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
);
159 static struct inode
*bdev_file_inode(struct file
*file
)
161 return file
->f_mapping
->host
;
165 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
167 struct file
*file
= iocb
->ki_filp
;
168 struct inode
*inode
= bdev_file_inode(file
);
171 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
172 NULL
, DIO_SKIP_DIO_COUNT
);
173 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
174 blkdev_get_block
, NULL
, NULL
,
178 int __sync_blockdev(struct block_device
*bdev
, int wait
)
183 return filemap_flush(bdev
->bd_inode
->i_mapping
);
184 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
188 * Write out and wait upon all the dirty data associated with a block
189 * device via its mapping. Does not take the superblock lock.
191 int sync_blockdev(struct block_device
*bdev
)
193 return __sync_blockdev(bdev
, 1);
195 EXPORT_SYMBOL(sync_blockdev
);
198 * Write out and wait upon all dirty data associated with this
199 * device. Filesystem data as well as the underlying block
200 * device. Takes the superblock lock.
202 int fsync_bdev(struct block_device
*bdev
)
204 struct super_block
*sb
= get_super(bdev
);
206 int res
= sync_filesystem(sb
);
210 return sync_blockdev(bdev
);
212 EXPORT_SYMBOL(fsync_bdev
);
215 * freeze_bdev -- lock a filesystem and force it into a consistent state
216 * @bdev: blockdevice to lock
218 * If a superblock is found on this device, we take the s_umount semaphore
219 * on it to make sure nobody unmounts until the snapshot creation is done.
220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
226 struct super_block
*freeze_bdev(struct block_device
*bdev
)
228 struct super_block
*sb
;
231 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
232 if (++bdev
->bd_fsfreeze_count
> 1) {
234 * We don't even need to grab a reference - the first call
235 * to freeze_bdev grab an active reference and only the last
236 * thaw_bdev drops it.
238 sb
= get_super(bdev
);
240 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
244 sb
= get_active_super(bdev
);
247 if (sb
->s_op
->freeze_super
)
248 error
= sb
->s_op
->freeze_super(sb
);
250 error
= freeze_super(sb
);
252 deactivate_super(sb
);
253 bdev
->bd_fsfreeze_count
--;
254 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
255 return ERR_PTR(error
);
257 deactivate_super(sb
);
260 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
261 return sb
; /* thaw_bdev releases s->s_umount */
263 EXPORT_SYMBOL(freeze_bdev
);
266 * thaw_bdev -- unlock filesystem
267 * @bdev: blockdevice to unlock
268 * @sb: associated superblock
270 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
272 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
276 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
277 if (!bdev
->bd_fsfreeze_count
)
281 if (--bdev
->bd_fsfreeze_count
> 0)
287 if (sb
->s_op
->thaw_super
)
288 error
= sb
->s_op
->thaw_super(sb
);
290 error
= thaw_super(sb
);
292 bdev
->bd_fsfreeze_count
++;
293 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
297 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
300 EXPORT_SYMBOL(thaw_bdev
);
302 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
304 return block_write_full_page(page
, blkdev_get_block
, wbc
);
307 static int blkdev_readpage(struct file
* file
, struct page
* page
)
309 return block_read_full_page(page
, blkdev_get_block
);
312 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
313 struct list_head
*pages
, unsigned nr_pages
)
315 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
318 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
319 loff_t pos
, unsigned len
, unsigned flags
,
320 struct page
**pagep
, void **fsdata
)
322 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
326 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
327 loff_t pos
, unsigned len
, unsigned copied
,
328 struct page
*page
, void *fsdata
)
331 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
334 page_cache_release(page
);
341 * for a block special file file_inode(file)->i_size is zero
342 * so we compute the size by hand (just as in block_read/write above)
344 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
346 struct inode
*bd_inode
= bdev_file_inode(file
);
349 mutex_lock(&bd_inode
->i_mutex
);
350 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
351 mutex_unlock(&bd_inode
->i_mutex
);
355 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
357 struct inode
*bd_inode
= bdev_file_inode(filp
);
358 struct block_device
*bdev
= I_BDEV(bd_inode
);
361 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
366 * There is no need to serialise calls to blkdev_issue_flush with
367 * i_mutex and doing so causes performance issues with concurrent
368 * O_SYNC writers to a block device.
370 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
371 if (error
== -EOPNOTSUPP
)
376 EXPORT_SYMBOL(blkdev_fsync
);
379 * bdev_read_page() - Start reading a page from a block device
380 * @bdev: The device to read the page from
381 * @sector: The offset on the device to read the page to (need not be aligned)
382 * @page: The page to read
384 * On entry, the page should be locked. It will be unlocked when the page
385 * has been read. If the block driver implements rw_page synchronously,
386 * that will be true on exit from this function, but it need not be.
388 * Errors returned by this function are usually "soft", eg out of memory, or
389 * queue full; callers should try a different route to read this page rather
390 * than propagate an error back up the stack.
392 * Return: negative errno if an error occurs, 0 if submission was successful.
394 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
397 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
398 int result
= -EOPNOTSUPP
;
400 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
403 result
= blk_queue_enter(bdev
->bd_queue
, GFP_KERNEL
);
406 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
407 blk_queue_exit(bdev
->bd_queue
);
410 EXPORT_SYMBOL_GPL(bdev_read_page
);
413 * bdev_write_page() - Start writing a page to a block device
414 * @bdev: The device to write the page to
415 * @sector: The offset on the device to write the page to (need not be aligned)
416 * @page: The page to write
417 * @wbc: The writeback_control for the write
419 * On entry, the page should be locked and not currently under writeback.
420 * On exit, if the write started successfully, the page will be unlocked and
421 * under writeback. If the write failed already (eg the driver failed to
422 * queue the page to the device), the page will still be locked. If the
423 * caller is a ->writepage implementation, it will need to unlock the page.
425 * Errors returned by this function are usually "soft", eg out of memory, or
426 * queue full; callers should try a different route to write this page rather
427 * than propagate an error back up the stack.
429 * Return: negative errno if an error occurs, 0 if submission was successful.
431 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
432 struct page
*page
, struct writeback_control
*wbc
)
435 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
436 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
438 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
440 result
= blk_queue_enter(bdev
->bd_queue
, GFP_NOIO
);
444 set_page_writeback(page
);
445 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
447 end_page_writeback(page
);
450 blk_queue_exit(bdev
->bd_queue
);
453 EXPORT_SYMBOL_GPL(bdev_write_page
);
456 * bdev_direct_access() - Get the address for directly-accessibly memory
457 * @bdev: The device containing the memory
458 * @sector: The offset within the device
459 * @addr: Where to put the address of the memory
460 * @pfn: The Page Frame Number for the memory
461 * @size: The number of bytes requested
463 * If a block device is made up of directly addressable memory, this function
464 * will tell the caller the PFN and the address of the memory. The address
465 * may be directly dereferenced within the kernel without the need to call
466 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
469 * Return: negative errno if an error occurs, otherwise the number of bytes
470 * accessible at this address.
472 long bdev_direct_access(struct block_device
*bdev
, sector_t sector
,
473 void __pmem
**addr
, unsigned long *pfn
, long size
)
476 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
479 * The device driver is allowed to sleep, in order to make the
480 * memory directly accessible.
486 if (!ops
->direct_access
)
488 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
489 part_nr_sects_read(bdev
->bd_part
))
491 sector
+= get_start_sect(bdev
);
492 if (sector
% (PAGE_SIZE
/ 512))
494 avail
= ops
->direct_access(bdev
, sector
, addr
, pfn
);
497 if (avail
> 0 && avail
& ~PAGE_MASK
)
499 return min(avail
, size
);
501 EXPORT_SYMBOL_GPL(bdev_direct_access
);
507 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
508 static struct kmem_cache
* bdev_cachep __read_mostly
;
510 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
512 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
515 return &ei
->vfs_inode
;
518 static void bdev_i_callback(struct rcu_head
*head
)
520 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
521 struct bdev_inode
*bdi
= BDEV_I(inode
);
523 kmem_cache_free(bdev_cachep
, bdi
);
526 static void bdev_destroy_inode(struct inode
*inode
)
528 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
531 static void init_once(void *foo
)
533 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
534 struct block_device
*bdev
= &ei
->bdev
;
536 memset(bdev
, 0, sizeof(*bdev
));
537 mutex_init(&bdev
->bd_mutex
);
538 INIT_LIST_HEAD(&bdev
->bd_inodes
);
539 INIT_LIST_HEAD(&bdev
->bd_list
);
541 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
543 inode_init_once(&ei
->vfs_inode
);
544 /* Initialize mutex for freeze. */
545 mutex_init(&bdev
->bd_fsfreeze_mutex
);
548 static inline void __bd_forget(struct inode
*inode
)
550 list_del_init(&inode
->i_devices
);
551 inode
->i_bdev
= NULL
;
552 inode
->i_mapping
= &inode
->i_data
;
555 static void bdev_evict_inode(struct inode
*inode
)
557 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
559 truncate_inode_pages_final(&inode
->i_data
);
560 invalidate_inode_buffers(inode
); /* is it needed here? */
562 spin_lock(&bdev_lock
);
563 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
564 __bd_forget(list_entry(p
, struct inode
, i_devices
));
566 list_del_init(&bdev
->bd_list
);
567 spin_unlock(&bdev_lock
);
570 static const struct super_operations bdev_sops
= {
571 .statfs
= simple_statfs
,
572 .alloc_inode
= bdev_alloc_inode
,
573 .destroy_inode
= bdev_destroy_inode
,
574 .drop_inode
= generic_delete_inode
,
575 .evict_inode
= bdev_evict_inode
,
578 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
579 int flags
, const char *dev_name
, void *data
)
581 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
584 static struct file_system_type bd_type
= {
587 .kill_sb
= kill_anon_super
,
590 struct super_block
*blockdev_superblock __read_mostly
;
591 EXPORT_SYMBOL_GPL(blockdev_superblock
);
593 void __init
bdev_cache_init(void)
596 static struct vfsmount
*bd_mnt
;
598 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
599 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
600 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
602 err
= register_filesystem(&bd_type
);
604 panic("Cannot register bdev pseudo-fs");
605 bd_mnt
= kern_mount(&bd_type
);
607 panic("Cannot create bdev pseudo-fs");
608 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
612 * Most likely _very_ bad one - but then it's hardly critical for small
613 * /dev and can be fixed when somebody will need really large one.
614 * Keep in mind that it will be fed through icache hash function too.
616 static inline unsigned long hash(dev_t dev
)
618 return MAJOR(dev
)+MINOR(dev
);
621 static int bdev_test(struct inode
*inode
, void *data
)
623 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
626 static int bdev_set(struct inode
*inode
, void *data
)
628 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
632 static LIST_HEAD(all_bdevs
);
634 struct block_device
*bdget(dev_t dev
)
636 struct block_device
*bdev
;
639 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
640 bdev_test
, bdev_set
, &dev
);
645 bdev
= &BDEV_I(inode
)->bdev
;
647 if (inode
->i_state
& I_NEW
) {
648 bdev
->bd_contains
= NULL
;
649 bdev
->bd_super
= NULL
;
650 bdev
->bd_inode
= inode
;
651 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
652 bdev
->bd_part_count
= 0;
653 bdev
->bd_invalidated
= 0;
654 inode
->i_mode
= S_IFBLK
;
656 inode
->i_bdev
= bdev
;
657 inode
->i_data
.a_ops
= &def_blk_aops
;
658 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
659 spin_lock(&bdev_lock
);
660 list_add(&bdev
->bd_list
, &all_bdevs
);
661 spin_unlock(&bdev_lock
);
662 unlock_new_inode(inode
);
667 EXPORT_SYMBOL(bdget
);
670 * bdgrab -- Grab a reference to an already referenced block device
671 * @bdev: Block device to grab a reference to.
673 struct block_device
*bdgrab(struct block_device
*bdev
)
675 ihold(bdev
->bd_inode
);
678 EXPORT_SYMBOL(bdgrab
);
680 long nr_blockdev_pages(void)
682 struct block_device
*bdev
;
684 spin_lock(&bdev_lock
);
685 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
686 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
688 spin_unlock(&bdev_lock
);
692 void bdput(struct block_device
*bdev
)
694 iput(bdev
->bd_inode
);
697 EXPORT_SYMBOL(bdput
);
699 static struct block_device
*bd_acquire(struct inode
*inode
)
701 struct block_device
*bdev
;
703 spin_lock(&bdev_lock
);
704 bdev
= inode
->i_bdev
;
706 ihold(bdev
->bd_inode
);
707 spin_unlock(&bdev_lock
);
710 spin_unlock(&bdev_lock
);
712 bdev
= bdget(inode
->i_rdev
);
714 spin_lock(&bdev_lock
);
715 if (!inode
->i_bdev
) {
717 * We take an additional reference to bd_inode,
718 * and it's released in clear_inode() of inode.
719 * So, we can access it via ->i_mapping always
722 ihold(bdev
->bd_inode
);
723 inode
->i_bdev
= bdev
;
724 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
725 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
727 spin_unlock(&bdev_lock
);
732 /* Call when you free inode */
734 void bd_forget(struct inode
*inode
)
736 struct block_device
*bdev
= NULL
;
738 spin_lock(&bdev_lock
);
739 if (!sb_is_blkdev_sb(inode
->i_sb
))
740 bdev
= inode
->i_bdev
;
742 spin_unlock(&bdev_lock
);
745 iput(bdev
->bd_inode
);
749 * bd_may_claim - test whether a block device can be claimed
750 * @bdev: block device of interest
751 * @whole: whole block device containing @bdev, may equal @bdev
752 * @holder: holder trying to claim @bdev
754 * Test whether @bdev can be claimed by @holder.
757 * spin_lock(&bdev_lock).
760 * %true if @bdev can be claimed, %false otherwise.
762 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
765 if (bdev
->bd_holder
== holder
)
766 return true; /* already a holder */
767 else if (bdev
->bd_holder
!= NULL
)
768 return false; /* held by someone else */
769 else if (bdev
->bd_contains
== bdev
)
770 return true; /* is a whole device which isn't held */
772 else if (whole
->bd_holder
== bd_may_claim
)
773 return true; /* is a partition of a device that is being partitioned */
774 else if (whole
->bd_holder
!= NULL
)
775 return false; /* is a partition of a held device */
777 return true; /* is a partition of an un-held device */
781 * bd_prepare_to_claim - prepare to claim a block device
782 * @bdev: block device of interest
783 * @whole: the whole device containing @bdev, may equal @bdev
784 * @holder: holder trying to claim @bdev
786 * Prepare to claim @bdev. This function fails if @bdev is already
787 * claimed by another holder and waits if another claiming is in
788 * progress. This function doesn't actually claim. On successful
789 * return, the caller has ownership of bd_claiming and bd_holder[s].
792 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
796 * 0 if @bdev can be claimed, -EBUSY otherwise.
798 static int bd_prepare_to_claim(struct block_device
*bdev
,
799 struct block_device
*whole
, void *holder
)
802 /* if someone else claimed, fail */
803 if (!bd_may_claim(bdev
, whole
, holder
))
806 /* if claiming is already in progress, wait for it to finish */
807 if (whole
->bd_claiming
) {
808 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
811 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
812 spin_unlock(&bdev_lock
);
814 finish_wait(wq
, &wait
);
815 spin_lock(&bdev_lock
);
824 * bd_start_claiming - start claiming a block device
825 * @bdev: block device of interest
826 * @holder: holder trying to claim @bdev
828 * @bdev is about to be opened exclusively. Check @bdev can be opened
829 * exclusively and mark that an exclusive open is in progress. Each
830 * successful call to this function must be matched with a call to
831 * either bd_finish_claiming() or bd_abort_claiming() (which do not
834 * This function is used to gain exclusive access to the block device
835 * without actually causing other exclusive open attempts to fail. It
836 * should be used when the open sequence itself requires exclusive
837 * access but may subsequently fail.
843 * Pointer to the block device containing @bdev on success, ERR_PTR()
846 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
849 struct gendisk
*disk
;
850 struct block_device
*whole
;
856 * @bdev might not have been initialized properly yet, look up
857 * and grab the outer block device the hard way.
859 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
861 return ERR_PTR(-ENXIO
);
864 * Normally, @bdev should equal what's returned from bdget_disk()
865 * if partno is 0; however, some drivers (floppy) use multiple
866 * bdev's for the same physical device and @bdev may be one of the
867 * aliases. Keep @bdev if partno is 0. This means claimer
868 * tracking is broken for those devices but it has always been that
872 whole
= bdget_disk(disk
, 0);
874 whole
= bdgrab(bdev
);
876 module_put(disk
->fops
->owner
);
879 return ERR_PTR(-ENOMEM
);
881 /* prepare to claim, if successful, mark claiming in progress */
882 spin_lock(&bdev_lock
);
884 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
886 whole
->bd_claiming
= holder
;
887 spin_unlock(&bdev_lock
);
890 spin_unlock(&bdev_lock
);
897 struct bd_holder_disk
{
898 struct list_head list
;
899 struct gendisk
*disk
;
903 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
904 struct gendisk
*disk
)
906 struct bd_holder_disk
*holder
;
908 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
909 if (holder
->disk
== disk
)
914 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
916 return sysfs_create_link(from
, to
, kobject_name(to
));
919 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
921 sysfs_remove_link(from
, kobject_name(to
));
925 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
926 * @bdev: the claimed slave bdev
927 * @disk: the holding disk
929 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
931 * This functions creates the following sysfs symlinks.
933 * - from "slaves" directory of the holder @disk to the claimed @bdev
934 * - from "holders" directory of the @bdev to the holder @disk
936 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
937 * passed to bd_link_disk_holder(), then:
939 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
940 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
942 * The caller must have claimed @bdev before calling this function and
943 * ensure that both @bdev and @disk are valid during the creation and
944 * lifetime of these symlinks.
950 * 0 on success, -errno on failure.
952 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
954 struct bd_holder_disk
*holder
;
957 mutex_lock(&bdev
->bd_mutex
);
959 WARN_ON_ONCE(!bdev
->bd_holder
);
961 /* FIXME: remove the following once add_disk() handles errors */
962 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
965 holder
= bd_find_holder_disk(bdev
, disk
);
971 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
977 INIT_LIST_HEAD(&holder
->list
);
981 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
985 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
989 * bdev could be deleted beneath us which would implicitly destroy
990 * the holder directory. Hold on to it.
992 kobject_get(bdev
->bd_part
->holder_dir
);
994 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
998 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1002 mutex_unlock(&bdev
->bd_mutex
);
1005 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1008 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1009 * @bdev: the calimed slave bdev
1010 * @disk: the holding disk
1012 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1017 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1019 struct bd_holder_disk
*holder
;
1021 mutex_lock(&bdev
->bd_mutex
);
1023 holder
= bd_find_holder_disk(bdev
, disk
);
1025 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1026 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1027 del_symlink(bdev
->bd_part
->holder_dir
,
1028 &disk_to_dev(disk
)->kobj
);
1029 kobject_put(bdev
->bd_part
->holder_dir
);
1030 list_del_init(&holder
->list
);
1034 mutex_unlock(&bdev
->bd_mutex
);
1036 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1040 * flush_disk - invalidates all buffer-cache entries on a disk
1042 * @bdev: struct block device to be flushed
1043 * @kill_dirty: flag to guide handling of dirty inodes
1045 * Invalidates all buffer-cache entries on a disk. It should be called
1046 * when a disk has been changed -- either by a media change or online
1049 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1051 if (__invalidate_device(bdev
, kill_dirty
)) {
1052 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1053 "resized disk %s\n",
1054 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1059 if (disk_part_scan_enabled(bdev
->bd_disk
))
1060 bdev
->bd_invalidated
= 1;
1064 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1065 * @disk: struct gendisk to check
1066 * @bdev: struct bdev to adjust.
1068 * This routine checks to see if the bdev size does not match the disk size
1069 * and adjusts it if it differs.
1071 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1073 loff_t disk_size
, bdev_size
;
1075 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1076 bdev_size
= i_size_read(bdev
->bd_inode
);
1077 if (disk_size
!= bdev_size
) {
1079 "%s: detected capacity change from %lld to %lld\n",
1080 disk
->disk_name
, bdev_size
, disk_size
);
1081 i_size_write(bdev
->bd_inode
, disk_size
);
1082 flush_disk(bdev
, false);
1085 EXPORT_SYMBOL(check_disk_size_change
);
1088 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1089 * @disk: struct gendisk to be revalidated
1091 * This routine is a wrapper for lower-level driver's revalidate_disk
1092 * call-backs. It is used to do common pre and post operations needed
1093 * for all revalidate_disk operations.
1095 int revalidate_disk(struct gendisk
*disk
)
1097 struct block_device
*bdev
;
1100 if (disk
->fops
->revalidate_disk
)
1101 ret
= disk
->fops
->revalidate_disk(disk
);
1102 blk_integrity_revalidate(disk
);
1103 bdev
= bdget_disk(disk
, 0);
1107 mutex_lock(&bdev
->bd_mutex
);
1108 check_disk_size_change(disk
, bdev
);
1109 bdev
->bd_invalidated
= 0;
1110 mutex_unlock(&bdev
->bd_mutex
);
1114 EXPORT_SYMBOL(revalidate_disk
);
1117 * This routine checks whether a removable media has been changed,
1118 * and invalidates all buffer-cache-entries in that case. This
1119 * is a relatively slow routine, so we have to try to minimize using
1120 * it. Thus it is called only upon a 'mount' or 'open'. This
1121 * is the best way of combining speed and utility, I think.
1122 * People changing diskettes in the middle of an operation deserve
1125 int check_disk_change(struct block_device
*bdev
)
1127 struct gendisk
*disk
= bdev
->bd_disk
;
1128 const struct block_device_operations
*bdops
= disk
->fops
;
1129 unsigned int events
;
1131 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1132 DISK_EVENT_EJECT_REQUEST
);
1133 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1136 flush_disk(bdev
, true);
1137 if (bdops
->revalidate_disk
)
1138 bdops
->revalidate_disk(bdev
->bd_disk
);
1142 EXPORT_SYMBOL(check_disk_change
);
1144 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1146 unsigned bsize
= bdev_logical_block_size(bdev
);
1148 mutex_lock(&bdev
->bd_inode
->i_mutex
);
1149 i_size_write(bdev
->bd_inode
, size
);
1150 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
1151 while (bsize
< PAGE_CACHE_SIZE
) {
1156 bdev
->bd_block_size
= bsize
;
1157 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1159 EXPORT_SYMBOL(bd_set_size
);
1161 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1166 * mutex_lock(part->bd_mutex)
1167 * mutex_lock_nested(whole->bd_mutex, 1)
1170 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1172 struct gendisk
*disk
;
1173 struct module
*owner
;
1178 if (mode
& FMODE_READ
)
1180 if (mode
& FMODE_WRITE
)
1183 * hooks: /n/, see "layering violations".
1186 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1196 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1199 owner
= disk
->fops
->owner
;
1201 disk_block_events(disk
);
1202 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1203 if (!bdev
->bd_openers
) {
1204 bdev
->bd_disk
= disk
;
1205 bdev
->bd_queue
= disk
->queue
;
1206 bdev
->bd_contains
= bdev
;
1207 bdev
->bd_inode
->i_flags
= disk
->fops
->direct_access
? S_DAX
: 0;
1210 bdev
->bd_part
= disk_get_part(disk
, partno
);
1215 if (disk
->fops
->open
) {
1216 ret
= disk
->fops
->open(bdev
, mode
);
1217 if (ret
== -ERESTARTSYS
) {
1218 /* Lost a race with 'disk' being
1219 * deleted, try again.
1222 disk_put_part(bdev
->bd_part
);
1223 bdev
->bd_part
= NULL
;
1224 bdev
->bd_disk
= NULL
;
1225 bdev
->bd_queue
= NULL
;
1226 mutex_unlock(&bdev
->bd_mutex
);
1227 disk_unblock_events(disk
);
1235 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1236 if (!blkdev_dax_capable(bdev
))
1237 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1241 * If the device is invalidated, rescan partition
1242 * if open succeeded or failed with -ENOMEDIUM.
1243 * The latter is necessary to prevent ghost
1244 * partitions on a removed medium.
1246 if (bdev
->bd_invalidated
) {
1248 rescan_partitions(disk
, bdev
);
1249 else if (ret
== -ENOMEDIUM
)
1250 invalidate_partitions(disk
, bdev
);
1256 struct block_device
*whole
;
1257 whole
= bdget_disk(disk
, 0);
1262 ret
= __blkdev_get(whole
, mode
, 1);
1265 bdev
->bd_contains
= whole
;
1266 bdev
->bd_part
= disk_get_part(disk
, partno
);
1267 if (!(disk
->flags
& GENHD_FL_UP
) ||
1268 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1272 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1273 if (!blkdev_dax_capable(bdev
))
1274 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1277 if (bdev
->bd_contains
== bdev
) {
1279 if (bdev
->bd_disk
->fops
->open
)
1280 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1281 /* the same as first opener case, read comment there */
1282 if (bdev
->bd_invalidated
) {
1284 rescan_partitions(bdev
->bd_disk
, bdev
);
1285 else if (ret
== -ENOMEDIUM
)
1286 invalidate_partitions(bdev
->bd_disk
, bdev
);
1289 goto out_unlock_bdev
;
1291 /* only one opener holds refs to the module and disk */
1297 bdev
->bd_part_count
++;
1298 mutex_unlock(&bdev
->bd_mutex
);
1299 disk_unblock_events(disk
);
1303 disk_put_part(bdev
->bd_part
);
1304 bdev
->bd_disk
= NULL
;
1305 bdev
->bd_part
= NULL
;
1306 bdev
->bd_queue
= NULL
;
1307 if (bdev
!= bdev
->bd_contains
)
1308 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1309 bdev
->bd_contains
= NULL
;
1311 mutex_unlock(&bdev
->bd_mutex
);
1312 disk_unblock_events(disk
);
1322 * blkdev_get - open a block device
1323 * @bdev: block_device to open
1324 * @mode: FMODE_* mask
1325 * @holder: exclusive holder identifier
1327 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1328 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1329 * @holder is invalid. Exclusive opens may nest for the same @holder.
1331 * On success, the reference count of @bdev is unchanged. On failure,
1338 * 0 on success, -errno on failure.
1340 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1342 struct block_device
*whole
= NULL
;
1345 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1347 if ((mode
& FMODE_EXCL
) && holder
) {
1348 whole
= bd_start_claiming(bdev
, holder
);
1349 if (IS_ERR(whole
)) {
1351 return PTR_ERR(whole
);
1355 res
= __blkdev_get(bdev
, mode
, 0);
1358 struct gendisk
*disk
= whole
->bd_disk
;
1360 /* finish claiming */
1361 mutex_lock(&bdev
->bd_mutex
);
1362 spin_lock(&bdev_lock
);
1365 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1367 * Note that for a whole device bd_holders
1368 * will be incremented twice, and bd_holder
1369 * will be set to bd_may_claim before being
1372 whole
->bd_holders
++;
1373 whole
->bd_holder
= bd_may_claim
;
1375 bdev
->bd_holder
= holder
;
1378 /* tell others that we're done */
1379 BUG_ON(whole
->bd_claiming
!= holder
);
1380 whole
->bd_claiming
= NULL
;
1381 wake_up_bit(&whole
->bd_claiming
, 0);
1383 spin_unlock(&bdev_lock
);
1386 * Block event polling for write claims if requested. Any
1387 * write holder makes the write_holder state stick until
1388 * all are released. This is good enough and tracking
1389 * individual writeable reference is too fragile given the
1390 * way @mode is used in blkdev_get/put().
1392 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1393 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1394 bdev
->bd_write_holder
= true;
1395 disk_block_events(disk
);
1398 mutex_unlock(&bdev
->bd_mutex
);
1404 EXPORT_SYMBOL(blkdev_get
);
1407 * blkdev_get_by_path - open a block device by name
1408 * @path: path to the block device to open
1409 * @mode: FMODE_* mask
1410 * @holder: exclusive holder identifier
1412 * Open the blockdevice described by the device file at @path. @mode
1413 * and @holder are identical to blkdev_get().
1415 * On success, the returned block_device has reference count of one.
1421 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1423 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1426 struct block_device
*bdev
;
1429 bdev
= lookup_bdev(path
);
1433 err
= blkdev_get(bdev
, mode
, holder
);
1435 return ERR_PTR(err
);
1437 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1438 blkdev_put(bdev
, mode
);
1439 return ERR_PTR(-EACCES
);
1444 EXPORT_SYMBOL(blkdev_get_by_path
);
1447 * blkdev_get_by_dev - open a block device by device number
1448 * @dev: device number of block device to open
1449 * @mode: FMODE_* mask
1450 * @holder: exclusive holder identifier
1452 * Open the blockdevice described by device number @dev. @mode and
1453 * @holder are identical to blkdev_get().
1455 * Use it ONLY if you really do not have anything better - i.e. when
1456 * you are behind a truly sucky interface and all you are given is a
1457 * device number. _Never_ to be used for internal purposes. If you
1458 * ever need it - reconsider your API.
1460 * On success, the returned block_device has reference count of one.
1466 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1468 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1470 struct block_device
*bdev
;
1475 return ERR_PTR(-ENOMEM
);
1477 err
= blkdev_get(bdev
, mode
, holder
);
1479 return ERR_PTR(err
);
1483 EXPORT_SYMBOL(blkdev_get_by_dev
);
1485 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1487 struct block_device
*bdev
;
1490 * Preserve backwards compatibility and allow large file access
1491 * even if userspace doesn't ask for it explicitly. Some mkfs
1492 * binary needs it. We might want to drop this workaround
1493 * during an unstable branch.
1495 filp
->f_flags
|= O_LARGEFILE
;
1497 if (filp
->f_flags
& O_NDELAY
)
1498 filp
->f_mode
|= FMODE_NDELAY
;
1499 if (filp
->f_flags
& O_EXCL
)
1500 filp
->f_mode
|= FMODE_EXCL
;
1501 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1502 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1504 bdev
= bd_acquire(inode
);
1508 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1510 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1513 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1515 struct gendisk
*disk
= bdev
->bd_disk
;
1516 struct block_device
*victim
= NULL
;
1518 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1520 bdev
->bd_part_count
--;
1522 if (!--bdev
->bd_openers
) {
1523 WARN_ON_ONCE(bdev
->bd_holders
);
1524 sync_blockdev(bdev
);
1527 bdev_write_inode(bdev
);
1529 * Detaching bdev inode from its wb in __destroy_inode()
1530 * is too late: the queue which embeds its bdi (along with
1531 * root wb) can be gone as soon as we put_disk() below.
1533 inode_detach_wb(bdev
->bd_inode
);
1535 if (bdev
->bd_contains
== bdev
) {
1536 if (disk
->fops
->release
)
1537 disk
->fops
->release(disk
, mode
);
1539 if (!bdev
->bd_openers
) {
1540 struct module
*owner
= disk
->fops
->owner
;
1542 disk_put_part(bdev
->bd_part
);
1543 bdev
->bd_part
= NULL
;
1544 bdev
->bd_disk
= NULL
;
1545 if (bdev
!= bdev
->bd_contains
)
1546 victim
= bdev
->bd_contains
;
1547 bdev
->bd_contains
= NULL
;
1552 mutex_unlock(&bdev
->bd_mutex
);
1555 __blkdev_put(victim
, mode
, 1);
1558 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1560 mutex_lock(&bdev
->bd_mutex
);
1562 if (mode
& FMODE_EXCL
) {
1566 * Release a claim on the device. The holder fields
1567 * are protected with bdev_lock. bd_mutex is to
1568 * synchronize disk_holder unlinking.
1570 spin_lock(&bdev_lock
);
1572 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1573 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1575 /* bd_contains might point to self, check in a separate step */
1576 if ((bdev_free
= !bdev
->bd_holders
))
1577 bdev
->bd_holder
= NULL
;
1578 if (!bdev
->bd_contains
->bd_holders
)
1579 bdev
->bd_contains
->bd_holder
= NULL
;
1581 spin_unlock(&bdev_lock
);
1584 * If this was the last claim, remove holder link and
1585 * unblock evpoll if it was a write holder.
1587 if (bdev_free
&& bdev
->bd_write_holder
) {
1588 disk_unblock_events(bdev
->bd_disk
);
1589 bdev
->bd_write_holder
= false;
1594 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1595 * event. This is to ensure detection of media removal commanded
1596 * from userland - e.g. eject(1).
1598 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1600 mutex_unlock(&bdev
->bd_mutex
);
1602 __blkdev_put(bdev
, mode
, 0);
1604 EXPORT_SYMBOL(blkdev_put
);
1606 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1608 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1609 blkdev_put(bdev
, filp
->f_mode
);
1613 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1615 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1616 fmode_t mode
= file
->f_mode
;
1619 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1620 * to updated it before every ioctl.
1622 if (file
->f_flags
& O_NDELAY
)
1623 mode
|= FMODE_NDELAY
;
1625 mode
&= ~FMODE_NDELAY
;
1627 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1631 * Write data to the block device. Only intended for the block device itself
1632 * and the raw driver which basically is a fake block device.
1634 * Does not take i_mutex for the write and thus is not for general purpose
1637 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1639 struct file
*file
= iocb
->ki_filp
;
1640 struct inode
*bd_inode
= bdev_file_inode(file
);
1641 loff_t size
= i_size_read(bd_inode
);
1642 struct blk_plug plug
;
1645 if (bdev_read_only(I_BDEV(bd_inode
)))
1648 if (!iov_iter_count(from
))
1651 if (iocb
->ki_pos
>= size
)
1654 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1656 blk_start_plug(&plug
);
1657 ret
= __generic_file_write_iter(iocb
, from
);
1660 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1664 blk_finish_plug(&plug
);
1667 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1669 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1671 struct file
*file
= iocb
->ki_filp
;
1672 struct inode
*bd_inode
= bdev_file_inode(file
);
1673 loff_t size
= i_size_read(bd_inode
);
1674 loff_t pos
= iocb
->ki_pos
;
1680 iov_iter_truncate(to
, size
);
1681 return generic_file_read_iter(iocb
, to
);
1683 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1686 * Try to release a page associated with block device when the system
1687 * is under memory pressure.
1689 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1691 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1693 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1694 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1696 return try_to_free_buffers(page
);
1699 static const struct address_space_operations def_blk_aops
= {
1700 .readpage
= blkdev_readpage
,
1701 .readpages
= blkdev_readpages
,
1702 .writepage
= blkdev_writepage
,
1703 .write_begin
= blkdev_write_begin
,
1704 .write_end
= blkdev_write_end
,
1705 .writepages
= generic_writepages
,
1706 .releasepage
= blkdev_releasepage
,
1707 .direct_IO
= blkdev_direct_IO
,
1708 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1711 #ifdef CONFIG_FS_DAX
1713 * In the raw block case we do not need to contend with truncation nor
1714 * unwritten file extents. Without those concerns there is no need for
1715 * additional locking beyond the mmap_sem context that these routines
1716 * are already executing under.
1718 * Note, there is no protection if the block device is dynamically
1719 * resized (partition grow/shrink) during a fault. A stable block device
1720 * size is already not enforced in the blkdev_direct_IO path.
1722 * For DAX, it is the responsibility of the block device driver to
1723 * ensure the whole-disk device size is stable while requests are in
1726 * Finally, unlike the filemap_page_mkwrite() case there is no
1727 * filesystem superblock to sync against freezing. We still include a
1728 * pfn_mkwrite callback for dax drivers to receive write fault
1731 static int blkdev_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1733 return __dax_fault(vma
, vmf
, blkdev_get_block
, NULL
);
1736 static int blkdev_dax_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
1737 pmd_t
*pmd
, unsigned int flags
)
1739 return __dax_pmd_fault(vma
, addr
, pmd
, flags
, blkdev_get_block
, NULL
);
1742 static void blkdev_vm_open(struct vm_area_struct
*vma
)
1744 struct inode
*bd_inode
= bdev_file_inode(vma
->vm_file
);
1745 struct block_device
*bdev
= I_BDEV(bd_inode
);
1747 mutex_lock(&bd_inode
->i_mutex
);
1748 bdev
->bd_map_count
++;
1749 mutex_unlock(&bd_inode
->i_mutex
);
1752 static void blkdev_vm_close(struct vm_area_struct
*vma
)
1754 struct inode
*bd_inode
= bdev_file_inode(vma
->vm_file
);
1755 struct block_device
*bdev
= I_BDEV(bd_inode
);
1757 mutex_lock(&bd_inode
->i_mutex
);
1758 bdev
->bd_map_count
--;
1759 mutex_unlock(&bd_inode
->i_mutex
);
1762 static const struct vm_operations_struct blkdev_dax_vm_ops
= {
1763 .open
= blkdev_vm_open
,
1764 .close
= blkdev_vm_close
,
1765 .fault
= blkdev_dax_fault
,
1766 .pmd_fault
= blkdev_dax_pmd_fault
,
1767 .pfn_mkwrite
= blkdev_dax_fault
,
1770 static const struct vm_operations_struct blkdev_default_vm_ops
= {
1771 .open
= blkdev_vm_open
,
1772 .close
= blkdev_vm_close
,
1773 .fault
= filemap_fault
,
1774 .map_pages
= filemap_map_pages
,
1777 static int blkdev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1779 struct inode
*bd_inode
= bdev_file_inode(file
);
1780 struct block_device
*bdev
= I_BDEV(bd_inode
);
1782 file_accessed(file
);
1783 mutex_lock(&bd_inode
->i_mutex
);
1784 bdev
->bd_map_count
++;
1785 if (IS_DAX(bd_inode
)) {
1786 vma
->vm_ops
= &blkdev_dax_vm_ops
;
1787 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1789 vma
->vm_ops
= &blkdev_default_vm_ops
;
1791 mutex_unlock(&bd_inode
->i_mutex
);
1796 #define blkdev_mmap generic_file_mmap
1799 const struct file_operations def_blk_fops
= {
1800 .open
= blkdev_open
,
1801 .release
= blkdev_close
,
1802 .llseek
= block_llseek
,
1803 .read_iter
= blkdev_read_iter
,
1804 .write_iter
= blkdev_write_iter
,
1805 .mmap
= blkdev_mmap
,
1806 .fsync
= blkdev_fsync
,
1807 .unlocked_ioctl
= block_ioctl
,
1808 #ifdef CONFIG_COMPAT
1809 .compat_ioctl
= compat_blkdev_ioctl
,
1811 .splice_read
= generic_file_splice_read
,
1812 .splice_write
= iter_file_splice_write
,
1815 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1818 mm_segment_t old_fs
= get_fs();
1820 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1825 EXPORT_SYMBOL(ioctl_by_bdev
);
1828 * lookup_bdev - lookup a struct block_device by name
1829 * @pathname: special file representing the block device
1831 * Get a reference to the blockdevice at @pathname in the current
1832 * namespace if possible and return it. Return ERR_PTR(error)
1835 struct block_device
*lookup_bdev(const char *pathname
)
1837 struct block_device
*bdev
;
1838 struct inode
*inode
;
1842 if (!pathname
|| !*pathname
)
1843 return ERR_PTR(-EINVAL
);
1845 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1847 return ERR_PTR(error
);
1849 inode
= d_backing_inode(path
.dentry
);
1851 if (!S_ISBLK(inode
->i_mode
))
1854 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1857 bdev
= bd_acquire(inode
);
1864 bdev
= ERR_PTR(error
);
1867 EXPORT_SYMBOL(lookup_bdev
);
1869 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1871 struct super_block
*sb
= get_super(bdev
);
1876 * no need to lock the super, get_super holds the
1877 * read mutex so the filesystem cannot go away
1878 * under us (->put_super runs with the write lock
1881 shrink_dcache_sb(sb
);
1882 res
= invalidate_inodes(sb
, kill_dirty
);
1885 invalidate_bdev(bdev
);
1888 EXPORT_SYMBOL(__invalidate_device
);
1890 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1892 struct inode
*inode
, *old_inode
= NULL
;
1894 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1895 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1896 struct address_space
*mapping
= inode
->i_mapping
;
1898 spin_lock(&inode
->i_lock
);
1899 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1900 mapping
->nrpages
== 0) {
1901 spin_unlock(&inode
->i_lock
);
1905 spin_unlock(&inode
->i_lock
);
1906 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1908 * We hold a reference to 'inode' so it couldn't have been
1909 * removed from s_inodes list while we dropped the
1910 * s_inode_list_lock We cannot iput the inode now as we can
1911 * be holding the last reference and we cannot iput it under
1912 * s_inode_list_lock. So we keep the reference and iput it
1918 func(I_BDEV(inode
), arg
);
1920 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1922 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);