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
) {
93 lru_add_drain_all(); /* make sure all lru add caches are flushed */
94 invalidate_mapping_pages(mapping
, 0, -1);
96 /* 99% of the time, we don't need to flush the cleancache on the bdev.
97 * But, for the strange corners, lets be cautious
99 cleancache_invalidate_inode(mapping
);
101 EXPORT_SYMBOL(invalidate_bdev
);
103 int set_blocksize(struct block_device
*bdev
, int size
)
105 /* Size must be a power of two, and between 512 and PAGE_SIZE */
106 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
109 /* Size cannot be smaller than the size supported by the device */
110 if (size
< bdev_logical_block_size(bdev
))
113 /* Don't change the size if it is same as current */
114 if (bdev
->bd_block_size
!= size
) {
116 bdev
->bd_block_size
= size
;
117 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
123 EXPORT_SYMBOL(set_blocksize
);
125 int sb_set_blocksize(struct super_block
*sb
, int size
)
127 if (set_blocksize(sb
->s_bdev
, size
))
129 /* If we get here, we know size is power of two
130 * and it's value is between 512 and PAGE_SIZE */
131 sb
->s_blocksize
= size
;
132 sb
->s_blocksize_bits
= blksize_bits(size
);
133 return sb
->s_blocksize
;
136 EXPORT_SYMBOL(sb_set_blocksize
);
138 int sb_min_blocksize(struct super_block
*sb
, int size
)
140 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
143 return sb_set_blocksize(sb
, size
);
146 EXPORT_SYMBOL(sb_min_blocksize
);
149 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
150 struct buffer_head
*bh
, int create
)
152 bh
->b_bdev
= I_BDEV(inode
);
153 bh
->b_blocknr
= iblock
;
154 set_buffer_mapped(bh
);
159 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
161 struct file
*file
= iocb
->ki_filp
;
162 struct inode
*inode
= file
->f_mapping
->host
;
165 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
166 NULL
, DIO_SKIP_DIO_COUNT
);
167 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
168 blkdev_get_block
, NULL
, NULL
,
172 int __sync_blockdev(struct block_device
*bdev
, int wait
)
177 return filemap_flush(bdev
->bd_inode
->i_mapping
);
178 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
182 * Write out and wait upon all the dirty data associated with a block
183 * device via its mapping. Does not take the superblock lock.
185 int sync_blockdev(struct block_device
*bdev
)
187 return __sync_blockdev(bdev
, 1);
189 EXPORT_SYMBOL(sync_blockdev
);
192 * Write out and wait upon all dirty data associated with this
193 * device. Filesystem data as well as the underlying block
194 * device. Takes the superblock lock.
196 int fsync_bdev(struct block_device
*bdev
)
198 struct super_block
*sb
= get_super(bdev
);
200 int res
= sync_filesystem(sb
);
204 return sync_blockdev(bdev
);
206 EXPORT_SYMBOL(fsync_bdev
);
209 * freeze_bdev -- lock a filesystem and force it into a consistent state
210 * @bdev: blockdevice to lock
212 * If a superblock is found on this device, we take the s_umount semaphore
213 * on it to make sure nobody unmounts until the snapshot creation is done.
214 * The reference counter (bd_fsfreeze_count) guarantees that only the last
215 * unfreeze process can unfreeze the frozen filesystem actually when multiple
216 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
217 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
220 struct super_block
*freeze_bdev(struct block_device
*bdev
)
222 struct super_block
*sb
;
225 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
226 if (++bdev
->bd_fsfreeze_count
> 1) {
228 * We don't even need to grab a reference - the first call
229 * to freeze_bdev grab an active reference and only the last
230 * thaw_bdev drops it.
232 sb
= get_super(bdev
);
234 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
238 sb
= get_active_super(bdev
);
241 if (sb
->s_op
->freeze_super
)
242 error
= sb
->s_op
->freeze_super(sb
);
244 error
= freeze_super(sb
);
246 deactivate_super(sb
);
247 bdev
->bd_fsfreeze_count
--;
248 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
249 return ERR_PTR(error
);
251 deactivate_super(sb
);
254 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
255 return sb
; /* thaw_bdev releases s->s_umount */
257 EXPORT_SYMBOL(freeze_bdev
);
260 * thaw_bdev -- unlock filesystem
261 * @bdev: blockdevice to unlock
262 * @sb: associated superblock
264 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
266 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
270 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
271 if (!bdev
->bd_fsfreeze_count
)
275 if (--bdev
->bd_fsfreeze_count
> 0)
281 if (sb
->s_op
->thaw_super
)
282 error
= sb
->s_op
->thaw_super(sb
);
284 error
= thaw_super(sb
);
286 bdev
->bd_fsfreeze_count
++;
287 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
291 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
294 EXPORT_SYMBOL(thaw_bdev
);
296 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
298 return block_write_full_page(page
, blkdev_get_block
, wbc
);
301 static int blkdev_readpage(struct file
* file
, struct page
* page
)
303 return block_read_full_page(page
, blkdev_get_block
);
306 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
307 struct list_head
*pages
, unsigned nr_pages
)
309 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
312 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
313 loff_t pos
, unsigned len
, unsigned flags
,
314 struct page
**pagep
, void **fsdata
)
316 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
320 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
321 loff_t pos
, unsigned len
, unsigned copied
,
322 struct page
*page
, void *fsdata
)
325 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
328 page_cache_release(page
);
335 * for a block special file file_inode(file)->i_size is zero
336 * so we compute the size by hand (just as in block_read/write above)
338 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
340 struct inode
*bd_inode
= file
->f_mapping
->host
;
343 mutex_lock(&bd_inode
->i_mutex
);
344 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
345 mutex_unlock(&bd_inode
->i_mutex
);
349 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
351 struct inode
*bd_inode
= filp
->f_mapping
->host
;
352 struct block_device
*bdev
= I_BDEV(bd_inode
);
355 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
360 * There is no need to serialise calls to blkdev_issue_flush with
361 * i_mutex and doing so causes performance issues with concurrent
362 * O_SYNC writers to a block device.
364 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
365 if (error
== -EOPNOTSUPP
)
370 EXPORT_SYMBOL(blkdev_fsync
);
373 * bdev_read_page() - Start reading a page from a block device
374 * @bdev: The device to read the page from
375 * @sector: The offset on the device to read the page to (need not be aligned)
376 * @page: The page to read
378 * On entry, the page should be locked. It will be unlocked when the page
379 * has been read. If the block driver implements rw_page synchronously,
380 * that will be true on exit from this function, but it need not be.
382 * Errors returned by this function are usually "soft", eg out of memory, or
383 * queue full; callers should try a different route to read this page rather
384 * than propagate an error back up the stack.
386 * Return: negative errno if an error occurs, 0 if submission was successful.
388 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
391 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
392 int result
= -EOPNOTSUPP
;
394 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
397 result
= blk_queue_enter(bdev
->bd_queue
, false);
400 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
401 blk_queue_exit(bdev
->bd_queue
);
404 EXPORT_SYMBOL_GPL(bdev_read_page
);
407 * bdev_write_page() - Start writing a page to a block device
408 * @bdev: The device to write the page to
409 * @sector: The offset on the device to write the page to (need not be aligned)
410 * @page: The page to write
411 * @wbc: The writeback_control for the write
413 * On entry, the page should be locked and not currently under writeback.
414 * On exit, if the write started successfully, the page will be unlocked and
415 * under writeback. If the write failed already (eg the driver failed to
416 * queue the page to the device), the page will still be locked. If the
417 * caller is a ->writepage implementation, it will need to unlock the page.
419 * Errors returned by this function are usually "soft", eg out of memory, or
420 * queue full; callers should try a different route to write this page rather
421 * than propagate an error back up the stack.
423 * Return: negative errno if an error occurs, 0 if submission was successful.
425 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
426 struct page
*page
, struct writeback_control
*wbc
)
429 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
430 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
432 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
434 result
= blk_queue_enter(bdev
->bd_queue
, false);
438 set_page_writeback(page
);
439 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
441 end_page_writeback(page
);
444 blk_queue_exit(bdev
->bd_queue
);
447 EXPORT_SYMBOL_GPL(bdev_write_page
);
450 * bdev_direct_access() - Get the address for directly-accessibly memory
451 * @bdev: The device containing the memory
452 * @sector: The offset within the device
453 * @addr: Where to put the address of the memory
454 * @pfn: The Page Frame Number for the memory
455 * @size: The number of bytes requested
457 * If a block device is made up of directly addressable memory, this function
458 * will tell the caller the PFN and the address of the memory. The address
459 * may be directly dereferenced within the kernel without the need to call
460 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
463 * Return: negative errno if an error occurs, otherwise the number of bytes
464 * accessible at this address.
466 long bdev_direct_access(struct block_device
*bdev
, sector_t sector
,
467 void __pmem
**addr
, unsigned long *pfn
, long size
)
470 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
473 * The device driver is allowed to sleep, in order to make the
474 * memory directly accessible.
480 if (!ops
->direct_access
)
482 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
483 part_nr_sects_read(bdev
->bd_part
))
485 sector
+= get_start_sect(bdev
);
486 if (sector
% (PAGE_SIZE
/ 512))
488 avail
= ops
->direct_access(bdev
, sector
, addr
, pfn
);
491 return min(avail
, size
);
493 EXPORT_SYMBOL_GPL(bdev_direct_access
);
499 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
500 static struct kmem_cache
* bdev_cachep __read_mostly
;
502 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
504 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
507 return &ei
->vfs_inode
;
510 static void bdev_i_callback(struct rcu_head
*head
)
512 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
513 struct bdev_inode
*bdi
= BDEV_I(inode
);
515 kmem_cache_free(bdev_cachep
, bdi
);
518 static void bdev_destroy_inode(struct inode
*inode
)
520 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
523 static void init_once(void *foo
)
525 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
526 struct block_device
*bdev
= &ei
->bdev
;
528 memset(bdev
, 0, sizeof(*bdev
));
529 mutex_init(&bdev
->bd_mutex
);
530 INIT_LIST_HEAD(&bdev
->bd_inodes
);
531 INIT_LIST_HEAD(&bdev
->bd_list
);
533 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
535 bdev
->bd_bdi
= &noop_backing_dev_info
;
536 inode_init_once(&ei
->vfs_inode
);
537 /* Initialize mutex for freeze. */
538 mutex_init(&bdev
->bd_fsfreeze_mutex
);
541 static inline void __bd_forget(struct inode
*inode
)
543 list_del_init(&inode
->i_devices
);
544 inode
->i_bdev
= NULL
;
545 inode
->i_mapping
= &inode
->i_data
;
548 static void bdev_evict_inode(struct inode
*inode
)
550 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
552 truncate_inode_pages_final(&inode
->i_data
);
553 invalidate_inode_buffers(inode
); /* is it needed here? */
555 spin_lock(&bdev_lock
);
556 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
557 __bd_forget(list_entry(p
, struct inode
, i_devices
));
559 list_del_init(&bdev
->bd_list
);
560 spin_unlock(&bdev_lock
);
561 /* Detach inode from wb early as bdi_put() may free bdi->wb */
562 inode_detach_wb(inode
);
563 if (bdev
->bd_bdi
!= &noop_backing_dev_info
) {
564 bdi_put(bdev
->bd_bdi
);
565 bdev
->bd_bdi
= &noop_backing_dev_info
;
569 static const struct super_operations bdev_sops
= {
570 .statfs
= simple_statfs
,
571 .alloc_inode
= bdev_alloc_inode
,
572 .destroy_inode
= bdev_destroy_inode
,
573 .drop_inode
= generic_delete_inode
,
574 .evict_inode
= bdev_evict_inode
,
577 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
578 int flags
, const char *dev_name
, void *data
)
580 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
583 static struct file_system_type bd_type
= {
586 .kill_sb
= kill_anon_super
,
589 struct super_block
*blockdev_superblock __read_mostly
;
590 EXPORT_SYMBOL_GPL(blockdev_superblock
);
592 void __init
bdev_cache_init(void)
595 static struct vfsmount
*bd_mnt
;
597 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
598 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
599 SLAB_MEM_SPREAD
|SLAB_PANIC
),
601 err
= register_filesystem(&bd_type
);
603 panic("Cannot register bdev pseudo-fs");
604 bd_mnt
= kern_mount(&bd_type
);
606 panic("Cannot create bdev pseudo-fs");
607 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
611 * Most likely _very_ bad one - but then it's hardly critical for small
612 * /dev and can be fixed when somebody will need really large one.
613 * Keep in mind that it will be fed through icache hash function too.
615 static inline unsigned long hash(dev_t dev
)
617 return MAJOR(dev
)+MINOR(dev
);
620 static int bdev_test(struct inode
*inode
, void *data
)
622 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
625 static int bdev_set(struct inode
*inode
, void *data
)
627 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
631 static LIST_HEAD(all_bdevs
);
634 * If there is a bdev inode for this device, unhash it so that it gets evicted
635 * as soon as last inode reference is dropped.
637 void bdev_unhash_inode(dev_t dev
)
641 inode
= ilookup5(blockdev_superblock
, hash(dev
), bdev_test
, &dev
);
643 remove_inode_hash(inode
);
648 struct block_device
*bdget(dev_t dev
)
650 struct block_device
*bdev
;
653 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
654 bdev_test
, bdev_set
, &dev
);
659 bdev
= &BDEV_I(inode
)->bdev
;
661 if (inode
->i_state
& I_NEW
) {
662 bdev
->bd_contains
= NULL
;
663 bdev
->bd_super
= NULL
;
664 bdev
->bd_inode
= inode
;
665 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
666 bdev
->bd_part_count
= 0;
667 bdev
->bd_invalidated
= 0;
668 inode
->i_mode
= S_IFBLK
;
670 inode
->i_bdev
= bdev
;
671 inode
->i_data
.a_ops
= &def_blk_aops
;
672 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
673 spin_lock(&bdev_lock
);
674 list_add(&bdev
->bd_list
, &all_bdevs
);
675 spin_unlock(&bdev_lock
);
676 unlock_new_inode(inode
);
681 EXPORT_SYMBOL(bdget
);
684 * bdgrab -- Grab a reference to an already referenced block device
685 * @bdev: Block device to grab a reference to.
687 struct block_device
*bdgrab(struct block_device
*bdev
)
689 ihold(bdev
->bd_inode
);
692 EXPORT_SYMBOL(bdgrab
);
694 long nr_blockdev_pages(void)
696 struct block_device
*bdev
;
698 spin_lock(&bdev_lock
);
699 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
700 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
702 spin_unlock(&bdev_lock
);
706 void bdput(struct block_device
*bdev
)
708 iput(bdev
->bd_inode
);
711 EXPORT_SYMBOL(bdput
);
713 static struct block_device
*bd_acquire(struct inode
*inode
)
715 struct block_device
*bdev
;
717 spin_lock(&bdev_lock
);
718 bdev
= inode
->i_bdev
;
719 if (bdev
&& !inode_unhashed(bdev
->bd_inode
)) {
720 ihold(bdev
->bd_inode
);
721 spin_unlock(&bdev_lock
);
724 spin_unlock(&bdev_lock
);
727 * i_bdev references block device inode that was already shut down
728 * (corresponding device got removed). Remove the reference and look
729 * up block device inode again just in case new device got
730 * reestablished under the same device number.
735 bdev
= bdget(inode
->i_rdev
);
737 spin_lock(&bdev_lock
);
738 if (!inode
->i_bdev
) {
740 * We take an additional reference to bd_inode,
741 * and it's released in clear_inode() of inode.
742 * So, we can access it via ->i_mapping always
745 ihold(bdev
->bd_inode
);
746 inode
->i_bdev
= bdev
;
747 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
748 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
750 spin_unlock(&bdev_lock
);
755 /* Call when you free inode */
757 void bd_forget(struct inode
*inode
)
759 struct block_device
*bdev
= NULL
;
761 spin_lock(&bdev_lock
);
762 if (!sb_is_blkdev_sb(inode
->i_sb
))
763 bdev
= inode
->i_bdev
;
765 spin_unlock(&bdev_lock
);
768 iput(bdev
->bd_inode
);
772 * bd_may_claim - test whether a block device can be claimed
773 * @bdev: block device of interest
774 * @whole: whole block device containing @bdev, may equal @bdev
775 * @holder: holder trying to claim @bdev
777 * Test whether @bdev can be claimed by @holder.
780 * spin_lock(&bdev_lock).
783 * %true if @bdev can be claimed, %false otherwise.
785 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
788 if (bdev
->bd_holder
== holder
)
789 return true; /* already a holder */
790 else if (bdev
->bd_holder
!= NULL
)
791 return false; /* held by someone else */
792 else if (whole
== bdev
)
793 return true; /* is a whole device which isn't held */
795 else if (whole
->bd_holder
== bd_may_claim
)
796 return true; /* is a partition of a device that is being partitioned */
797 else if (whole
->bd_holder
!= NULL
)
798 return false; /* is a partition of a held device */
800 return true; /* is a partition of an un-held device */
804 * bd_prepare_to_claim - prepare to claim a block device
805 * @bdev: block device of interest
806 * @whole: the whole device containing @bdev, may equal @bdev
807 * @holder: holder trying to claim @bdev
809 * Prepare to claim @bdev. This function fails if @bdev is already
810 * claimed by another holder and waits if another claiming is in
811 * progress. This function doesn't actually claim. On successful
812 * return, the caller has ownership of bd_claiming and bd_holder[s].
815 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
819 * 0 if @bdev can be claimed, -EBUSY otherwise.
821 static int bd_prepare_to_claim(struct block_device
*bdev
,
822 struct block_device
*whole
, void *holder
)
825 /* if someone else claimed, fail */
826 if (!bd_may_claim(bdev
, whole
, holder
))
829 /* if claiming is already in progress, wait for it to finish */
830 if (whole
->bd_claiming
) {
831 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
834 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
835 spin_unlock(&bdev_lock
);
837 finish_wait(wq
, &wait
);
838 spin_lock(&bdev_lock
);
847 * bd_start_claiming - start claiming a block device
848 * @bdev: block device of interest
849 * @holder: holder trying to claim @bdev
851 * @bdev is about to be opened exclusively. Check @bdev can be opened
852 * exclusively and mark that an exclusive open is in progress. Each
853 * successful call to this function must be matched with a call to
854 * either bd_finish_claiming() or bd_abort_claiming() (which do not
857 * This function is used to gain exclusive access to the block device
858 * without actually causing other exclusive open attempts to fail. It
859 * should be used when the open sequence itself requires exclusive
860 * access but may subsequently fail.
866 * Pointer to the block device containing @bdev on success, ERR_PTR()
869 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
872 struct gendisk
*disk
;
873 struct block_device
*whole
;
879 * @bdev might not have been initialized properly yet, look up
880 * and grab the outer block device the hard way.
882 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
884 return ERR_PTR(-ENXIO
);
887 * Normally, @bdev should equal what's returned from bdget_disk()
888 * if partno is 0; however, some drivers (floppy) use multiple
889 * bdev's for the same physical device and @bdev may be one of the
890 * aliases. Keep @bdev if partno is 0. This means claimer
891 * tracking is broken for those devices but it has always been that
895 whole
= bdget_disk(disk
, 0);
897 whole
= bdgrab(bdev
);
899 module_put(disk
->fops
->owner
);
902 return ERR_PTR(-ENOMEM
);
904 /* prepare to claim, if successful, mark claiming in progress */
905 spin_lock(&bdev_lock
);
907 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
909 whole
->bd_claiming
= holder
;
910 spin_unlock(&bdev_lock
);
913 spin_unlock(&bdev_lock
);
920 struct bd_holder_disk
{
921 struct list_head list
;
922 struct gendisk
*disk
;
926 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
927 struct gendisk
*disk
)
929 struct bd_holder_disk
*holder
;
931 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
932 if (holder
->disk
== disk
)
937 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
939 return sysfs_create_link(from
, to
, kobject_name(to
));
942 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
944 sysfs_remove_link(from
, kobject_name(to
));
948 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
949 * @bdev: the claimed slave bdev
950 * @disk: the holding disk
952 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
954 * This functions creates the following sysfs symlinks.
956 * - from "slaves" directory of the holder @disk to the claimed @bdev
957 * - from "holders" directory of the @bdev to the holder @disk
959 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
960 * passed to bd_link_disk_holder(), then:
962 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
963 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
965 * The caller must have claimed @bdev before calling this function and
966 * ensure that both @bdev and @disk are valid during the creation and
967 * lifetime of these symlinks.
973 * 0 on success, -errno on failure.
975 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
977 struct bd_holder_disk
*holder
;
980 mutex_lock(&bdev
->bd_mutex
);
982 WARN_ON_ONCE(!bdev
->bd_holder
);
984 /* FIXME: remove the following once add_disk() handles errors */
985 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
988 holder
= bd_find_holder_disk(bdev
, disk
);
994 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1000 INIT_LIST_HEAD(&holder
->list
);
1001 holder
->disk
= disk
;
1004 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1008 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1012 * bdev could be deleted beneath us which would implicitly destroy
1013 * the holder directory. Hold on to it.
1015 kobject_get(bdev
->bd_part
->holder_dir
);
1017 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1021 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1025 mutex_unlock(&bdev
->bd_mutex
);
1028 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1031 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1032 * @bdev: the calimed slave bdev
1033 * @disk: the holding disk
1035 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1040 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1042 struct bd_holder_disk
*holder
;
1044 mutex_lock(&bdev
->bd_mutex
);
1046 holder
= bd_find_holder_disk(bdev
, disk
);
1048 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1049 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1050 del_symlink(bdev
->bd_part
->holder_dir
,
1051 &disk_to_dev(disk
)->kobj
);
1052 kobject_put(bdev
->bd_part
->holder_dir
);
1053 list_del_init(&holder
->list
);
1057 mutex_unlock(&bdev
->bd_mutex
);
1059 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1063 * flush_disk - invalidates all buffer-cache entries on a disk
1065 * @bdev: struct block device to be flushed
1066 * @kill_dirty: flag to guide handling of dirty inodes
1068 * Invalidates all buffer-cache entries on a disk. It should be called
1069 * when a disk has been changed -- either by a media change or online
1072 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1074 if (__invalidate_device(bdev
, kill_dirty
)) {
1075 char name
[BDEVNAME_SIZE
] = "";
1078 disk_name(bdev
->bd_disk
, 0, name
);
1079 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1080 "resized disk %s\n", name
);
1085 if (disk_part_scan_enabled(bdev
->bd_disk
))
1086 bdev
->bd_invalidated
= 1;
1090 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1091 * @disk: struct gendisk to check
1092 * @bdev: struct bdev to adjust.
1094 * This routine checks to see if the bdev size does not match the disk size
1095 * and adjusts it if it differs.
1097 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1099 loff_t disk_size
, bdev_size
;
1101 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1102 bdev_size
= i_size_read(bdev
->bd_inode
);
1103 if (disk_size
!= bdev_size
) {
1104 char name
[BDEVNAME_SIZE
];
1106 disk_name(disk
, 0, name
);
1108 "%s: detected capacity change from %lld to %lld\n",
1109 name
, bdev_size
, disk_size
);
1110 i_size_write(bdev
->bd_inode
, disk_size
);
1111 flush_disk(bdev
, false);
1114 EXPORT_SYMBOL(check_disk_size_change
);
1117 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1118 * @disk: struct gendisk to be revalidated
1120 * This routine is a wrapper for lower-level driver's revalidate_disk
1121 * call-backs. It is used to do common pre and post operations needed
1122 * for all revalidate_disk operations.
1124 int revalidate_disk(struct gendisk
*disk
)
1126 struct block_device
*bdev
;
1129 if (disk
->fops
->revalidate_disk
)
1130 ret
= disk
->fops
->revalidate_disk(disk
);
1131 bdev
= bdget_disk(disk
, 0);
1135 mutex_lock(&bdev
->bd_mutex
);
1136 check_disk_size_change(disk
, bdev
);
1137 bdev
->bd_invalidated
= 0;
1138 mutex_unlock(&bdev
->bd_mutex
);
1142 EXPORT_SYMBOL(revalidate_disk
);
1145 * This routine checks whether a removable media has been changed,
1146 * and invalidates all buffer-cache-entries in that case. This
1147 * is a relatively slow routine, so we have to try to minimize using
1148 * it. Thus it is called only upon a 'mount' or 'open'. This
1149 * is the best way of combining speed and utility, I think.
1150 * People changing diskettes in the middle of an operation deserve
1153 int check_disk_change(struct block_device
*bdev
)
1155 struct gendisk
*disk
= bdev
->bd_disk
;
1156 const struct block_device_operations
*bdops
= disk
->fops
;
1157 unsigned int events
;
1159 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1160 DISK_EVENT_EJECT_REQUEST
);
1161 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1164 flush_disk(bdev
, true);
1165 if (bdops
->revalidate_disk
)
1166 bdops
->revalidate_disk(bdev
->bd_disk
);
1170 EXPORT_SYMBOL(check_disk_change
);
1172 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1174 unsigned bsize
= bdev_logical_block_size(bdev
);
1176 mutex_lock(&bdev
->bd_inode
->i_mutex
);
1177 i_size_write(bdev
->bd_inode
, size
);
1178 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
1179 while (bsize
< PAGE_CACHE_SIZE
) {
1184 bdev
->bd_block_size
= bsize
;
1185 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1187 EXPORT_SYMBOL(bd_set_size
);
1189 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1194 * mutex_lock(part->bd_mutex)
1195 * mutex_lock_nested(whole->bd_mutex, 1)
1198 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1200 struct gendisk
*disk
;
1201 struct module
*owner
;
1206 if (mode
& FMODE_READ
)
1208 if (mode
& FMODE_WRITE
)
1211 * hooks: /n/, see "layering violations".
1214 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1224 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1227 owner
= disk
->fops
->owner
;
1229 disk_block_events(disk
);
1230 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1231 if (!bdev
->bd_openers
) {
1232 bdev
->bd_disk
= disk
;
1233 bdev
->bd_queue
= disk
->queue
;
1234 bdev
->bd_contains
= bdev
;
1235 bdev
->bd_inode
->i_flags
= disk
->fops
->direct_access
? S_DAX
: 0;
1239 bdev
->bd_part
= disk_get_part(disk
, partno
);
1244 if (disk
->fops
->open
) {
1245 ret
= disk
->fops
->open(bdev
, mode
);
1246 if (ret
== -ERESTARTSYS
) {
1247 /* Lost a race with 'disk' being
1248 * deleted, try again.
1251 disk_put_part(bdev
->bd_part
);
1252 bdev
->bd_part
= NULL
;
1253 bdev
->bd_disk
= NULL
;
1254 bdev
->bd_queue
= NULL
;
1255 mutex_unlock(&bdev
->bd_mutex
);
1256 disk_unblock_events(disk
);
1264 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1267 * If the device is invalidated, rescan partition
1268 * if open succeeded or failed with -ENOMEDIUM.
1269 * The latter is necessary to prevent ghost
1270 * partitions on a removed medium.
1272 if (bdev
->bd_invalidated
) {
1274 rescan_partitions(disk
, bdev
);
1275 else if (ret
== -ENOMEDIUM
)
1276 invalidate_partitions(disk
, bdev
);
1281 struct block_device
*whole
;
1282 whole
= bdget_disk(disk
, 0);
1287 ret
= __blkdev_get(whole
, mode
, 1);
1290 bdev
->bd_contains
= whole
;
1291 bdev
->bd_part
= disk_get_part(disk
, partno
);
1292 if (!(disk
->flags
& GENHD_FL_UP
) ||
1293 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1297 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1299 * If the partition is not aligned on a page
1300 * boundary, we can't do dax I/O to it.
1302 if ((bdev
->bd_part
->start_sect
% (PAGE_SIZE
/ 512)) ||
1303 (bdev
->bd_part
->nr_sects
% (PAGE_SIZE
/ 512)))
1304 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1307 if (bdev
->bd_bdi
== &noop_backing_dev_info
)
1308 bdev
->bd_bdi
= bdi_get(disk
->queue
->backing_dev_info
);
1310 if (bdev
->bd_contains
== bdev
) {
1312 if (bdev
->bd_disk
->fops
->open
)
1313 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1314 /* the same as first opener case, read comment there */
1315 if (bdev
->bd_invalidated
) {
1317 rescan_partitions(bdev
->bd_disk
, bdev
);
1318 else if (ret
== -ENOMEDIUM
)
1319 invalidate_partitions(bdev
->bd_disk
, bdev
);
1322 goto out_unlock_bdev
;
1324 /* only one opener holds refs to the module and disk */
1330 bdev
->bd_part_count
++;
1331 mutex_unlock(&bdev
->bd_mutex
);
1332 disk_unblock_events(disk
);
1336 disk_put_part(bdev
->bd_part
);
1337 bdev
->bd_disk
= NULL
;
1338 bdev
->bd_part
= NULL
;
1339 bdev
->bd_queue
= NULL
;
1340 if (bdev
!= bdev
->bd_contains
)
1341 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1342 bdev
->bd_contains
= NULL
;
1344 mutex_unlock(&bdev
->bd_mutex
);
1345 disk_unblock_events(disk
);
1355 * blkdev_get - open a block device
1356 * @bdev: block_device to open
1357 * @mode: FMODE_* mask
1358 * @holder: exclusive holder identifier
1360 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1361 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1362 * @holder is invalid. Exclusive opens may nest for the same @holder.
1364 * On success, the reference count of @bdev is unchanged. On failure,
1371 * 0 on success, -errno on failure.
1373 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1375 struct block_device
*whole
= NULL
;
1378 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1380 if ((mode
& FMODE_EXCL
) && holder
) {
1381 whole
= bd_start_claiming(bdev
, holder
);
1382 if (IS_ERR(whole
)) {
1384 return PTR_ERR(whole
);
1388 res
= __blkdev_get(bdev
, mode
, 0);
1391 struct gendisk
*disk
= whole
->bd_disk
;
1393 /* finish claiming */
1394 mutex_lock(&bdev
->bd_mutex
);
1395 spin_lock(&bdev_lock
);
1398 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1400 * Note that for a whole device bd_holders
1401 * will be incremented twice, and bd_holder
1402 * will be set to bd_may_claim before being
1405 whole
->bd_holders
++;
1406 whole
->bd_holder
= bd_may_claim
;
1408 bdev
->bd_holder
= holder
;
1411 /* tell others that we're done */
1412 BUG_ON(whole
->bd_claiming
!= holder
);
1413 whole
->bd_claiming
= NULL
;
1414 wake_up_bit(&whole
->bd_claiming
, 0);
1416 spin_unlock(&bdev_lock
);
1419 * Block event polling for write claims if requested. Any
1420 * write holder makes the write_holder state stick until
1421 * all are released. This is good enough and tracking
1422 * individual writeable reference is too fragile given the
1423 * way @mode is used in blkdev_get/put().
1425 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1426 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1427 bdev
->bd_write_holder
= true;
1428 disk_block_events(disk
);
1431 mutex_unlock(&bdev
->bd_mutex
);
1437 EXPORT_SYMBOL(blkdev_get
);
1440 * blkdev_get_by_path - open a block device by name
1441 * @path: path to the block device to open
1442 * @mode: FMODE_* mask
1443 * @holder: exclusive holder identifier
1445 * Open the blockdevice described by the device file at @path. @mode
1446 * and @holder are identical to blkdev_get().
1448 * On success, the returned block_device has reference count of one.
1454 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1456 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1459 struct block_device
*bdev
;
1463 if (mode
& FMODE_READ
)
1465 if (mode
& FMODE_WRITE
)
1467 bdev
= lookup_bdev(path
, perm
);
1471 err
= blkdev_get(bdev
, mode
, holder
);
1473 return ERR_PTR(err
);
1475 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1476 blkdev_put(bdev
, mode
);
1477 return ERR_PTR(-EACCES
);
1482 EXPORT_SYMBOL(blkdev_get_by_path
);
1485 * blkdev_get_by_dev - open a block device by device number
1486 * @dev: device number of block device to open
1487 * @mode: FMODE_* mask
1488 * @holder: exclusive holder identifier
1490 * Open the blockdevice described by device number @dev. @mode and
1491 * @holder are identical to blkdev_get().
1493 * Use it ONLY if you really do not have anything better - i.e. when
1494 * you are behind a truly sucky interface and all you are given is a
1495 * device number. _Never_ to be used for internal purposes. If you
1496 * ever need it - reconsider your API.
1498 * On success, the returned block_device has reference count of one.
1504 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1506 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1508 struct block_device
*bdev
;
1513 return ERR_PTR(-ENOMEM
);
1515 err
= blkdev_get(bdev
, mode
, holder
);
1517 return ERR_PTR(err
);
1521 EXPORT_SYMBOL(blkdev_get_by_dev
);
1523 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1525 struct block_device
*bdev
;
1528 * Preserve backwards compatibility and allow large file access
1529 * even if userspace doesn't ask for it explicitly. Some mkfs
1530 * binary needs it. We might want to drop this workaround
1531 * during an unstable branch.
1533 filp
->f_flags
|= O_LARGEFILE
;
1535 if (filp
->f_flags
& O_NDELAY
)
1536 filp
->f_mode
|= FMODE_NDELAY
;
1537 if (filp
->f_flags
& O_EXCL
)
1538 filp
->f_mode
|= FMODE_EXCL
;
1539 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1540 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1542 bdev
= bd_acquire(inode
);
1547 * A negative i_writecount for bdev->bd_inode means that the bdev
1548 * or one of its paritions is mounted in a user namespace. Deny
1549 * writing for non-root in this case, otherwise an unprivileged
1550 * user can attack the kernel by modifying the backing store of a
1551 * mounted filesystem.
1553 if ((filp
->f_mode
& FMODE_WRITE
) &&
1554 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
) &&
1555 !atomic_inc_unless_negative(&bdev
->bd_inode
->i_writecount
)) {
1560 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1562 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1565 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1567 struct gendisk
*disk
= bdev
->bd_disk
;
1568 struct block_device
*victim
= NULL
;
1570 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1572 bdev
->bd_part_count
--;
1574 if (!--bdev
->bd_openers
) {
1575 WARN_ON_ONCE(bdev
->bd_holders
);
1576 sync_blockdev(bdev
);
1579 bdev_write_inode(bdev
);
1581 if (bdev
->bd_contains
== bdev
) {
1582 if (disk
->fops
->release
)
1583 disk
->fops
->release(disk
, mode
);
1585 if (!bdev
->bd_openers
) {
1586 struct module
*owner
= disk
->fops
->owner
;
1588 disk_put_part(bdev
->bd_part
);
1589 bdev
->bd_part
= NULL
;
1590 bdev
->bd_disk
= NULL
;
1591 if (bdev
!= bdev
->bd_contains
)
1592 victim
= bdev
->bd_contains
;
1593 bdev
->bd_contains
= NULL
;
1598 mutex_unlock(&bdev
->bd_mutex
);
1601 __blkdev_put(victim
, mode
, 1);
1604 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1606 mutex_lock(&bdev
->bd_mutex
);
1608 if (mode
& FMODE_EXCL
) {
1612 * Release a claim on the device. The holder fields
1613 * are protected with bdev_lock. bd_mutex is to
1614 * synchronize disk_holder unlinking.
1616 spin_lock(&bdev_lock
);
1618 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1619 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1621 /* bd_contains might point to self, check in a separate step */
1622 if ((bdev_free
= !bdev
->bd_holders
))
1623 bdev
->bd_holder
= NULL
;
1624 if (!bdev
->bd_contains
->bd_holders
)
1625 bdev
->bd_contains
->bd_holder
= NULL
;
1627 spin_unlock(&bdev_lock
);
1630 * If this was the last claim, remove holder link and
1631 * unblock evpoll if it was a write holder.
1633 if (bdev_free
&& bdev
->bd_write_holder
) {
1634 disk_unblock_events(bdev
->bd_disk
);
1635 bdev
->bd_write_holder
= false;
1640 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1641 * event. This is to ensure detection of media removal commanded
1642 * from userland - e.g. eject(1).
1644 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1646 mutex_unlock(&bdev
->bd_mutex
);
1648 __blkdev_put(bdev
, mode
, 0);
1650 EXPORT_SYMBOL(blkdev_put
);
1652 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1654 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1655 if (filp
->f_mode
& FMODE_WRITE
&&
1656 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
))
1657 atomic_dec(&bdev
->bd_inode
->i_writecount
);
1658 blkdev_put(bdev
, filp
->f_mode
);
1662 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1664 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1665 fmode_t mode
= file
->f_mode
;
1668 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1669 * to updated it before every ioctl.
1671 if (file
->f_flags
& O_NDELAY
)
1672 mode
|= FMODE_NDELAY
;
1674 mode
&= ~FMODE_NDELAY
;
1676 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1680 * Write data to the block device. Only intended for the block device itself
1681 * and the raw driver which basically is a fake block device.
1683 * Does not take i_mutex for the write and thus is not for general purpose
1686 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1688 struct file
*file
= iocb
->ki_filp
;
1689 struct inode
*bd_inode
= file
->f_mapping
->host
;
1690 loff_t size
= i_size_read(bd_inode
);
1691 struct blk_plug plug
;
1694 if (bdev_read_only(I_BDEV(bd_inode
)))
1697 if (!iov_iter_count(from
))
1700 if (iocb
->ki_pos
>= size
)
1703 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1705 blk_start_plug(&plug
);
1706 ret
= __generic_file_write_iter(iocb
, from
);
1709 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1713 blk_finish_plug(&plug
);
1716 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1718 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1720 struct file
*file
= iocb
->ki_filp
;
1721 struct inode
*bd_inode
= file
->f_mapping
->host
;
1722 loff_t size
= i_size_read(bd_inode
);
1723 loff_t pos
= iocb
->ki_pos
;
1729 iov_iter_truncate(to
, size
);
1730 return generic_file_read_iter(iocb
, to
);
1732 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1735 * Try to release a page associated with block device when the system
1736 * is under memory pressure.
1738 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1740 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1742 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1743 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1745 return try_to_free_buffers(page
);
1748 static const struct address_space_operations def_blk_aops
= {
1749 .readpage
= blkdev_readpage
,
1750 .readpages
= blkdev_readpages
,
1751 .writepage
= blkdev_writepage
,
1752 .write_begin
= blkdev_write_begin
,
1753 .write_end
= blkdev_write_end
,
1754 .writepages
= generic_writepages
,
1755 .releasepage
= blkdev_releasepage
,
1756 .direct_IO
= blkdev_direct_IO
,
1757 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1760 const struct file_operations def_blk_fops
= {
1761 .open
= blkdev_open
,
1762 .release
= blkdev_close
,
1763 .llseek
= block_llseek
,
1764 .read_iter
= blkdev_read_iter
,
1765 .write_iter
= blkdev_write_iter
,
1766 .mmap
= generic_file_mmap
,
1767 .fsync
= blkdev_fsync
,
1768 .unlocked_ioctl
= block_ioctl
,
1769 #ifdef CONFIG_COMPAT
1770 .compat_ioctl
= compat_blkdev_ioctl
,
1772 .splice_read
= generic_file_splice_read
,
1773 .splice_write
= iter_file_splice_write
,
1776 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1779 mm_segment_t old_fs
= get_fs();
1781 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1786 EXPORT_SYMBOL(ioctl_by_bdev
);
1789 * lookup_bdev - lookup a struct block_device by name
1790 * @pathname: special file representing the block device
1791 * @mask: rights to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
1793 * Get a reference to the blockdevice at @pathname in the current
1794 * namespace if possible and return it. Return ERR_PTR(error)
1795 * otherwise. If @mask is non-zero, check for access rights to the
1796 * inode at @pathname.
1798 struct block_device
*lookup_bdev(const char *pathname
, int mask
)
1800 struct block_device
*bdev
;
1801 struct inode
*inode
;
1805 if (!pathname
|| !*pathname
)
1806 return ERR_PTR(-EINVAL
);
1808 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1810 return ERR_PTR(error
);
1812 inode
= d_backing_inode(path
.dentry
);
1813 if (mask
!= 0 && !capable(CAP_SYS_ADMIN
)) {
1814 error
= __inode_permission(inode
, mask
);
1819 if (!S_ISBLK(inode
->i_mode
))
1822 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1825 bdev
= bd_acquire(inode
);
1832 bdev
= ERR_PTR(error
);
1835 EXPORT_SYMBOL(lookup_bdev
);
1837 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1839 struct super_block
*sb
= get_super(bdev
);
1844 * no need to lock the super, get_super holds the
1845 * read mutex so the filesystem cannot go away
1846 * under us (->put_super runs with the write lock
1849 shrink_dcache_sb(sb
);
1850 res
= invalidate_inodes(sb
, kill_dirty
);
1853 invalidate_bdev(bdev
);
1856 EXPORT_SYMBOL(__invalidate_device
);
1858 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1860 struct inode
*inode
, *old_inode
= NULL
;
1862 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1863 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1864 struct address_space
*mapping
= inode
->i_mapping
;
1865 struct block_device
*bdev
;
1867 spin_lock(&inode
->i_lock
);
1868 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1869 mapping
->nrpages
== 0) {
1870 spin_unlock(&inode
->i_lock
);
1874 spin_unlock(&inode
->i_lock
);
1875 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1877 * We hold a reference to 'inode' so it couldn't have been
1878 * removed from s_inodes list while we dropped the
1879 * s_inode_list_lock We cannot iput the inode now as we can
1880 * be holding the last reference and we cannot iput it under
1881 * s_inode_list_lock. So we keep the reference and iput it
1886 bdev
= I_BDEV(inode
);
1888 mutex_lock(&bdev
->bd_mutex
);
1889 if (bdev
->bd_openers
)
1891 mutex_unlock(&bdev
->bd_mutex
);
1893 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1895 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);