_iter
= blkdev_read_iter
,
1699 .write_iter
= blkdev_write_iter
,
1700 .mmap
= generic_file_mmap
,
1701 .fsync
= blkdev_fsync
,
1702 .unlocked_ioctl
= block_ioctl
,
1703 #ifdef CONFIG_COMPAT
1704 .compat_ioctl
= compat_blkdev_ioctl
,
1706 .splice_read
= generic_file_splice_read
,
1707 .splice_write
= iter_file_splice_write
,
1710 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1713 mm_segment_t old_fs
= get_fs();
1715 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1720 EXPORT_SYMBOL(ioctl_by_bdev
);
1723 * lookup_bdev - lookup a struct block_device by name
1724 * @pathname: special file representing the block device
1726 * Get a reference to the blockdevice at @pathname in the current
1727 * namespace if possible and return it. Return ERR_PTR(error)
1730 struct block_device
*lookup_bdev(const char *pathname
)
1732 struct block_device
*bdev
;
1733 struct inode
*inode
;
1737 if (!pathname
|| !*pathname
)
1738 return ERR_PTR(-EINVAL
);
1740 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1742 return ERR_PTR(error
);
1744 inode
= d_backing_inode(path
.dentry
);
1746 if (!S_ISBLK(inode
->i_mode
))
1749 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1752 bdev
= bd_acquire(inode
);
1759 bdev
= ERR_PTR(error
);
1762 EXPORT_SYMBOL(lookup_bdev
);
1764 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1766 struct super_block
*sb
= get_super(bdev
);
1771 * no need to lock the super, get_super holds the
1772 * read mutex so the filesystem cannot go away
1773 * under us (->put_super runs with the write lock
1776 shrink_dcache_sb(sb
);
1777 res
= invalidate_inodes(sb
, kill_dirty
);
1780 invalidate_bdev(bdev
);
1783 EXPORT_SYMBOL(__invalidate_device
);
1785 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1787 struct inode
*inode
, *old_inode
= NULL
;
1789 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1790 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1791 struct address_space
*mapping
= inode
->i_mapping
;
1793 spin_lock(&inode
->i_lock
);
1794 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1795 mapping
->nrpages
== 0) {
1796 spin_unlock(&inode
->i_lock
);
1800 spin_unlock(&inode
->i_lock
);
1801 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1803 * We hold a reference to 'inode' so it couldn't have been
1804 * removed from s_inodes list while we dropped the
1805 * s_inode_list_lock We cannot iput the inode now as we can
1806 * be holding the last reference and we cannot iput it under
1807 * s_inode_list_lock. So we keep the reference and iput it
1813 func(I_BDEV(inode
), arg
);
1815 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1817 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);