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/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <linux/aio.h>
31 #include <asm/uaccess.h>
35 struct block_device bdev
;
36 struct inode vfs_inode
;
39 static const struct address_space_operations def_blk_aops
;
41 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
43 return container_of(inode
, struct bdev_inode
, vfs_inode
);
46 inline struct block_device
*I_BDEV(struct inode
*inode
)
48 return &BDEV_I(inode
)->bdev
;
50 EXPORT_SYMBOL(I_BDEV
);
53 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
54 * need to move it onto the dirty list of @dst so that the inode is always on
57 static void bdev_inode_switch_bdi(struct inode
*inode
,
58 struct backing_dev_info
*dst
)
60 struct backing_dev_info
*old
= inode
->i_data
.backing_dev_info
;
61 bool wakeup_bdi
= false;
63 if (unlikely(dst
== old
)) /* deadlock avoidance */
65 bdi_lock_two(&old
->wb
, &dst
->wb
);
66 spin_lock(&inode
->i_lock
);
67 inode
->i_data
.backing_dev_info
= dst
;
68 if (inode
->i_state
& I_DIRTY
) {
69 if (bdi_cap_writeback_dirty(dst
) && !wb_has_dirty_io(&dst
->wb
))
71 list_move(&inode
->i_wb_list
, &dst
->wb
.b_dirty
);
73 spin_unlock(&inode
->i_lock
);
74 spin_unlock(&old
->wb
.list_lock
);
75 spin_unlock(&dst
->wb
.list_lock
);
78 bdi_wakeup_thread_delayed(dst
);
81 /* Kill _all_ buffers and pagecache , dirty or not.. */
82 void kill_bdev(struct block_device
*bdev
)
84 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
86 if (mapping
->nrpages
== 0 && mapping
->nrshadows
== 0)
90 truncate_inode_pages(mapping
, 0);
92 EXPORT_SYMBOL(kill_bdev
);
94 /* Invalidate clean unused buffers and pagecache. */
95 void invalidate_bdev(struct block_device
*bdev
)
97 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
99 if (mapping
->nrpages
== 0)
102 invalidate_bh_lrus();
103 lru_add_drain_all(); /* make sure all lru add caches are flushed */
104 invalidate_mapping_pages(mapping
, 0, -1);
105 /* 99% of the time, we don't need to flush the cleancache on the bdev.
106 * But, for the strange corners, lets be cautious
108 cleancache_invalidate_inode(mapping
);
110 EXPORT_SYMBOL(invalidate_bdev
);
112 int set_blocksize(struct block_device
*bdev
, int size
)
114 /* Size must be a power of two, and between 512 and PAGE_SIZE */
115 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
118 /* Size cannot be smaller than the size supported by the device */
119 if (size
< bdev_logical_block_size(bdev
))
122 /* Don't change the size if it is same as current */
123 if (bdev
->bd_block_size
!= size
) {
125 bdev
->bd_block_size
= size
;
126 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
132 EXPORT_SYMBOL(set_blocksize
);
134 int sb_set_blocksize(struct super_block
*sb
, int size
)
136 if (set_blocksize(sb
->s_bdev
, size
))
138 /* If we get here, we know size is power of two
139 * and it's value is between 512 and PAGE_SIZE */
140 sb
->s_blocksize
= size
;
141 sb
->s_blocksize_bits
= blksize_bits(size
);
142 return sb
->s_blocksize
;
145 EXPORT_SYMBOL(sb_set_blocksize
);
147 int sb_min_blocksize(struct super_block
*sb
, int size
)
149 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
152 return sb_set_blocksize(sb
, size
);
155 EXPORT_SYMBOL(sb_min_blocksize
);
158 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
159 struct buffer_head
*bh
, int create
)
161 bh
->b_bdev
= I_BDEV(inode
);
162 bh
->b_blocknr
= iblock
;
163 set_buffer_mapped(bh
);
168 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
169 loff_t offset
, unsigned long nr_segs
)
171 struct file
*file
= iocb
->ki_filp
;
172 struct inode
*inode
= file
->f_mapping
->host
;
174 return __blockdev_direct_IO(rw
, iocb
, inode
, I_BDEV(inode
), iov
, offset
,
175 nr_segs
, blkdev_get_block
, NULL
, NULL
, 0);
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 error
= freeze_super(sb
);
249 deactivate_super(sb
);
250 bdev
->bd_fsfreeze_count
--;
251 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
252 return ERR_PTR(error
);
254 deactivate_super(sb
);
257 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
258 return sb
; /* thaw_bdev releases s->s_umount */
260 EXPORT_SYMBOL(freeze_bdev
);
263 * thaw_bdev -- unlock filesystem
264 * @bdev: blockdevice to unlock
265 * @sb: associated superblock
267 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
269 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
273 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
274 if (!bdev
->bd_fsfreeze_count
)
278 if (--bdev
->bd_fsfreeze_count
> 0)
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_write_begin(struct file
*file
, struct address_space
*mapping
,
307 loff_t pos
, unsigned len
, unsigned flags
,
308 struct page
**pagep
, void **fsdata
)
310 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
314 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
315 loff_t pos
, unsigned len
, unsigned copied
,
316 struct page
*page
, void *fsdata
)
319 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
322 page_cache_release(page
);
329 * for a block special file file_inode(file)->i_size is zero
330 * so we compute the size by hand (just as in block_read/write above)
332 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
334 struct inode
*bd_inode
= file
->f_mapping
->host
;
337 mutex_lock(&bd_inode
->i_mutex
);
338 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
339 mutex_unlock(&bd_inode
->i_mutex
);
343 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
345 struct inode
*bd_inode
= filp
->f_mapping
->host
;
346 struct block_device
*bdev
= I_BDEV(bd_inode
);
349 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
354 * There is no need to serialise calls to blkdev_issue_flush with
355 * i_mutex and doing so causes performance issues with concurrent
356 * O_SYNC writers to a block device.
358 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
359 if (error
== -EOPNOTSUPP
)
364 EXPORT_SYMBOL(blkdev_fsync
);
367 * bdev_read_page() - Start reading a page from a block device
368 * @bdev: The device to read the page from
369 * @sector: The offset on the device to read the page to (need not be aligned)
370 * @page: The page to read
372 * On entry, the page should be locked. It will be unlocked when the page
373 * has been read. If the block driver implements rw_page synchronously,
374 * that will be true on exit from this function, but it need not be.
376 * Errors returned by this function are usually "soft", eg out of memory, or
377 * queue full; callers should try a different route to read this page rather
378 * than propagate an error back up the stack.
380 * Return: negative errno if an error occurs, 0 if submission was successful.
382 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
385 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
388 return ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
390 EXPORT_SYMBOL_GPL(bdev_read_page
);
393 * bdev_write_page() - Start writing a page to a block device
394 * @bdev: The device to write the page to
395 * @sector: The offset on the device to write the page to (need not be aligned)
396 * @page: The page to write
397 * @wbc: The writeback_control for the write
399 * On entry, the page should be locked and not currently under writeback.
400 * On exit, if the write started successfully, the page will be unlocked and
401 * under writeback. If the write failed already (eg the driver failed to
402 * queue the page to the device), the page will still be locked. If the
403 * caller is a ->writepage implementation, it will need to unlock the page.
405 * Errors returned by this function are usually "soft", eg out of memory, or
406 * queue full; callers should try a different route to write this page rather
407 * than propagate an error back up the stack.
409 * Return: negative errno if an error occurs, 0 if submission was successful.
411 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
412 struct page
*page
, struct writeback_control
*wbc
)
415 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
416 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
419 set_page_writeback(page
);
420 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
422 end_page_writeback(page
);
427 EXPORT_SYMBOL_GPL(bdev_write_page
);
433 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
434 static struct kmem_cache
* bdev_cachep __read_mostly
;
436 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
438 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
441 return &ei
->vfs_inode
;
444 static void bdev_i_callback(struct rcu_head
*head
)
446 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
447 struct bdev_inode
*bdi
= BDEV_I(inode
);
449 kmem_cache_free(bdev_cachep
, bdi
);
452 static void bdev_destroy_inode(struct inode
*inode
)
454 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
457 static void init_once(void *foo
)
459 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
460 struct block_device
*bdev
= &ei
->bdev
;
462 memset(bdev
, 0, sizeof(*bdev
));
463 mutex_init(&bdev
->bd_mutex
);
464 INIT_LIST_HEAD(&bdev
->bd_inodes
);
465 INIT_LIST_HEAD(&bdev
->bd_list
);
467 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
469 inode_init_once(&ei
->vfs_inode
);
470 /* Initialize mutex for freeze. */
471 mutex_init(&bdev
->bd_fsfreeze_mutex
);
474 static inline void __bd_forget(struct inode
*inode
)
476 list_del_init(&inode
->i_devices
);
477 inode
->i_bdev
= NULL
;
478 inode
->i_mapping
= &inode
->i_data
;
481 static void bdev_evict_inode(struct inode
*inode
)
483 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
485 truncate_inode_pages_final(&inode
->i_data
);
486 invalidate_inode_buffers(inode
); /* is it needed here? */
488 spin_lock(&bdev_lock
);
489 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
490 __bd_forget(list_entry(p
, struct inode
, i_devices
));
492 list_del_init(&bdev
->bd_list
);
493 spin_unlock(&bdev_lock
);
496 static const struct super_operations bdev_sops
= {
497 .statfs
= simple_statfs
,
498 .alloc_inode
= bdev_alloc_inode
,
499 .destroy_inode
= bdev_destroy_inode
,
500 .drop_inode
= generic_delete_inode
,
501 .evict_inode
= bdev_evict_inode
,
504 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
505 int flags
, const char *dev_name
, void *data
)
507 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
510 static struct file_system_type bd_type
= {
513 .kill_sb
= kill_anon_super
,
516 static struct super_block
*blockdev_superblock __read_mostly
;
518 void __init
bdev_cache_init(void)
521 static struct vfsmount
*bd_mnt
;
523 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
524 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
525 SLAB_MEM_SPREAD
|SLAB_PANIC
),
527 err
= register_filesystem(&bd_type
);
529 panic("Cannot register bdev pseudo-fs");
530 bd_mnt
= kern_mount(&bd_type
);
532 panic("Cannot create bdev pseudo-fs");
533 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
537 * Most likely _very_ bad one - but then it's hardly critical for small
538 * /dev and can be fixed when somebody will need really large one.
539 * Keep in mind that it will be fed through icache hash function too.
541 static inline unsigned long hash(dev_t dev
)
543 return MAJOR(dev
)+MINOR(dev
);
546 static int bdev_test(struct inode
*inode
, void *data
)
548 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
551 static int bdev_set(struct inode
*inode
, void *data
)
553 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
557 static LIST_HEAD(all_bdevs
);
559 struct block_device
*bdget(dev_t dev
)
561 struct block_device
*bdev
;
564 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
565 bdev_test
, bdev_set
, &dev
);
570 bdev
= &BDEV_I(inode
)->bdev
;
572 if (inode
->i_state
& I_NEW
) {
573 bdev
->bd_contains
= NULL
;
574 bdev
->bd_super
= NULL
;
575 bdev
->bd_inode
= inode
;
576 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
577 bdev
->bd_part_count
= 0;
578 bdev
->bd_invalidated
= 0;
579 inode
->i_mode
= S_IFBLK
;
581 inode
->i_bdev
= bdev
;
582 inode
->i_data
.a_ops
= &def_blk_aops
;
583 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
584 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
585 spin_lock(&bdev_lock
);
586 list_add(&bdev
->bd_list
, &all_bdevs
);
587 spin_unlock(&bdev_lock
);
588 unlock_new_inode(inode
);
593 EXPORT_SYMBOL(bdget
);
596 * bdgrab -- Grab a reference to an already referenced block device
597 * @bdev: Block device to grab a reference to.
599 struct block_device
*bdgrab(struct block_device
*bdev
)
601 ihold(bdev
->bd_inode
);
604 EXPORT_SYMBOL(bdgrab
);
606 long nr_blockdev_pages(void)
608 struct block_device
*bdev
;
610 spin_lock(&bdev_lock
);
611 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
612 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
614 spin_unlock(&bdev_lock
);
618 void bdput(struct block_device
*bdev
)
620 iput(bdev
->bd_inode
);
623 EXPORT_SYMBOL(bdput
);
625 static struct block_device
*bd_acquire(struct inode
*inode
)
627 struct block_device
*bdev
;
629 spin_lock(&bdev_lock
);
630 bdev
= inode
->i_bdev
;
632 ihold(bdev
->bd_inode
);
633 spin_unlock(&bdev_lock
);
636 spin_unlock(&bdev_lock
);
638 bdev
= bdget(inode
->i_rdev
);
640 spin_lock(&bdev_lock
);
641 if (!inode
->i_bdev
) {
643 * We take an additional reference to bd_inode,
644 * and it's released in clear_inode() of inode.
645 * So, we can access it via ->i_mapping always
648 ihold(bdev
->bd_inode
);
649 inode
->i_bdev
= bdev
;
650 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
651 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
653 spin_unlock(&bdev_lock
);
658 int sb_is_blkdev_sb(struct super_block
*sb
)
660 return sb
== blockdev_superblock
;
663 /* Call when you free inode */
665 void bd_forget(struct inode
*inode
)
667 struct block_device
*bdev
= NULL
;
669 spin_lock(&bdev_lock
);
670 if (!sb_is_blkdev_sb(inode
->i_sb
))
671 bdev
= inode
->i_bdev
;
673 spin_unlock(&bdev_lock
);
676 iput(bdev
->bd_inode
);
680 * bd_may_claim - test whether a block device can be claimed
681 * @bdev: block device of interest
682 * @whole: whole block device containing @bdev, may equal @bdev
683 * @holder: holder trying to claim @bdev
685 * Test whether @bdev can be claimed by @holder.
688 * spin_lock(&bdev_lock).
691 * %true if @bdev can be claimed, %false otherwise.
693 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
696 if (bdev
->bd_holder
== holder
)
697 return true; /* already a holder */
698 else if (bdev
->bd_holder
!= NULL
)
699 return false; /* held by someone else */
700 else if (bdev
->bd_contains
== bdev
)
701 return true; /* is a whole device which isn't held */
703 else if (whole
->bd_holder
== bd_may_claim
)
704 return true; /* is a partition of a device that is being partitioned */
705 else if (whole
->bd_holder
!= NULL
)
706 return false; /* is a partition of a held device */
708 return true; /* is a partition of an un-held device */
712 * bd_prepare_to_claim - prepare to claim a block device
713 * @bdev: block device of interest
714 * @whole: the whole device containing @bdev, may equal @bdev
715 * @holder: holder trying to claim @bdev
717 * Prepare to claim @bdev. This function fails if @bdev is already
718 * claimed by another holder and waits if another claiming is in
719 * progress. This function doesn't actually claim. On successful
720 * return, the caller has ownership of bd_claiming and bd_holder[s].
723 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
727 * 0 if @bdev can be claimed, -EBUSY otherwise.
729 static int bd_prepare_to_claim(struct block_device
*bdev
,
730 struct block_device
*whole
, void *holder
)
733 /* if someone else claimed, fail */
734 if (!bd_may_claim(bdev
, whole
, holder
))
737 /* if claiming is already in progress, wait for it to finish */
738 if (whole
->bd_claiming
) {
739 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
742 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
743 spin_unlock(&bdev_lock
);
745 finish_wait(wq
, &wait
);
746 spin_lock(&bdev_lock
);
755 * bd_start_claiming - start claiming a block device
756 * @bdev: block device of interest
757 * @holder: holder trying to claim @bdev
759 * @bdev is about to be opened exclusively. Check @bdev can be opened
760 * exclusively and mark that an exclusive open is in progress. Each
761 * successful call to this function must be matched with a call to
762 * either bd_finish_claiming() or bd_abort_claiming() (which do not
765 * This function is used to gain exclusive access to the block device
766 * without actually causing other exclusive open attempts to fail. It
767 * should be used when the open sequence itself requires exclusive
768 * access but may subsequently fail.
774 * Pointer to the block device containing @bdev on success, ERR_PTR()
777 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
780 struct gendisk
*disk
;
781 struct block_device
*whole
;
787 * @bdev might not have been initialized properly yet, look up
788 * and grab the outer block device the hard way.
790 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
792 return ERR_PTR(-ENXIO
);
795 * Normally, @bdev should equal what's returned from bdget_disk()
796 * if partno is 0; however, some drivers (floppy) use multiple
797 * bdev's for the same physical device and @bdev may be one of the
798 * aliases. Keep @bdev if partno is 0. This means claimer
799 * tracking is broken for those devices but it has always been that
803 whole
= bdget_disk(disk
, 0);
805 whole
= bdgrab(bdev
);
807 module_put(disk
->fops
->owner
);
810 return ERR_PTR(-ENOMEM
);
812 /* prepare to claim, if successful, mark claiming in progress */
813 spin_lock(&bdev_lock
);
815 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
817 whole
->bd_claiming
= holder
;
818 spin_unlock(&bdev_lock
);
821 spin_unlock(&bdev_lock
);
828 struct bd_holder_disk
{
829 struct list_head list
;
830 struct gendisk
*disk
;
834 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
835 struct gendisk
*disk
)
837 struct bd_holder_disk
*holder
;
839 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
840 if (holder
->disk
== disk
)
845 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
847 return sysfs_create_link(from
, to
, kobject_name(to
));
850 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
852 sysfs_remove_link(from
, kobject_name(to
));
856 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
857 * @bdev: the claimed slave bdev
858 * @disk: the holding disk
860 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
862 * This functions creates the following sysfs symlinks.
864 * - from "slaves" directory of the holder @disk to the claimed @bdev
865 * - from "holders" directory of the @bdev to the holder @disk
867 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
868 * passed to bd_link_disk_holder(), then:
870 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
871 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
873 * The caller must have claimed @bdev before calling this function and
874 * ensure that both @bdev and @disk are valid during the creation and
875 * lifetime of these symlinks.
881 * 0 on success, -errno on failure.
883 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
885 struct bd_holder_disk
*holder
;
888 mutex_lock(&bdev
->bd_mutex
);
890 WARN_ON_ONCE(!bdev
->bd_holder
);
892 /* FIXME: remove the following once add_disk() handles errors */
893 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
896 holder
= bd_find_holder_disk(bdev
, disk
);
902 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
908 INIT_LIST_HEAD(&holder
->list
);
912 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
916 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
920 * bdev could be deleted beneath us which would implicitly destroy
921 * the holder directory. Hold on to it.
923 kobject_get(bdev
->bd_part
->holder_dir
);
925 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
929 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
933 mutex_unlock(&bdev
->bd_mutex
);
936 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
939 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
940 * @bdev: the calimed slave bdev
941 * @disk: the holding disk
943 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
948 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
950 struct bd_holder_disk
*holder
;
952 mutex_lock(&bdev
->bd_mutex
);
954 holder
= bd_find_holder_disk(bdev
, disk
);
956 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
957 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
958 del_symlink(bdev
->bd_part
->holder_dir
,
959 &disk_to_dev(disk
)->kobj
);
960 kobject_put(bdev
->bd_part
->holder_dir
);
961 list_del_init(&holder
->list
);
965 mutex_unlock(&bdev
->bd_mutex
);
967 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
971 * flush_disk - invalidates all buffer-cache entries on a disk
973 * @bdev: struct block device to be flushed
974 * @kill_dirty: flag to guide handling of dirty inodes
976 * Invalidates all buffer-cache entries on a disk. It should be called
977 * when a disk has been changed -- either by a media change or online
980 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
982 if (__invalidate_device(bdev
, kill_dirty
)) {
983 char name
[BDEVNAME_SIZE
] = "";
986 disk_name(bdev
->bd_disk
, 0, name
);
987 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
988 "resized disk %s\n", name
);
993 if (disk_part_scan_enabled(bdev
->bd_disk
))
994 bdev
->bd_invalidated
= 1;
998 * check_disk_size_change - checks for disk size change and adjusts bdev size.
999 * @disk: struct gendisk to check
1000 * @bdev: struct bdev to adjust.
1002 * This routine checks to see if the bdev size does not match the disk size
1003 * and adjusts it if it differs.
1005 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1007 loff_t disk_size
, bdev_size
;
1009 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1010 bdev_size
= i_size_read(bdev
->bd_inode
);
1011 if (disk_size
!= bdev_size
) {
1012 char name
[BDEVNAME_SIZE
];
1014 disk_name(disk
, 0, name
);
1016 "%s: detected capacity change from %lld to %lld\n",
1017 name
, bdev_size
, disk_size
);
1018 i_size_write(bdev
->bd_inode
, disk_size
);
1019 flush_disk(bdev
, false);
1022 EXPORT_SYMBOL(check_disk_size_change
);
1025 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1026 * @disk: struct gendisk to be revalidated
1028 * This routine is a wrapper for lower-level driver's revalidate_disk
1029 * call-backs. It is used to do common pre and post operations needed
1030 * for all revalidate_disk operations.
1032 int revalidate_disk(struct gendisk
*disk
)
1034 struct block_device
*bdev
;
1037 if (disk
->fops
->revalidate_disk
)
1038 ret
= disk
->fops
->revalidate_disk(disk
);
1040 bdev
= bdget_disk(disk
, 0);
1044 mutex_lock(&bdev
->bd_mutex
);
1045 check_disk_size_change(disk
, bdev
);
1046 bdev
->bd_invalidated
= 0;
1047 mutex_unlock(&bdev
->bd_mutex
);
1051 EXPORT_SYMBOL(revalidate_disk
);
1054 * This routine checks whether a removable media has been changed,
1055 * and invalidates all buffer-cache-entries in that case. This
1056 * is a relatively slow routine, so we have to try to minimize using
1057 * it. Thus it is called only upon a 'mount' or 'open'. This
1058 * is the best way of combining speed and utility, I think.
1059 * People changing diskettes in the middle of an operation deserve
1062 int check_disk_change(struct block_device
*bdev
)
1064 struct gendisk
*disk
= bdev
->bd_disk
;
1065 const struct block_device_operations
*bdops
= disk
->fops
;
1066 unsigned int events
;
1068 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1069 DISK_EVENT_EJECT_REQUEST
);
1070 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1073 flush_disk(bdev
, true);
1074 if (bdops
->revalidate_disk
)
1075 bdops
->revalidate_disk(bdev
->bd_disk
);
1079 EXPORT_SYMBOL(check_disk_change
);
1081 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1083 unsigned bsize
= bdev_logical_block_size(bdev
);
1085 mutex_lock(&bdev
->bd_inode
->i_mutex
);
1086 i_size_write(bdev
->bd_inode
, size
);
1087 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
1088 while (bsize
< PAGE_CACHE_SIZE
) {
1093 bdev
->bd_block_size
= bsize
;
1094 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1096 EXPORT_SYMBOL(bd_set_size
);
1098 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1103 * mutex_lock(part->bd_mutex)
1104 * mutex_lock_nested(whole->bd_mutex, 1)
1107 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1109 struct gendisk
*disk
;
1110 struct module
*owner
;
1115 if (mode
& FMODE_READ
)
1117 if (mode
& FMODE_WRITE
)
1120 * hooks: /n/, see "layering violations".
1123 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1133 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1136 owner
= disk
->fops
->owner
;
1138 disk_block_events(disk
);
1139 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1140 if (!bdev
->bd_openers
) {
1141 bdev
->bd_disk
= disk
;
1142 bdev
->bd_queue
= disk
->queue
;
1143 bdev
->bd_contains
= bdev
;
1145 struct backing_dev_info
*bdi
;
1148 bdev
->bd_part
= disk_get_part(disk
, partno
);
1153 if (disk
->fops
->open
) {
1154 ret
= disk
->fops
->open(bdev
, mode
);
1155 if (ret
== -ERESTARTSYS
) {
1156 /* Lost a race with 'disk' being
1157 * deleted, try again.
1160 disk_put_part(bdev
->bd_part
);
1161 bdev
->bd_part
= NULL
;
1162 bdev
->bd_disk
= NULL
;
1163 bdev
->bd_queue
= NULL
;
1164 mutex_unlock(&bdev
->bd_mutex
);
1165 disk_unblock_events(disk
);
1173 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1174 bdi
= blk_get_backing_dev_info(bdev
);
1176 bdi
= &default_backing_dev_info
;
1177 bdev_inode_switch_bdi(bdev
->bd_inode
, bdi
);
1181 * If the device is invalidated, rescan partition
1182 * if open succeeded or failed with -ENOMEDIUM.
1183 * The latter is necessary to prevent ghost
1184 * partitions on a removed medium.
1186 if (bdev
->bd_invalidated
) {
1188 rescan_partitions(disk
, bdev
);
1189 else if (ret
== -ENOMEDIUM
)
1190 invalidate_partitions(disk
, bdev
);
1195 struct block_device
*whole
;
1196 whole
= bdget_disk(disk
, 0);
1201 ret
= __blkdev_get(whole
, mode
, 1);
1204 bdev
->bd_contains
= whole
;
1205 bdev_inode_switch_bdi(bdev
->bd_inode
,
1206 whole
->bd_inode
->i_data
.backing_dev_info
);
1207 bdev
->bd_part
= disk_get_part(disk
, partno
);
1208 if (!(disk
->flags
& GENHD_FL_UP
) ||
1209 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1213 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1216 if (bdev
->bd_contains
== bdev
) {
1218 if (bdev
->bd_disk
->fops
->open
)
1219 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1220 /* the same as first opener case, read comment there */
1221 if (bdev
->bd_invalidated
) {
1223 rescan_partitions(bdev
->bd_disk
, bdev
);
1224 else if (ret
== -ENOMEDIUM
)
1225 invalidate_partitions(bdev
->bd_disk
, bdev
);
1228 goto out_unlock_bdev
;
1230 /* only one opener holds refs to the module and disk */
1236 bdev
->bd_part_count
++;
1237 mutex_unlock(&bdev
->bd_mutex
);
1238 disk_unblock_events(disk
);
1242 disk_put_part(bdev
->bd_part
);
1243 bdev
->bd_disk
= NULL
;
1244 bdev
->bd_part
= NULL
;
1245 bdev
->bd_queue
= NULL
;
1246 bdev_inode_switch_bdi(bdev
->bd_inode
, &default_backing_dev_info
);
1247 if (bdev
!= bdev
->bd_contains
)
1248 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1249 bdev
->bd_contains
= NULL
;
1251 mutex_unlock(&bdev
->bd_mutex
);
1252 disk_unblock_events(disk
);
1262 * blkdev_get - open a block device
1263 * @bdev: block_device to open
1264 * @mode: FMODE_* mask
1265 * @holder: exclusive holder identifier
1267 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1268 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1269 * @holder is invalid. Exclusive opens may nest for the same @holder.
1271 * On success, the reference count of @bdev is unchanged. On failure,
1278 * 0 on success, -errno on failure.
1280 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1282 struct block_device
*whole
= NULL
;
1285 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1287 if ((mode
& FMODE_EXCL
) && holder
) {
1288 whole
= bd_start_claiming(bdev
, holder
);
1289 if (IS_ERR(whole
)) {
1291 return PTR_ERR(whole
);
1295 res
= __blkdev_get(bdev
, mode
, 0);
1298 struct gendisk
*disk
= whole
->bd_disk
;
1300 /* finish claiming */
1301 mutex_lock(&bdev
->bd_mutex
);
1302 spin_lock(&bdev_lock
);
1305 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1307 * Note that for a whole device bd_holders
1308 * will be incremented twice, and bd_holder
1309 * will be set to bd_may_claim before being
1312 whole
->bd_holders
++;
1313 whole
->bd_holder
= bd_may_claim
;
1315 bdev
->bd_holder
= holder
;
1318 /* tell others that we're done */
1319 BUG_ON(whole
->bd_claiming
!= holder
);
1320 whole
->bd_claiming
= NULL
;
1321 wake_up_bit(&whole
->bd_claiming
, 0);
1323 spin_unlock(&bdev_lock
);
1326 * Block event polling for write claims if requested. Any
1327 * write holder makes the write_holder state stick until
1328 * all are released. This is good enough and tracking
1329 * individual writeable reference is too fragile given the
1330 * way @mode is used in blkdev_get/put().
1332 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1333 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1334 bdev
->bd_write_holder
= true;
1335 disk_block_events(disk
);
1338 mutex_unlock(&bdev
->bd_mutex
);
1344 EXPORT_SYMBOL(blkdev_get
);
1347 * blkdev_get_by_path - open a block device by name
1348 * @path: path to the block device to open
1349 * @mode: FMODE_* mask
1350 * @holder: exclusive holder identifier
1352 * Open the blockdevice described by the device file at @path. @mode
1353 * and @holder are identical to blkdev_get().
1355 * On success, the returned block_device has reference count of one.
1361 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1363 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1366 struct block_device
*bdev
;
1369 bdev
= lookup_bdev(path
);
1373 err
= blkdev_get(bdev
, mode
, holder
);
1375 return ERR_PTR(err
);
1377 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1378 blkdev_put(bdev
, mode
);
1379 return ERR_PTR(-EACCES
);
1384 EXPORT_SYMBOL(blkdev_get_by_path
);
1387 * blkdev_get_by_dev - open a block device by device number
1388 * @dev: device number of block device to open
1389 * @mode: FMODE_* mask
1390 * @holder: exclusive holder identifier
1392 * Open the blockdevice described by device number @dev. @mode and
1393 * @holder are identical to blkdev_get().
1395 * Use it ONLY if you really do not have anything better - i.e. when
1396 * you are behind a truly sucky interface and all you are given is a
1397 * device number. _Never_ to be used for internal purposes. If you
1398 * ever need it - reconsider your API.
1400 * On success, the returned block_device has reference count of one.
1406 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1408 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1410 struct block_device
*bdev
;
1415 return ERR_PTR(-ENOMEM
);
1417 err
= blkdev_get(bdev
, mode
, holder
);
1419 return ERR_PTR(err
);
1423 EXPORT_SYMBOL(blkdev_get_by_dev
);
1425 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1427 struct block_device
*bdev
;
1430 * Preserve backwards compatibility and allow large file access
1431 * even if userspace doesn't ask for it explicitly. Some mkfs
1432 * binary needs it. We might want to drop this workaround
1433 * during an unstable branch.
1435 filp
->f_flags
|= O_LARGEFILE
;
1437 if (filp
->f_flags
& O_NDELAY
)
1438 filp
->f_mode
|= FMODE_NDELAY
;
1439 if (filp
->f_flags
& O_EXCL
)
1440 filp
->f_mode
|= FMODE_EXCL
;
1441 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1442 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1444 bdev
= bd_acquire(inode
);
1448 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1450 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1453 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1455 struct gendisk
*disk
= bdev
->bd_disk
;
1456 struct block_device
*victim
= NULL
;
1458 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1460 bdev
->bd_part_count
--;
1462 if (!--bdev
->bd_openers
) {
1463 WARN_ON_ONCE(bdev
->bd_holders
);
1464 sync_blockdev(bdev
);
1466 /* ->release can cause the old bdi to disappear,
1467 * so must switch it out first
1469 bdev_inode_switch_bdi(bdev
->bd_inode
,
1470 &default_backing_dev_info
);
1472 if (bdev
->bd_contains
== bdev
) {
1473 if (disk
->fops
->release
)
1474 disk
->fops
->release(disk
, mode
);
1476 if (!bdev
->bd_openers
) {
1477 struct module
*owner
= disk
->fops
->owner
;
1479 disk_put_part(bdev
->bd_part
);
1480 bdev
->bd_part
= NULL
;
1481 bdev
->bd_disk
= NULL
;
1482 if (bdev
!= bdev
->bd_contains
)
1483 victim
= bdev
->bd_contains
;
1484 bdev
->bd_contains
= NULL
;
1489 mutex_unlock(&bdev
->bd_mutex
);
1492 __blkdev_put(victim
, mode
, 1);
1495 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1497 mutex_lock(&bdev
->bd_mutex
);
1499 if (mode
& FMODE_EXCL
) {
1503 * Release a claim on the device. The holder fields
1504 * are protected with bdev_lock. bd_mutex is to
1505 * synchronize disk_holder unlinking.
1507 spin_lock(&bdev_lock
);
1509 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1510 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1512 /* bd_contains might point to self, check in a separate step */
1513 if ((bdev_free
= !bdev
->bd_holders
))
1514 bdev
->bd_holder
= NULL
;
1515 if (!bdev
->bd_contains
->bd_holders
)
1516 bdev
->bd_contains
->bd_holder
= NULL
;
1518 spin_unlock(&bdev_lock
);
1521 * If this was the last claim, remove holder link and
1522 * unblock evpoll if it was a write holder.
1524 if (bdev_free
&& bdev
->bd_write_holder
) {
1525 disk_unblock_events(bdev
->bd_disk
);
1526 bdev
->bd_write_holder
= false;
1531 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1532 * event. This is to ensure detection of media removal commanded
1533 * from userland - e.g. eject(1).
1535 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1537 mutex_unlock(&bdev
->bd_mutex
);
1539 __blkdev_put(bdev
, mode
, 0);
1541 EXPORT_SYMBOL(blkdev_put
);
1543 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1545 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1546 blkdev_put(bdev
, filp
->f_mode
);
1550 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1552 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1553 fmode_t mode
= file
->f_mode
;
1556 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1557 * to updated it before every ioctl.
1559 if (file
->f_flags
& O_NDELAY
)
1560 mode
|= FMODE_NDELAY
;
1562 mode
&= ~FMODE_NDELAY
;
1564 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1568 * Write data to the block device. Only intended for the block device itself
1569 * and the raw driver which basically is a fake block device.
1571 * Does not take i_mutex for the write and thus is not for general purpose
1574 ssize_t
blkdev_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1575 unsigned long nr_segs
, loff_t pos
)
1577 struct file
*file
= iocb
->ki_filp
;
1578 struct blk_plug plug
;
1581 BUG_ON(iocb
->ki_pos
!= pos
);
1583 blk_start_plug(&plug
);
1584 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
);
1588 err
= generic_write_sync(file
, pos
, ret
);
1592 blk_finish_plug(&plug
);
1595 EXPORT_SYMBOL_GPL(blkdev_aio_write
);
1597 static ssize_t
blkdev_aio_read(struct kiocb
*iocb
, const struct iovec
*iov
,
1598 unsigned long nr_segs
, loff_t pos
)
1600 struct file
*file
= iocb
->ki_filp
;
1601 struct inode
*bd_inode
= file
->f_mapping
->host
;
1602 loff_t size
= i_size_read(bd_inode
);
1608 if (size
< iocb
->ki_nbytes
)
1609 nr_segs
= iov_shorten((struct iovec
*)iov
, nr_segs
, size
);
1610 return generic_file_aio_read(iocb
, iov
, nr_segs
, pos
);
1614 * Try to release a page associated with block device when the system
1615 * is under memory pressure.
1617 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1619 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1621 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1622 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1624 return try_to_free_buffers(page
);
1627 static const struct address_space_operations def_blk_aops
= {
1628 .readpage
= blkdev_readpage
,
1629 .writepage
= blkdev_writepage
,
1630 .write_begin
= blkdev_write_begin
,
1631 .write_end
= blkdev_write_end
,
1632 .writepages
= generic_writepages
,
1633 .releasepage
= blkdev_releasepage
,
1634 .direct_IO
= blkdev_direct_IO
,
1635 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1638 const struct file_operations def_blk_fops
= {
1639 .open
= blkdev_open
,
1640 .release
= blkdev_close
,
1641 .llseek
= block_llseek
,
1642 .read
= do_sync_read
,
1643 .write
= do_sync_write
,
1644 .aio_read
= blkdev_aio_read
,
1645 .aio_write
= blkdev_aio_write
,
1646 .mmap
= generic_file_mmap
,
1647 .fsync
= blkdev_fsync
,
1648 .unlocked_ioctl
= block_ioctl
,
1649 #ifdef CONFIG_COMPAT
1650 .compat_ioctl
= compat_blkdev_ioctl
,
1652 .splice_read
= generic_file_splice_read
,
1653 .splice_write
= generic_file_splice_write
,
1656 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1659 mm_segment_t old_fs
= get_fs();
1661 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1666 EXPORT_SYMBOL(ioctl_by_bdev
);
1669 * lookup_bdev - lookup a struct block_device by name
1670 * @pathname: special file representing the block device
1672 * Get a reference to the blockdevice at @pathname in the current
1673 * namespace if possible and return it. Return ERR_PTR(error)
1676 struct block_device
*lookup_bdev(const char *pathname
)
1678 struct block_device
*bdev
;
1679 struct inode
*inode
;
1683 if (!pathname
|| !*pathname
)
1684 return ERR_PTR(-EINVAL
);
1686 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1688 return ERR_PTR(error
);
1690 inode
= path
.dentry
->d_inode
;
1692 if (!S_ISBLK(inode
->i_mode
))
1695 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1698 bdev
= bd_acquire(inode
);
1705 bdev
= ERR_PTR(error
);
1708 EXPORT_SYMBOL(lookup_bdev
);
1710 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1712 struct super_block
*sb
= get_super(bdev
);
1717 * no need to lock the super, get_super holds the
1718 * read mutex so the filesystem cannot go away
1719 * under us (->put_super runs with the write lock
1722 shrink_dcache_sb(sb
);
1723 res
= invalidate_inodes(sb
, kill_dirty
);
1726 invalidate_bdev(bdev
);
1729 EXPORT_SYMBOL(__invalidate_device
);
1731 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1733 struct inode
*inode
, *old_inode
= NULL
;
1735 spin_lock(&inode_sb_list_lock
);
1736 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1737 struct address_space
*mapping
= inode
->i_mapping
;
1739 spin_lock(&inode
->i_lock
);
1740 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1741 mapping
->nrpages
== 0) {
1742 spin_unlock(&inode
->i_lock
);
1746 spin_unlock(&inode
->i_lock
);
1747 spin_unlock(&inode_sb_list_lock
);
1749 * We hold a reference to 'inode' so it couldn't have been
1750 * removed from s_inodes list while we dropped the
1751 * inode_sb_list_lock. We cannot iput the inode now as we can
1752 * be holding the last reference and we cannot iput it under
1753 * inode_sb_list_lock. So we keep the reference and iput it
1759 func(I_BDEV(inode
), arg
);
1761 spin_lock(&inode_sb_list_lock
);
1763 spin_unlock(&inode_sb_list_lock
);