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/buffer_head.h>
20 #include <linux/pagevec.h>
21 #include <linux/writeback.h>
22 #include <linux/mpage.h>
23 #include <linux/mount.h>
24 #include <linux/uio.h>
25 #include <linux/namei.h>
26 #include <linux/log2.h>
27 #include <linux/kmemleak.h>
28 #include <asm/uaccess.h>
32 struct block_device bdev
;
33 struct inode vfs_inode
;
36 static const struct address_space_operations def_blk_aops
;
38 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
40 return container_of(inode
, struct bdev_inode
, vfs_inode
);
43 inline struct block_device
*I_BDEV(struct inode
*inode
)
45 return &BDEV_I(inode
)->bdev
;
47 EXPORT_SYMBOL(I_BDEV
);
50 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
51 * need to move it onto the dirty list of @dst so that the inode is always on
54 static void bdev_inode_switch_bdi(struct inode
*inode
,
55 struct backing_dev_info
*dst
)
57 struct backing_dev_info
*old
= inode
->i_data
.backing_dev_info
;
59 if (unlikely(dst
== old
)) /* deadlock avoidance */
61 bdi_lock_two(&old
->wb
, &dst
->wb
);
62 spin_lock(&inode
->i_lock
);
63 inode
->i_data
.backing_dev_info
= dst
;
64 if (inode
->i_state
& I_DIRTY
)
65 list_move(&inode
->i_wb_list
, &dst
->wb
.b_dirty
);
66 spin_unlock(&inode
->i_lock
);
67 spin_unlock(&old
->wb
.list_lock
);
68 spin_unlock(&dst
->wb
.list_lock
);
71 static sector_t
max_block(struct block_device
*bdev
)
73 sector_t retval
= ~((sector_t
)0);
74 loff_t sz
= i_size_read(bdev
->bd_inode
);
77 unsigned int size
= block_size(bdev
);
78 unsigned int sizebits
= blksize_bits(size
);
79 retval
= (sz
>> sizebits
);
84 /* Kill _all_ buffers and pagecache , dirty or not.. */
85 static void kill_bdev(struct block_device
*bdev
)
87 if (bdev
->bd_inode
->i_mapping
->nrpages
== 0)
90 truncate_inode_pages(bdev
->bd_inode
->i_mapping
, 0);
93 int set_blocksize(struct block_device
*bdev
, int size
)
95 /* Size must be a power of two, and between 512 and PAGE_SIZE */
96 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
99 /* Size cannot be smaller than the size supported by the device */
100 if (size
< bdev_logical_block_size(bdev
))
103 /* Don't change the size if it is same as current */
104 if (bdev
->bd_block_size
!= size
) {
106 bdev
->bd_block_size
= size
;
107 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
113 EXPORT_SYMBOL(set_blocksize
);
115 int sb_set_blocksize(struct super_block
*sb
, int size
)
117 if (set_blocksize(sb
->s_bdev
, size
))
119 /* If we get here, we know size is power of two
120 * and it's value is between 512 and PAGE_SIZE */
121 sb
->s_blocksize
= size
;
122 sb
->s_blocksize_bits
= blksize_bits(size
);
123 return sb
->s_blocksize
;
126 EXPORT_SYMBOL(sb_set_blocksize
);
128 int sb_min_blocksize(struct super_block
*sb
, int size
)
130 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
133 return sb_set_blocksize(sb
, size
);
136 EXPORT_SYMBOL(sb_min_blocksize
);
139 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
140 struct buffer_head
*bh
, int create
)
142 if (iblock
>= max_block(I_BDEV(inode
))) {
147 * for reads, we're just trying to fill a partial page.
148 * return a hole, they will have to call get_block again
149 * before they can fill it, and they will get -EIO at that
154 bh
->b_bdev
= I_BDEV(inode
);
155 bh
->b_blocknr
= iblock
;
156 set_buffer_mapped(bh
);
161 blkdev_get_blocks(struct inode
*inode
, sector_t iblock
,
162 struct buffer_head
*bh
, int create
)
164 sector_t end_block
= max_block(I_BDEV(inode
));
165 unsigned long max_blocks
= bh
->b_size
>> inode
->i_blkbits
;
167 if ((iblock
+ max_blocks
) > end_block
) {
168 max_blocks
= end_block
- iblock
;
169 if ((long)max_blocks
<= 0) {
171 return -EIO
; /* write fully beyond EOF */
173 * It is a read which is fully beyond EOF. We return
174 * a !buffer_mapped buffer
180 bh
->b_bdev
= I_BDEV(inode
);
181 bh
->b_blocknr
= iblock
;
182 bh
->b_size
= max_blocks
<< inode
->i_blkbits
;
184 set_buffer_mapped(bh
);
189 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
190 loff_t offset
, unsigned long nr_segs
)
192 struct file
*file
= iocb
->ki_filp
;
193 struct inode
*inode
= file
->f_mapping
->host
;
195 return __blockdev_direct_IO(rw
, iocb
, inode
, I_BDEV(inode
), iov
, offset
,
196 nr_segs
, blkdev_get_blocks
, NULL
, NULL
, 0);
199 int __sync_blockdev(struct block_device
*bdev
, int wait
)
204 return filemap_flush(bdev
->bd_inode
->i_mapping
);
205 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
209 * Write out and wait upon all the dirty data associated with a block
210 * device via its mapping. Does not take the superblock lock.
212 int sync_blockdev(struct block_device
*bdev
)
214 return __sync_blockdev(bdev
, 1);
216 EXPORT_SYMBOL(sync_blockdev
);
219 * Write out and wait upon all dirty data associated with this
220 * device. Filesystem data as well as the underlying block
221 * device. Takes the superblock lock.
223 int fsync_bdev(struct block_device
*bdev
)
225 struct super_block
*sb
= get_super(bdev
);
227 int res
= sync_filesystem(sb
);
231 return sync_blockdev(bdev
);
233 EXPORT_SYMBOL(fsync_bdev
);
236 * freeze_bdev -- lock a filesystem and force it into a consistent state
237 * @bdev: blockdevice to lock
239 * If a superblock is found on this device, we take the s_umount semaphore
240 * on it to make sure nobody unmounts until the snapshot creation is done.
241 * The reference counter (bd_fsfreeze_count) guarantees that only the last
242 * unfreeze process can unfreeze the frozen filesystem actually when multiple
243 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
244 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
247 struct super_block
*freeze_bdev(struct block_device
*bdev
)
249 struct super_block
*sb
;
252 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
253 if (++bdev
->bd_fsfreeze_count
> 1) {
255 * We don't even need to grab a reference - the first call
256 * to freeze_bdev grab an active reference and only the last
257 * thaw_bdev drops it.
259 sb
= get_super(bdev
);
261 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
265 sb
= get_active_super(bdev
);
268 error
= freeze_super(sb
);
270 deactivate_super(sb
);
271 bdev
->bd_fsfreeze_count
--;
272 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
273 return ERR_PTR(error
);
275 deactivate_super(sb
);
278 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
279 return sb
; /* thaw_bdev releases s->s_umount */
281 EXPORT_SYMBOL(freeze_bdev
);
284 * thaw_bdev -- unlock filesystem
285 * @bdev: blockdevice to unlock
286 * @sb: associated superblock
288 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
290 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
294 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
295 if (!bdev
->bd_fsfreeze_count
)
299 if (--bdev
->bd_fsfreeze_count
> 0)
305 error
= thaw_super(sb
);
307 bdev
->bd_fsfreeze_count
++;
308 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
312 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
315 EXPORT_SYMBOL(thaw_bdev
);
317 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
319 return block_write_full_page(page
, blkdev_get_block
, wbc
);
322 static int blkdev_readpage(struct file
* file
, struct page
* page
)
324 return block_read_full_page(page
, blkdev_get_block
);
327 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
328 loff_t pos
, unsigned len
, unsigned flags
,
329 struct page
**pagep
, void **fsdata
)
331 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
335 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
336 loff_t pos
, unsigned len
, unsigned copied
,
337 struct page
*page
, void *fsdata
)
340 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
343 page_cache_release(page
);
350 * for a block special file file->f_path.dentry->d_inode->i_size is zero
351 * so we compute the size by hand (just as in block_read/write above)
353 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int origin
)
355 struct inode
*bd_inode
= file
->f_mapping
->host
;
359 mutex_lock(&bd_inode
->i_mutex
);
360 size
= i_size_read(bd_inode
);
368 offset
+= file
->f_pos
;
374 if (offset
>= 0 && offset
<= size
) {
375 if (offset
!= file
->f_pos
) {
376 file
->f_pos
= offset
;
381 mutex_unlock(&bd_inode
->i_mutex
);
385 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
387 struct inode
*bd_inode
= filp
->f_mapping
->host
;
388 struct block_device
*bdev
= I_BDEV(bd_inode
);
391 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
396 * There is no need to serialise calls to blkdev_issue_flush with
397 * i_mutex and doing so causes performance issues with concurrent
398 * O_SYNC writers to a block device.
400 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
401 if (error
== -EOPNOTSUPP
)
406 EXPORT_SYMBOL(blkdev_fsync
);
412 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
413 static struct kmem_cache
* bdev_cachep __read_mostly
;
415 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
417 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
420 return &ei
->vfs_inode
;
423 static void bdev_i_callback(struct rcu_head
*head
)
425 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
426 struct bdev_inode
*bdi
= BDEV_I(inode
);
428 INIT_LIST_HEAD(&inode
->i_dentry
);
429 kmem_cache_free(bdev_cachep
, bdi
);
432 static void bdev_destroy_inode(struct inode
*inode
)
434 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
437 static void init_once(void *foo
)
439 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
440 struct block_device
*bdev
= &ei
->bdev
;
442 memset(bdev
, 0, sizeof(*bdev
));
443 mutex_init(&bdev
->bd_mutex
);
444 INIT_LIST_HEAD(&bdev
->bd_inodes
);
445 INIT_LIST_HEAD(&bdev
->bd_list
);
447 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
449 inode_init_once(&ei
->vfs_inode
);
450 /* Initialize mutex for freeze. */
451 mutex_init(&bdev
->bd_fsfreeze_mutex
);
454 static inline void __bd_forget(struct inode
*inode
)
456 list_del_init(&inode
->i_devices
);
457 inode
->i_bdev
= NULL
;
458 inode
->i_mapping
= &inode
->i_data
;
461 static void bdev_evict_inode(struct inode
*inode
)
463 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
465 truncate_inode_pages(&inode
->i_data
, 0);
466 invalidate_inode_buffers(inode
); /* is it needed here? */
467 end_writeback(inode
);
468 spin_lock(&bdev_lock
);
469 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
470 __bd_forget(list_entry(p
, struct inode
, i_devices
));
472 list_del_init(&bdev
->bd_list
);
473 spin_unlock(&bdev_lock
);
476 static const struct super_operations bdev_sops
= {
477 .statfs
= simple_statfs
,
478 .alloc_inode
= bdev_alloc_inode
,
479 .destroy_inode
= bdev_destroy_inode
,
480 .drop_inode
= generic_delete_inode
,
481 .evict_inode
= bdev_evict_inode
,
484 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
485 int flags
, const char *dev_name
, void *data
)
487 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, 0x62646576);
490 static struct file_system_type bd_type
= {
493 .kill_sb
= kill_anon_super
,
496 struct super_block
*blockdev_superblock __read_mostly
;
498 void __init
bdev_cache_init(void)
501 struct vfsmount
*bd_mnt
;
503 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
504 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
505 SLAB_MEM_SPREAD
|SLAB_PANIC
),
507 err
= register_filesystem(&bd_type
);
509 panic("Cannot register bdev pseudo-fs");
510 bd_mnt
= kern_mount(&bd_type
);
512 panic("Cannot create bdev pseudo-fs");
514 * This vfsmount structure is only used to obtain the
515 * blockdev_superblock, so tell kmemleak not to report it.
517 kmemleak_not_leak(bd_mnt
);
518 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
522 * Most likely _very_ bad one - but then it's hardly critical for small
523 * /dev and can be fixed when somebody will need really large one.
524 * Keep in mind that it will be fed through icache hash function too.
526 static inline unsigned long hash(dev_t dev
)
528 return MAJOR(dev
)+MINOR(dev
);
531 static int bdev_test(struct inode
*inode
, void *data
)
533 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
536 static int bdev_set(struct inode
*inode
, void *data
)
538 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
542 static LIST_HEAD(all_bdevs
);
544 struct block_device
*bdget(dev_t dev
)
546 struct block_device
*bdev
;
549 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
550 bdev_test
, bdev_set
, &dev
);
555 bdev
= &BDEV_I(inode
)->bdev
;
557 if (inode
->i_state
& I_NEW
) {
558 bdev
->bd_contains
= NULL
;
559 bdev
->bd_super
= NULL
;
560 bdev
->bd_inode
= inode
;
561 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
562 bdev
->bd_part_count
= 0;
563 bdev
->bd_invalidated
= 0;
564 inode
->i_mode
= S_IFBLK
;
566 inode
->i_bdev
= bdev
;
567 inode
->i_data
.a_ops
= &def_blk_aops
;
568 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
569 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
570 spin_lock(&bdev_lock
);
571 list_add(&bdev
->bd_list
, &all_bdevs
);
572 spin_unlock(&bdev_lock
);
573 unlock_new_inode(inode
);
578 EXPORT_SYMBOL(bdget
);
581 * bdgrab -- Grab a reference to an already referenced block device
582 * @bdev: Block device to grab a reference to.
584 struct block_device
*bdgrab(struct block_device
*bdev
)
586 ihold(bdev
->bd_inode
);
590 long nr_blockdev_pages(void)
592 struct block_device
*bdev
;
594 spin_lock(&bdev_lock
);
595 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
596 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
598 spin_unlock(&bdev_lock
);
602 void bdput(struct block_device
*bdev
)
604 iput(bdev
->bd_inode
);
607 EXPORT_SYMBOL(bdput
);
609 static struct block_device
*bd_acquire(struct inode
*inode
)
611 struct block_device
*bdev
;
613 spin_lock(&bdev_lock
);
614 bdev
= inode
->i_bdev
;
616 ihold(bdev
->bd_inode
);
617 spin_unlock(&bdev_lock
);
620 spin_unlock(&bdev_lock
);
622 bdev
= bdget(inode
->i_rdev
);
624 spin_lock(&bdev_lock
);
625 if (!inode
->i_bdev
) {
627 * We take an additional reference to bd_inode,
628 * and it's released in clear_inode() of inode.
629 * So, we can access it via ->i_mapping always
632 ihold(bdev
->bd_inode
);
633 inode
->i_bdev
= bdev
;
634 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
635 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
637 spin_unlock(&bdev_lock
);
642 /* Call when you free inode */
644 void bd_forget(struct inode
*inode
)
646 struct block_device
*bdev
= NULL
;
648 spin_lock(&bdev_lock
);
650 if (!sb_is_blkdev_sb(inode
->i_sb
))
651 bdev
= inode
->i_bdev
;
654 spin_unlock(&bdev_lock
);
657 iput(bdev
->bd_inode
);
661 * bd_may_claim - test whether a block device can be claimed
662 * @bdev: block device of interest
663 * @whole: whole block device containing @bdev, may equal @bdev
664 * @holder: holder trying to claim @bdev
666 * Test whether @bdev can be claimed by @holder.
669 * spin_lock(&bdev_lock).
672 * %true if @bdev can be claimed, %false otherwise.
674 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
677 if (bdev
->bd_holder
== holder
)
678 return true; /* already a holder */
679 else if (bdev
->bd_holder
!= NULL
)
680 return false; /* held by someone else */
681 else if (bdev
->bd_contains
== bdev
)
682 return true; /* is a whole device which isn't held */
684 else if (whole
->bd_holder
== bd_may_claim
)
685 return true; /* is a partition of a device that is being partitioned */
686 else if (whole
->bd_holder
!= NULL
)
687 return false; /* is a partition of a held device */
689 return true; /* is a partition of an un-held device */
693 * bd_prepare_to_claim - prepare to claim a block device
694 * @bdev: block device of interest
695 * @whole: the whole device containing @bdev, may equal @bdev
696 * @holder: holder trying to claim @bdev
698 * Prepare to claim @bdev. This function fails if @bdev is already
699 * claimed by another holder and waits if another claiming is in
700 * progress. This function doesn't actually claim. On successful
701 * return, the caller has ownership of bd_claiming and bd_holder[s].
704 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
708 * 0 if @bdev can be claimed, -EBUSY otherwise.
710 static int bd_prepare_to_claim(struct block_device
*bdev
,
711 struct block_device
*whole
, void *holder
)
714 /* if someone else claimed, fail */
715 if (!bd_may_claim(bdev
, whole
, holder
))
718 /* if claiming is already in progress, wait for it to finish */
719 if (whole
->bd_claiming
) {
720 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
723 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
724 spin_unlock(&bdev_lock
);
726 finish_wait(wq
, &wait
);
727 spin_lock(&bdev_lock
);
736 * bd_start_claiming - start claiming a block device
737 * @bdev: block device of interest
738 * @holder: holder trying to claim @bdev
740 * @bdev is about to be opened exclusively. Check @bdev can be opened
741 * exclusively and mark that an exclusive open is in progress. Each
742 * successful call to this function must be matched with a call to
743 * either bd_finish_claiming() or bd_abort_claiming() (which do not
746 * This function is used to gain exclusive access to the block device
747 * without actually causing other exclusive open attempts to fail. It
748 * should be used when the open sequence itself requires exclusive
749 * access but may subsequently fail.
755 * Pointer to the block device containing @bdev on success, ERR_PTR()
758 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
761 struct gendisk
*disk
;
762 struct block_device
*whole
;
768 * @bdev might not have been initialized properly yet, look up
769 * and grab the outer block device the hard way.
771 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
773 return ERR_PTR(-ENXIO
);
776 * Normally, @bdev should equal what's returned from bdget_disk()
777 * if partno is 0; however, some drivers (floppy) use multiple
778 * bdev's for the same physical device and @bdev may be one of the
779 * aliases. Keep @bdev if partno is 0. This means claimer
780 * tracking is broken for those devices but it has always been that
784 whole
= bdget_disk(disk
, 0);
786 whole
= bdgrab(bdev
);
788 module_put(disk
->fops
->owner
);
791 return ERR_PTR(-ENOMEM
);
793 /* prepare to claim, if successful, mark claiming in progress */
794 spin_lock(&bdev_lock
);
796 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
798 whole
->bd_claiming
= holder
;
799 spin_unlock(&bdev_lock
);
802 spin_unlock(&bdev_lock
);
809 struct bd_holder_disk
{
810 struct list_head list
;
811 struct gendisk
*disk
;
815 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
816 struct gendisk
*disk
)
818 struct bd_holder_disk
*holder
;
820 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
821 if (holder
->disk
== disk
)
826 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
828 return sysfs_create_link(from
, to
, kobject_name(to
));
831 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
833 sysfs_remove_link(from
, kobject_name(to
));
837 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
838 * @bdev: the claimed slave bdev
839 * @disk: the holding disk
841 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
843 * This functions creates the following sysfs symlinks.
845 * - from "slaves" directory of the holder @disk to the claimed @bdev
846 * - from "holders" directory of the @bdev to the holder @disk
848 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
849 * passed to bd_link_disk_holder(), then:
851 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
852 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
854 * The caller must have claimed @bdev before calling this function and
855 * ensure that both @bdev and @disk are valid during the creation and
856 * lifetime of these symlinks.
862 * 0 on success, -errno on failure.
864 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
866 struct bd_holder_disk
*holder
;
869 mutex_lock(&bdev
->bd_mutex
);
871 WARN_ON_ONCE(!bdev
->bd_holder
);
873 /* FIXME: remove the following once add_disk() handles errors */
874 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
877 holder
= bd_find_holder_disk(bdev
, disk
);
883 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
889 INIT_LIST_HEAD(&holder
->list
);
893 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
897 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
901 * bdev could be deleted beneath us which would implicitly destroy
902 * the holder directory. Hold on to it.
904 kobject_get(bdev
->bd_part
->holder_dir
);
906 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
910 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
914 mutex_unlock(&bdev
->bd_mutex
);
917 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
920 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
921 * @bdev: the calimed slave bdev
922 * @disk: the holding disk
924 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
929 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
931 struct bd_holder_disk
*holder
;
933 mutex_lock(&bdev
->bd_mutex
);
935 holder
= bd_find_holder_disk(bdev
, disk
);
937 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
938 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
939 del_symlink(bdev
->bd_part
->holder_dir
,
940 &disk_to_dev(disk
)->kobj
);
941 kobject_put(bdev
->bd_part
->holder_dir
);
942 list_del_init(&holder
->list
);
946 mutex_unlock(&bdev
->bd_mutex
);
948 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
952 * flush_disk - invalidates all buffer-cache entries on a disk
954 * @bdev: struct block device to be flushed
955 * @kill_dirty: flag to guide handling of dirty inodes
957 * Invalidates all buffer-cache entries on a disk. It should be called
958 * when a disk has been changed -- either by a media change or online
961 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
963 if (__invalidate_device(bdev
, kill_dirty
)) {
964 char name
[BDEVNAME_SIZE
] = "";
967 disk_name(bdev
->bd_disk
, 0, name
);
968 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
969 "resized disk %s\n", name
);
974 if (disk_part_scan_enabled(bdev
->bd_disk
))
975 bdev
->bd_invalidated
= 1;
979 * check_disk_size_change - checks for disk size change and adjusts bdev size.
980 * @disk: struct gendisk to check
981 * @bdev: struct bdev to adjust.
983 * This routine checks to see if the bdev size does not match the disk size
984 * and adjusts it if it differs.
986 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
988 loff_t disk_size
, bdev_size
;
990 disk_size
= (loff_t
)get_capacity(disk
) << 9;
991 bdev_size
= i_size_read(bdev
->bd_inode
);
992 if (disk_size
!= bdev_size
) {
993 char name
[BDEVNAME_SIZE
];
995 disk_name(disk
, 0, name
);
997 "%s: detected capacity change from %lld to %lld\n",
998 name
, bdev_size
, disk_size
);
999 i_size_write(bdev
->bd_inode
, disk_size
);
1000 flush_disk(bdev
, false);
1003 EXPORT_SYMBOL(check_disk_size_change
);
1006 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1007 * @disk: struct gendisk to be revalidated
1009 * This routine is a wrapper for lower-level driver's revalidate_disk
1010 * call-backs. It is used to do common pre and post operations needed
1011 * for all revalidate_disk operations.
1013 int revalidate_disk(struct gendisk
*disk
)
1015 struct block_device
*bdev
;
1018 if (disk
->fops
->revalidate_disk
)
1019 ret
= disk
->fops
->revalidate_disk(disk
);
1021 bdev
= bdget_disk(disk
, 0);
1025 mutex_lock(&bdev
->bd_mutex
);
1026 check_disk_size_change(disk
, bdev
);
1027 mutex_unlock(&bdev
->bd_mutex
);
1031 EXPORT_SYMBOL(revalidate_disk
);
1034 * This routine checks whether a removable media has been changed,
1035 * and invalidates all buffer-cache-entries in that case. This
1036 * is a relatively slow routine, so we have to try to minimize using
1037 * it. Thus it is called only upon a 'mount' or 'open'. This
1038 * is the best way of combining speed and utility, I think.
1039 * People changing diskettes in the middle of an operation deserve
1042 int check_disk_change(struct block_device
*bdev
)
1044 struct gendisk
*disk
= bdev
->bd_disk
;
1045 const struct block_device_operations
*bdops
= disk
->fops
;
1046 unsigned int events
;
1048 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1049 DISK_EVENT_EJECT_REQUEST
);
1050 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1053 flush_disk(bdev
, true);
1054 if (bdops
->revalidate_disk
)
1055 bdops
->revalidate_disk(bdev
->bd_disk
);
1059 EXPORT_SYMBOL(check_disk_change
);
1061 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1063 unsigned bsize
= bdev_logical_block_size(bdev
);
1065 bdev
->bd_inode
->i_size
= size
;
1066 while (bsize
< PAGE_CACHE_SIZE
) {
1071 bdev
->bd_block_size
= bsize
;
1072 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1074 EXPORT_SYMBOL(bd_set_size
);
1076 static int __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1081 * mutex_lock(part->bd_mutex)
1082 * mutex_lock_nested(whole->bd_mutex, 1)
1085 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1087 struct gendisk
*disk
;
1092 if (mode
& FMODE_READ
)
1094 if (mode
& FMODE_WRITE
)
1097 * hooks: /n/, see "layering violations".
1100 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1110 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1114 disk_block_events(disk
);
1115 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1116 if (!bdev
->bd_openers
) {
1117 bdev
->bd_disk
= disk
;
1118 bdev
->bd_contains
= bdev
;
1120 struct backing_dev_info
*bdi
;
1123 bdev
->bd_part
= disk_get_part(disk
, partno
);
1128 if (disk
->fops
->open
) {
1129 ret
= disk
->fops
->open(bdev
, mode
);
1130 if (ret
== -ERESTARTSYS
) {
1131 /* Lost a race with 'disk' being
1132 * deleted, try again.
1135 disk_put_part(bdev
->bd_part
);
1136 bdev
->bd_part
= NULL
;
1137 bdev
->bd_disk
= NULL
;
1138 mutex_unlock(&bdev
->bd_mutex
);
1139 disk_unblock_events(disk
);
1140 module_put(disk
->fops
->owner
);
1146 if (!ret
&& !bdev
->bd_openers
) {
1147 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1148 bdi
= blk_get_backing_dev_info(bdev
);
1150 bdi
= &default_backing_dev_info
;
1151 bdev_inode_switch_bdi(bdev
->bd_inode
, bdi
);
1155 * If the device is invalidated, rescan partition
1156 * if open succeeded or failed with -ENOMEDIUM.
1157 * The latter is necessary to prevent ghost
1158 * partitions on a removed medium.
1160 if (bdev
->bd_invalidated
&& (!ret
|| ret
== -ENOMEDIUM
))
1161 rescan_partitions(disk
, bdev
);
1165 struct block_device
*whole
;
1166 whole
= bdget_disk(disk
, 0);
1171 ret
= __blkdev_get(whole
, mode
, 1);
1174 bdev
->bd_contains
= whole
;
1175 bdev_inode_switch_bdi(bdev
->bd_inode
,
1176 whole
->bd_inode
->i_data
.backing_dev_info
);
1177 bdev
->bd_part
= disk_get_part(disk
, partno
);
1178 if (!(disk
->flags
& GENHD_FL_UP
) ||
1179 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1183 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1186 if (bdev
->bd_contains
== bdev
) {
1188 if (bdev
->bd_disk
->fops
->open
)
1189 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1190 /* the same as first opener case, read comment there */
1191 if (bdev
->bd_invalidated
&& (!ret
|| ret
== -ENOMEDIUM
))
1192 rescan_partitions(bdev
->bd_disk
, bdev
);
1194 goto out_unlock_bdev
;
1196 /* only one opener holds refs to the module and disk */
1197 module_put(disk
->fops
->owner
);
1202 bdev
->bd_part_count
++;
1203 mutex_unlock(&bdev
->bd_mutex
);
1204 disk_unblock_events(disk
);
1208 disk_put_part(bdev
->bd_part
);
1209 bdev
->bd_disk
= NULL
;
1210 bdev
->bd_part
= NULL
;
1211 bdev_inode_switch_bdi(bdev
->bd_inode
, &default_backing_dev_info
);
1212 if (bdev
!= bdev
->bd_contains
)
1213 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1214 bdev
->bd_contains
= NULL
;
1216 mutex_unlock(&bdev
->bd_mutex
);
1217 disk_unblock_events(disk
);
1218 module_put(disk
->fops
->owner
);
1227 * blkdev_get - open a block device
1228 * @bdev: block_device to open
1229 * @mode: FMODE_* mask
1230 * @holder: exclusive holder identifier
1232 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1233 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1234 * @holder is invalid. Exclusive opens may nest for the same @holder.
1236 * On success, the reference count of @bdev is unchanged. On failure,
1243 * 0 on success, -errno on failure.
1245 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1247 struct block_device
*whole
= NULL
;
1250 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1252 if ((mode
& FMODE_EXCL
) && holder
) {
1253 whole
= bd_start_claiming(bdev
, holder
);
1254 if (IS_ERR(whole
)) {
1256 return PTR_ERR(whole
);
1260 res
= __blkdev_get(bdev
, mode
, 0);
1263 struct gendisk
*disk
= whole
->bd_disk
;
1265 /* finish claiming */
1266 mutex_lock(&bdev
->bd_mutex
);
1267 spin_lock(&bdev_lock
);
1270 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1272 * Note that for a whole device bd_holders
1273 * will be incremented twice, and bd_holder
1274 * will be set to bd_may_claim before being
1277 whole
->bd_holders
++;
1278 whole
->bd_holder
= bd_may_claim
;
1280 bdev
->bd_holder
= holder
;
1283 /* tell others that we're done */
1284 BUG_ON(whole
->bd_claiming
!= holder
);
1285 whole
->bd_claiming
= NULL
;
1286 wake_up_bit(&whole
->bd_claiming
, 0);
1288 spin_unlock(&bdev_lock
);
1291 * Block event polling for write claims if requested. Any
1292 * write holder makes the write_holder state stick until
1293 * all are released. This is good enough and tracking
1294 * individual writeable reference is too fragile given the
1295 * way @mode is used in blkdev_get/put().
1297 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1298 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1299 bdev
->bd_write_holder
= true;
1300 disk_block_events(disk
);
1303 mutex_unlock(&bdev
->bd_mutex
);
1309 EXPORT_SYMBOL(blkdev_get
);
1312 * blkdev_get_by_path - open a block device by name
1313 * @path: path to the block device to open
1314 * @mode: FMODE_* mask
1315 * @holder: exclusive holder identifier
1317 * Open the blockdevice described by the device file at @path. @mode
1318 * and @holder are identical to blkdev_get().
1320 * On success, the returned block_device has reference count of one.
1326 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1328 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1331 struct block_device
*bdev
;
1334 bdev
= lookup_bdev(path
);
1338 err
= blkdev_get(bdev
, mode
, holder
);
1340 return ERR_PTR(err
);
1342 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1343 blkdev_put(bdev
, mode
);
1344 return ERR_PTR(-EACCES
);
1349 EXPORT_SYMBOL(blkdev_get_by_path
);
1352 * blkdev_get_by_dev - open a block device by device number
1353 * @dev: device number of block device to open
1354 * @mode: FMODE_* mask
1355 * @holder: exclusive holder identifier
1357 * Open the blockdevice described by device number @dev. @mode and
1358 * @holder are identical to blkdev_get().
1360 * Use it ONLY if you really do not have anything better - i.e. when
1361 * you are behind a truly sucky interface and all you are given is a
1362 * device number. _Never_ to be used for internal purposes. If you
1363 * ever need it - reconsider your API.
1365 * On success, the returned block_device has reference count of one.
1371 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1373 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1375 struct block_device
*bdev
;
1380 return ERR_PTR(-ENOMEM
);
1382 err
= blkdev_get(bdev
, mode
, holder
);
1384 return ERR_PTR(err
);
1388 EXPORT_SYMBOL(blkdev_get_by_dev
);
1390 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1392 struct block_device
*bdev
;
1395 * Preserve backwards compatibility and allow large file access
1396 * even if userspace doesn't ask for it explicitly. Some mkfs
1397 * binary needs it. We might want to drop this workaround
1398 * during an unstable branch.
1400 filp
->f_flags
|= O_LARGEFILE
;
1402 if (filp
->f_flags
& O_NDELAY
)
1403 filp
->f_mode
|= FMODE_NDELAY
;
1404 if (filp
->f_flags
& O_EXCL
)
1405 filp
->f_mode
|= FMODE_EXCL
;
1406 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1407 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1409 bdev
= bd_acquire(inode
);
1413 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1415 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1418 static int __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1421 struct gendisk
*disk
= bdev
->bd_disk
;
1422 struct block_device
*victim
= NULL
;
1424 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1426 bdev
->bd_part_count
--;
1428 if (!--bdev
->bd_openers
) {
1429 WARN_ON_ONCE(bdev
->bd_holders
);
1430 sync_blockdev(bdev
);
1433 if (bdev
->bd_contains
== bdev
) {
1434 if (disk
->fops
->release
)
1435 ret
= disk
->fops
->release(disk
, mode
);
1437 if (!bdev
->bd_openers
) {
1438 struct module
*owner
= disk
->fops
->owner
;
1442 disk_put_part(bdev
->bd_part
);
1443 bdev
->bd_part
= NULL
;
1444 bdev
->bd_disk
= NULL
;
1445 bdev_inode_switch_bdi(bdev
->bd_inode
,
1446 &default_backing_dev_info
);
1447 if (bdev
!= bdev
->bd_contains
)
1448 victim
= bdev
->bd_contains
;
1449 bdev
->bd_contains
= NULL
;
1451 mutex_unlock(&bdev
->bd_mutex
);
1454 __blkdev_put(victim
, mode
, 1);
1458 int blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1460 mutex_lock(&bdev
->bd_mutex
);
1462 if (mode
& FMODE_EXCL
) {
1466 * Release a claim on the device. The holder fields
1467 * are protected with bdev_lock. bd_mutex is to
1468 * synchronize disk_holder unlinking.
1470 spin_lock(&bdev_lock
);
1472 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1473 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1475 /* bd_contains might point to self, check in a separate step */
1476 if ((bdev_free
= !bdev
->bd_holders
))
1477 bdev
->bd_holder
= NULL
;
1478 if (!bdev
->bd_contains
->bd_holders
)
1479 bdev
->bd_contains
->bd_holder
= NULL
;
1481 spin_unlock(&bdev_lock
);
1484 * If this was the last claim, remove holder link and
1485 * unblock evpoll if it was a write holder.
1487 if (bdev_free
&& bdev
->bd_write_holder
) {
1488 disk_unblock_events(bdev
->bd_disk
);
1489 bdev
->bd_write_holder
= false;
1494 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1495 * event. This is to ensure detection of media removal commanded
1496 * from userland - e.g. eject(1).
1498 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1500 mutex_unlock(&bdev
->bd_mutex
);
1502 return __blkdev_put(bdev
, mode
, 0);
1504 EXPORT_SYMBOL(blkdev_put
);
1506 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1508 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1510 return blkdev_put(bdev
, filp
->f_mode
);
1513 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1515 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1516 fmode_t mode
= file
->f_mode
;
1519 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1520 * to updated it before every ioctl.
1522 if (file
->f_flags
& O_NDELAY
)
1523 mode
|= FMODE_NDELAY
;
1525 mode
&= ~FMODE_NDELAY
;
1527 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1531 * Write data to the block device. Only intended for the block device itself
1532 * and the raw driver which basically is a fake block device.
1534 * Does not take i_mutex for the write and thus is not for general purpose
1537 ssize_t
blkdev_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1538 unsigned long nr_segs
, loff_t pos
)
1540 struct file
*file
= iocb
->ki_filp
;
1543 BUG_ON(iocb
->ki_pos
!= pos
);
1545 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
1546 if (ret
> 0 || ret
== -EIOCBQUEUED
) {
1549 err
= generic_write_sync(file
, pos
, ret
);
1550 if (err
< 0 && ret
> 0)
1555 EXPORT_SYMBOL_GPL(blkdev_aio_write
);
1558 * Try to release a page associated with block device when the system
1559 * is under memory pressure.
1561 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1563 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1565 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1566 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1568 return try_to_free_buffers(page
);
1571 static const struct address_space_operations def_blk_aops
= {
1572 .readpage
= blkdev_readpage
,
1573 .writepage
= blkdev_writepage
,
1574 .write_begin
= blkdev_write_begin
,
1575 .write_end
= blkdev_write_end
,
1576 .writepages
= generic_writepages
,
1577 .releasepage
= blkdev_releasepage
,
1578 .direct_IO
= blkdev_direct_IO
,
1581 const struct file_operations def_blk_fops
= {
1582 .open
= blkdev_open
,
1583 .release
= blkdev_close
,
1584 .llseek
= block_llseek
,
1585 .read
= do_sync_read
,
1586 .write
= do_sync_write
,
1587 .aio_read
= generic_file_aio_read
,
1588 .aio_write
= blkdev_aio_write
,
1589 .mmap
= generic_file_mmap
,
1590 .fsync
= blkdev_fsync
,
1591 .unlocked_ioctl
= block_ioctl
,
1592 #ifdef CONFIG_COMPAT
1593 .compat_ioctl
= compat_blkdev_ioctl
,
1595 .splice_read
= generic_file_splice_read
,
1596 .splice_write
= generic_file_splice_write
,
1599 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1602 mm_segment_t old_fs
= get_fs();
1604 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1609 EXPORT_SYMBOL(ioctl_by_bdev
);
1612 * lookup_bdev - lookup a struct block_device by name
1613 * @pathname: special file representing the block device
1615 * Get a reference to the blockdevice at @pathname in the current
1616 * namespace if possible and return it. Return ERR_PTR(error)
1619 struct block_device
*lookup_bdev(const char *pathname
)
1621 struct block_device
*bdev
;
1622 struct inode
*inode
;
1626 if (!pathname
|| !*pathname
)
1627 return ERR_PTR(-EINVAL
);
1629 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1631 return ERR_PTR(error
);
1633 inode
= path
.dentry
->d_inode
;
1635 if (!S_ISBLK(inode
->i_mode
))
1638 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1641 bdev
= bd_acquire(inode
);
1648 bdev
= ERR_PTR(error
);
1651 EXPORT_SYMBOL(lookup_bdev
);
1653 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1655 struct super_block
*sb
= get_super(bdev
);
1660 * no need to lock the super, get_super holds the
1661 * read mutex so the filesystem cannot go away
1662 * under us (->put_super runs with the write lock
1665 shrink_dcache_sb(sb
);
1666 res
= invalidate_inodes(sb
, kill_dirty
);
1669 invalidate_bdev(bdev
);
1672 EXPORT_SYMBOL(__invalidate_device
);