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/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev
;
34 struct inode vfs_inode
;
37 static const struct address_space_operations def_blk_aops
;
39 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
41 return container_of(inode
, struct bdev_inode
, vfs_inode
);
44 inline struct block_device
*I_BDEV(struct inode
*inode
)
46 return &BDEV_I(inode
)->bdev
;
49 EXPORT_SYMBOL(I_BDEV
);
51 static sector_t
max_block(struct block_device
*bdev
)
53 sector_t retval
= ~((sector_t
)0);
54 loff_t sz
= i_size_read(bdev
->bd_inode
);
57 unsigned int size
= block_size(bdev
);
58 unsigned int sizebits
= blksize_bits(size
);
59 retval
= (sz
>> sizebits
);
64 /* Kill _all_ buffers and pagecache , dirty or not.. */
65 static void kill_bdev(struct block_device
*bdev
)
67 if (bdev
->bd_inode
->i_mapping
->nrpages
== 0)
70 truncate_inode_pages(bdev
->bd_inode
->i_mapping
, 0);
73 int set_blocksize(struct block_device
*bdev
, int size
)
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
76 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
79 /* Size cannot be smaller than the size supported by the device */
80 if (size
< bdev_logical_block_size(bdev
))
83 /* Don't change the size if it is same as current */
84 if (bdev
->bd_block_size
!= size
) {
86 bdev
->bd_block_size
= size
;
87 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
93 EXPORT_SYMBOL(set_blocksize
);
95 int sb_set_blocksize(struct super_block
*sb
, int size
)
97 if (set_blocksize(sb
->s_bdev
, size
))
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb
->s_blocksize
= size
;
102 sb
->s_blocksize_bits
= blksize_bits(size
);
103 return sb
->s_blocksize
;
106 EXPORT_SYMBOL(sb_set_blocksize
);
108 int sb_min_blocksize(struct super_block
*sb
, int size
)
110 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
113 return sb_set_blocksize(sb
, size
);
116 EXPORT_SYMBOL(sb_min_blocksize
);
119 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
120 struct buffer_head
*bh
, int create
)
122 if (iblock
>= max_block(I_BDEV(inode
))) {
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
134 bh
->b_bdev
= I_BDEV(inode
);
135 bh
->b_blocknr
= iblock
;
136 set_buffer_mapped(bh
);
141 blkdev_get_blocks(struct inode
*inode
, sector_t iblock
,
142 struct buffer_head
*bh
, int create
)
144 sector_t end_block
= max_block(I_BDEV(inode
));
145 unsigned long max_blocks
= bh
->b_size
>> inode
->i_blkbits
;
147 if ((iblock
+ max_blocks
) > end_block
) {
148 max_blocks
= end_block
- iblock
;
149 if ((long)max_blocks
<= 0) {
151 return -EIO
; /* write fully beyond EOF */
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
160 bh
->b_bdev
= I_BDEV(inode
);
161 bh
->b_blocknr
= iblock
;
162 bh
->b_size
= max_blocks
<< inode
->i_blkbits
;
164 set_buffer_mapped(bh
);
169 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
170 loff_t offset
, unsigned long nr_segs
)
172 struct file
*file
= iocb
->ki_filp
;
173 struct inode
*inode
= file
->f_mapping
->host
;
175 return blockdev_direct_IO_no_locking(rw
, iocb
, inode
, I_BDEV(inode
),
176 iov
, offset
, nr_segs
, blkdev_get_blocks
, NULL
);
179 int __sync_blockdev(struct block_device
*bdev
, int wait
)
184 return filemap_flush(bdev
->bd_inode
->i_mapping
);
185 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
192 int sync_blockdev(struct block_device
*bdev
)
194 return __sync_blockdev(bdev
, 1);
196 EXPORT_SYMBOL(sync_blockdev
);
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
203 int fsync_bdev(struct block_device
*bdev
)
205 struct super_block
*sb
= get_super(bdev
);
207 int res
= sync_filesystem(sb
);
211 return sync_blockdev(bdev
);
213 EXPORT_SYMBOL(fsync_bdev
);
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
227 struct super_block
*freeze_bdev(struct block_device
*bdev
)
229 struct super_block
*sb
;
232 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
233 if (++bdev
->bd_fsfreeze_count
> 1) {
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
239 sb
= get_super(bdev
);
241 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
245 sb
= get_active_super(bdev
);
248 if (sb
->s_flags
& MS_RDONLY
) {
249 sb
->s_frozen
= SB_FREEZE_TRANS
;
250 up_write(&sb
->s_umount
);
251 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
255 sb
->s_frozen
= SB_FREEZE_WRITE
;
260 sb
->s_frozen
= SB_FREEZE_TRANS
;
263 sync_blockdev(sb
->s_bdev
);
265 if (sb
->s_op
->freeze_fs
) {
266 error
= sb
->s_op
->freeze_fs(sb
);
269 "VFS:Filesystem freeze failed\n");
270 sb
->s_frozen
= SB_UNFROZEN
;
271 deactivate_locked_super(sb
);
272 bdev
->bd_fsfreeze_count
--;
273 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
274 return ERR_PTR(error
);
277 up_write(&sb
->s_umount
);
281 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
282 return sb
; /* thaw_bdev releases s->s_umount */
284 EXPORT_SYMBOL(freeze_bdev
);
287 * thaw_bdev -- unlock filesystem
288 * @bdev: blockdevice to unlock
289 * @sb: associated superblock
291 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
293 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
297 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
298 if (!bdev
->bd_fsfreeze_count
)
302 if (--bdev
->bd_fsfreeze_count
> 0)
308 BUG_ON(sb
->s_bdev
!= bdev
);
309 down_write(&sb
->s_umount
);
310 if (sb
->s_flags
& MS_RDONLY
)
313 if (sb
->s_op
->unfreeze_fs
) {
314 error
= sb
->s_op
->unfreeze_fs(sb
);
317 "VFS:Filesystem thaw failed\n");
318 sb
->s_frozen
= SB_FREEZE_TRANS
;
319 bdev
->bd_fsfreeze_count
++;
320 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
326 sb
->s_frozen
= SB_UNFROZEN
;
328 wake_up(&sb
->s_wait_unfrozen
);
331 deactivate_locked_super(sb
);
333 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
336 EXPORT_SYMBOL(thaw_bdev
);
338 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
340 return block_write_full_page(page
, blkdev_get_block
, wbc
);
343 static int blkdev_readpage(struct file
* file
, struct page
* page
)
345 return block_read_full_page(page
, blkdev_get_block
);
348 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
349 loff_t pos
, unsigned len
, unsigned flags
,
350 struct page
**pagep
, void **fsdata
)
353 return block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
357 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
358 loff_t pos
, unsigned len
, unsigned copied
,
359 struct page
*page
, void *fsdata
)
362 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
365 page_cache_release(page
);
372 * for a block special file file->f_path.dentry->d_inode->i_size is zero
373 * so we compute the size by hand (just as in block_read/write above)
375 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int origin
)
377 struct inode
*bd_inode
= file
->f_mapping
->host
;
381 mutex_lock(&bd_inode
->i_mutex
);
382 size
= i_size_read(bd_inode
);
389 offset
+= file
->f_pos
;
392 if (offset
>= 0 && offset
<= size
) {
393 if (offset
!= file
->f_pos
) {
394 file
->f_pos
= offset
;
398 mutex_unlock(&bd_inode
->i_mutex
);
403 * Filp is never NULL; the only case when ->fsync() is called with
404 * NULL first argument is nfsd_sync_dir() and that's not a directory.
407 static int block_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
409 return sync_blockdev(I_BDEV(filp
->f_mapping
->host
));
416 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
417 static struct kmem_cache
* bdev_cachep __read_mostly
;
419 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
421 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
424 return &ei
->vfs_inode
;
427 static void bdev_destroy_inode(struct inode
*inode
)
429 struct bdev_inode
*bdi
= BDEV_I(inode
);
431 kmem_cache_free(bdev_cachep
, bdi
);
434 static void init_once(void *foo
)
436 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
437 struct block_device
*bdev
= &ei
->bdev
;
439 memset(bdev
, 0, sizeof(*bdev
));
440 mutex_init(&bdev
->bd_mutex
);
441 INIT_LIST_HEAD(&bdev
->bd_inodes
);
442 INIT_LIST_HEAD(&bdev
->bd_list
);
444 INIT_LIST_HEAD(&bdev
->bd_holder_list
);
446 inode_init_once(&ei
->vfs_inode
);
447 /* Initialize mutex for freeze. */
448 mutex_init(&bdev
->bd_fsfreeze_mutex
);
451 static inline void __bd_forget(struct inode
*inode
)
453 list_del_init(&inode
->i_devices
);
454 inode
->i_bdev
= NULL
;
455 inode
->i_mapping
= &inode
->i_data
;
458 static void bdev_clear_inode(struct inode
*inode
)
460 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
462 spin_lock(&bdev_lock
);
463 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
464 __bd_forget(list_entry(p
, struct inode
, i_devices
));
466 list_del_init(&bdev
->bd_list
);
467 spin_unlock(&bdev_lock
);
470 static const struct super_operations bdev_sops
= {
471 .statfs
= simple_statfs
,
472 .alloc_inode
= bdev_alloc_inode
,
473 .destroy_inode
= bdev_destroy_inode
,
474 .drop_inode
= generic_delete_inode
,
475 .clear_inode
= bdev_clear_inode
,
478 static int bd_get_sb(struct file_system_type
*fs_type
,
479 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
481 return get_sb_pseudo(fs_type
, "bdev:", &bdev_sops
, 0x62646576, mnt
);
484 static struct file_system_type bd_type
= {
487 .kill_sb
= kill_anon_super
,
490 struct super_block
*blockdev_superblock __read_mostly
;
492 void __init
bdev_cache_init(void)
495 struct vfsmount
*bd_mnt
;
497 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
498 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
499 SLAB_MEM_SPREAD
|SLAB_PANIC
),
501 err
= register_filesystem(&bd_type
);
503 panic("Cannot register bdev pseudo-fs");
504 bd_mnt
= kern_mount(&bd_type
);
506 panic("Cannot create bdev pseudo-fs");
508 * This vfsmount structure is only used to obtain the
509 * blockdev_superblock, so tell kmemleak not to report it.
511 kmemleak_not_leak(bd_mnt
);
512 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
516 * Most likely _very_ bad one - but then it's hardly critical for small
517 * /dev and can be fixed when somebody will need really large one.
518 * Keep in mind that it will be fed through icache hash function too.
520 static inline unsigned long hash(dev_t dev
)
522 return MAJOR(dev
)+MINOR(dev
);
525 static int bdev_test(struct inode
*inode
, void *data
)
527 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
530 static int bdev_set(struct inode
*inode
, void *data
)
532 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
536 static LIST_HEAD(all_bdevs
);
538 struct block_device
*bdget(dev_t dev
)
540 struct block_device
*bdev
;
543 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
544 bdev_test
, bdev_set
, &dev
);
549 bdev
= &BDEV_I(inode
)->bdev
;
551 if (inode
->i_state
& I_NEW
) {
552 bdev
->bd_contains
= NULL
;
553 bdev
->bd_inode
= inode
;
554 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
555 bdev
->bd_part_count
= 0;
556 bdev
->bd_invalidated
= 0;
557 inode
->i_mode
= S_IFBLK
;
559 inode
->i_bdev
= bdev
;
560 inode
->i_data
.a_ops
= &def_blk_aops
;
561 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
562 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
563 spin_lock(&bdev_lock
);
564 list_add(&bdev
->bd_list
, &all_bdevs
);
565 spin_unlock(&bdev_lock
);
566 unlock_new_inode(inode
);
571 EXPORT_SYMBOL(bdget
);
574 * bdgrab -- Grab a reference to an already referenced block device
575 * @bdev: Block device to grab a reference to.
577 struct block_device
*bdgrab(struct block_device
*bdev
)
579 atomic_inc(&bdev
->bd_inode
->i_count
);
583 long nr_blockdev_pages(void)
585 struct block_device
*bdev
;
587 spin_lock(&bdev_lock
);
588 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
589 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
591 spin_unlock(&bdev_lock
);
595 void bdput(struct block_device
*bdev
)
597 iput(bdev
->bd_inode
);
600 EXPORT_SYMBOL(bdput
);
602 static struct block_device
*bd_acquire(struct inode
*inode
)
604 struct block_device
*bdev
;
606 spin_lock(&bdev_lock
);
607 bdev
= inode
->i_bdev
;
609 atomic_inc(&bdev
->bd_inode
->i_count
);
610 spin_unlock(&bdev_lock
);
613 spin_unlock(&bdev_lock
);
615 bdev
= bdget(inode
->i_rdev
);
617 spin_lock(&bdev_lock
);
618 if (!inode
->i_bdev
) {
620 * We take an additional bd_inode->i_count for inode,
621 * and it's released in clear_inode() of inode.
622 * So, we can access it via ->i_mapping always
625 atomic_inc(&bdev
->bd_inode
->i_count
);
626 inode
->i_bdev
= bdev
;
627 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
628 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
630 spin_unlock(&bdev_lock
);
635 /* Call when you free inode */
637 void bd_forget(struct inode
*inode
)
639 struct block_device
*bdev
= NULL
;
641 spin_lock(&bdev_lock
);
643 if (!sb_is_blkdev_sb(inode
->i_sb
))
644 bdev
= inode
->i_bdev
;
647 spin_unlock(&bdev_lock
);
650 iput(bdev
->bd_inode
);
653 int bd_claim(struct block_device
*bdev
, void *holder
)
656 spin_lock(&bdev_lock
);
658 /* first decide result */
659 if (bdev
->bd_holder
== holder
)
660 res
= 0; /* already a holder */
661 else if (bdev
->bd_holder
!= NULL
)
662 res
= -EBUSY
; /* held by someone else */
663 else if (bdev
->bd_contains
== bdev
)
664 res
= 0; /* is a whole device which isn't held */
666 else if (bdev
->bd_contains
->bd_holder
== bd_claim
)
667 res
= 0; /* is a partition of a device that is being partitioned */
668 else if (bdev
->bd_contains
->bd_holder
!= NULL
)
669 res
= -EBUSY
; /* is a partition of a held device */
671 res
= 0; /* is a partition of an un-held device */
673 /* now impose change */
675 /* note that for a whole device bd_holders
676 * will be incremented twice, and bd_holder will
677 * be set to bd_claim before being set to holder
679 bdev
->bd_contains
->bd_holders
++;
680 bdev
->bd_contains
->bd_holder
= bd_claim
;
682 bdev
->bd_holder
= holder
;
684 spin_unlock(&bdev_lock
);
688 EXPORT_SYMBOL(bd_claim
);
690 void bd_release(struct block_device
*bdev
)
692 spin_lock(&bdev_lock
);
693 if (!--bdev
->bd_contains
->bd_holders
)
694 bdev
->bd_contains
->bd_holder
= NULL
;
695 if (!--bdev
->bd_holders
)
696 bdev
->bd_holder
= NULL
;
697 spin_unlock(&bdev_lock
);
700 EXPORT_SYMBOL(bd_release
);
704 * Functions for bd_claim_by_kobject / bd_release_from_kobject
706 * If a kobject is passed to bd_claim_by_kobject()
707 * and the kobject has a parent directory,
708 * following symlinks are created:
709 * o from the kobject to the claimed bdev
710 * o from "holders" directory of the bdev to the parent of the kobject
711 * bd_release_from_kobject() removes these symlinks.
714 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
715 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
716 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
717 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
720 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
724 return sysfs_create_link(from
, to
, kobject_name(to
));
727 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
731 sysfs_remove_link(from
, kobject_name(to
));
735 * 'struct bd_holder' contains pointers to kobjects symlinked by
736 * bd_claim_by_kobject.
737 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
740 struct list_head list
; /* chain of holders of the bdev */
741 int count
; /* references from the holder */
742 struct kobject
*sdir
; /* holder object, e.g. "/block/dm-0/slaves" */
743 struct kobject
*hdev
; /* e.g. "/block/dm-0" */
744 struct kobject
*hdir
; /* e.g. "/block/sda/holders" */
745 struct kobject
*sdev
; /* e.g. "/block/sda" */
749 * Get references of related kobjects at once.
750 * Returns 1 on success. 0 on failure.
752 * Should call bd_holder_release_dirs() after successful use.
754 static int bd_holder_grab_dirs(struct block_device
*bdev
,
755 struct bd_holder
*bo
)
760 bo
->sdir
= kobject_get(bo
->sdir
);
764 bo
->hdev
= kobject_get(bo
->sdir
->parent
);
768 bo
->sdev
= kobject_get(&part_to_dev(bdev
->bd_part
)->kobj
);
772 bo
->hdir
= kobject_get(bdev
->bd_part
->holder_dir
);
779 kobject_put(bo
->sdev
);
781 kobject_put(bo
->hdev
);
783 kobject_put(bo
->sdir
);
788 /* Put references of related kobjects at once. */
789 static void bd_holder_release_dirs(struct bd_holder
*bo
)
791 kobject_put(bo
->hdir
);
792 kobject_put(bo
->sdev
);
793 kobject_put(bo
->hdev
);
794 kobject_put(bo
->sdir
);
797 static struct bd_holder
*alloc_bd_holder(struct kobject
*kobj
)
799 struct bd_holder
*bo
;
801 bo
= kzalloc(sizeof(*bo
), GFP_KERNEL
);
811 static void free_bd_holder(struct bd_holder
*bo
)
817 * find_bd_holder - find matching struct bd_holder from the block device
819 * @bdev: struct block device to be searched
820 * @bo: target struct bd_holder
822 * Returns matching entry with @bo in @bdev->bd_holder_list.
823 * If found, increment the reference count and return the pointer.
824 * If not found, returns NULL.
826 static struct bd_holder
*find_bd_holder(struct block_device
*bdev
,
827 struct bd_holder
*bo
)
829 struct bd_holder
*tmp
;
831 list_for_each_entry(tmp
, &bdev
->bd_holder_list
, list
)
832 if (tmp
->sdir
== bo
->sdir
) {
841 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
843 * @bdev: block device to be bd_claimed
844 * @bo: preallocated and initialized by alloc_bd_holder()
846 * Add @bo to @bdev->bd_holder_list, create symlinks.
848 * Returns 0 if symlinks are created.
849 * Returns -ve if something fails.
851 static int add_bd_holder(struct block_device
*bdev
, struct bd_holder
*bo
)
858 if (!bd_holder_grab_dirs(bdev
, bo
))
861 err
= add_symlink(bo
->sdir
, bo
->sdev
);
865 err
= add_symlink(bo
->hdir
, bo
->hdev
);
867 del_symlink(bo
->sdir
, bo
->sdev
);
871 list_add_tail(&bo
->list
, &bdev
->bd_holder_list
);
876 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
878 * @bdev: block device to be bd_claimed
879 * @kobj: holder's kobject
881 * If there is matching entry with @kobj in @bdev->bd_holder_list
882 * and no other bd_claim() from the same kobject,
883 * remove the struct bd_holder from the list, delete symlinks for it.
885 * Returns a pointer to the struct bd_holder when it's removed from the list
886 * and ready to be freed.
887 * Returns NULL if matching claim isn't found or there is other bd_claim()
888 * by the same kobject.
890 static struct bd_holder
*del_bd_holder(struct block_device
*bdev
,
891 struct kobject
*kobj
)
893 struct bd_holder
*bo
;
895 list_for_each_entry(bo
, &bdev
->bd_holder_list
, list
) {
896 if (bo
->sdir
== kobj
) {
898 BUG_ON(bo
->count
< 0);
901 del_symlink(bo
->sdir
, bo
->sdev
);
902 del_symlink(bo
->hdir
, bo
->hdev
);
903 bd_holder_release_dirs(bo
);
914 * bd_claim_by_kobject - bd_claim() with additional kobject signature
916 * @bdev: block device to be claimed
917 * @holder: holder's signature
918 * @kobj: holder's kobject
920 * Do bd_claim() and if it succeeds, create sysfs symlinks between
921 * the bdev and the holder's kobject.
922 * Use bd_release_from_kobject() when relesing the claimed bdev.
924 * Returns 0 on success. (same as bd_claim())
925 * Returns errno on failure.
927 static int bd_claim_by_kobject(struct block_device
*bdev
, void *holder
,
928 struct kobject
*kobj
)
931 struct bd_holder
*bo
, *found
;
936 bo
= alloc_bd_holder(kobj
);
940 mutex_lock(&bdev
->bd_mutex
);
942 err
= bd_claim(bdev
, holder
);
946 found
= find_bd_holder(bdev
, bo
);
950 err
= add_bd_holder(bdev
, bo
);
956 mutex_unlock(&bdev
->bd_mutex
);
962 * bd_release_from_kobject - bd_release() with additional kobject signature
964 * @bdev: block device to be released
965 * @kobj: holder's kobject
967 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
969 static void bd_release_from_kobject(struct block_device
*bdev
,
970 struct kobject
*kobj
)
975 mutex_lock(&bdev
->bd_mutex
);
977 free_bd_holder(del_bd_holder(bdev
, kobj
));
978 mutex_unlock(&bdev
->bd_mutex
);
982 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
984 * @bdev: block device to be claimed
985 * @holder: holder's signature
986 * @disk: holder's gendisk
988 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
990 int bd_claim_by_disk(struct block_device
*bdev
, void *holder
,
991 struct gendisk
*disk
)
993 return bd_claim_by_kobject(bdev
, holder
, kobject_get(disk
->slave_dir
));
995 EXPORT_SYMBOL_GPL(bd_claim_by_disk
);
998 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1000 * @bdev: block device to be claimed
1001 * @disk: holder's gendisk
1003 * Call bd_release_from_kobject() and put @disk->slave_dir.
1005 void bd_release_from_disk(struct block_device
*bdev
, struct gendisk
*disk
)
1007 bd_release_from_kobject(bdev
, disk
->slave_dir
);
1008 kobject_put(disk
->slave_dir
);
1010 EXPORT_SYMBOL_GPL(bd_release_from_disk
);
1014 * Tries to open block device by device number. Use it ONLY if you
1015 * really do not have anything better - i.e. when you are behind a
1016 * truly sucky interface and all you are given is a device number. _Never_
1017 * to be used for internal purposes. If you ever need it - reconsider
1020 struct block_device
*open_by_devnum(dev_t dev
, fmode_t mode
)
1022 struct block_device
*bdev
= bdget(dev
);
1025 err
= blkdev_get(bdev
, mode
);
1026 return err
? ERR_PTR(err
) : bdev
;
1029 EXPORT_SYMBOL(open_by_devnum
);
1032 * flush_disk - invalidates all buffer-cache entries on a disk
1034 * @bdev: struct block device to be flushed
1036 * Invalidates all buffer-cache entries on a disk. It should be called
1037 * when a disk has been changed -- either by a media change or online
1040 static void flush_disk(struct block_device
*bdev
)
1042 if (__invalidate_device(bdev
)) {
1043 char name
[BDEVNAME_SIZE
] = "";
1046 disk_name(bdev
->bd_disk
, 0, name
);
1047 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1048 "resized disk %s\n", name
);
1053 if (disk_partitionable(bdev
->bd_disk
))
1054 bdev
->bd_invalidated
= 1;
1058 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1059 * @disk: struct gendisk to check
1060 * @bdev: struct bdev to adjust.
1062 * This routine checks to see if the bdev size does not match the disk size
1063 * and adjusts it if it differs.
1065 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1067 loff_t disk_size
, bdev_size
;
1069 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1070 bdev_size
= i_size_read(bdev
->bd_inode
);
1071 if (disk_size
!= bdev_size
) {
1072 char name
[BDEVNAME_SIZE
];
1074 disk_name(disk
, 0, name
);
1076 "%s: detected capacity change from %lld to %lld\n",
1077 name
, bdev_size
, disk_size
);
1078 i_size_write(bdev
->bd_inode
, disk_size
);
1082 EXPORT_SYMBOL(check_disk_size_change
);
1085 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1086 * @disk: struct gendisk to be revalidated
1088 * This routine is a wrapper for lower-level driver's revalidate_disk
1089 * call-backs. It is used to do common pre and post operations needed
1090 * for all revalidate_disk operations.
1092 int revalidate_disk(struct gendisk
*disk
)
1094 struct block_device
*bdev
;
1097 if (disk
->fops
->revalidate_disk
)
1098 ret
= disk
->fops
->revalidate_disk(disk
);
1100 bdev
= bdget_disk(disk
, 0);
1104 mutex_lock(&bdev
->bd_mutex
);
1105 check_disk_size_change(disk
, bdev
);
1106 mutex_unlock(&bdev
->bd_mutex
);
1110 EXPORT_SYMBOL(revalidate_disk
);
1113 * This routine checks whether a removable media has been changed,
1114 * and invalidates all buffer-cache-entries in that case. This
1115 * is a relatively slow routine, so we have to try to minimize using
1116 * it. Thus it is called only upon a 'mount' or 'open'. This
1117 * is the best way of combining speed and utility, I think.
1118 * People changing diskettes in the middle of an operation deserve
1121 int check_disk_change(struct block_device
*bdev
)
1123 struct gendisk
*disk
= bdev
->bd_disk
;
1124 const struct block_device_operations
*bdops
= disk
->fops
;
1126 if (!bdops
->media_changed
)
1128 if (!bdops
->media_changed(bdev
->bd_disk
))
1132 if (bdops
->revalidate_disk
)
1133 bdops
->revalidate_disk(bdev
->bd_disk
);
1137 EXPORT_SYMBOL(check_disk_change
);
1139 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1141 unsigned bsize
= bdev_logical_block_size(bdev
);
1143 bdev
->bd_inode
->i_size
= size
;
1144 while (bsize
< PAGE_CACHE_SIZE
) {
1149 bdev
->bd_block_size
= bsize
;
1150 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1152 EXPORT_SYMBOL(bd_set_size
);
1154 static int __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1159 * mutex_lock(part->bd_mutex)
1160 * mutex_lock_nested(whole->bd_mutex, 1)
1163 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1165 struct gendisk
*disk
;
1170 if (mode
& FMODE_READ
)
1172 if (mode
& FMODE_WRITE
)
1175 * hooks: /n/, see "layering violations".
1177 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1187 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1189 goto out_unlock_kernel
;
1191 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1192 if (!bdev
->bd_openers
) {
1193 bdev
->bd_disk
= disk
;
1194 bdev
->bd_contains
= bdev
;
1196 struct backing_dev_info
*bdi
;
1199 bdev
->bd_part
= disk_get_part(disk
, partno
);
1203 if (disk
->fops
->open
) {
1204 ret
= disk
->fops
->open(bdev
, mode
);
1205 if (ret
== -ERESTARTSYS
) {
1206 /* Lost a race with 'disk' being
1207 * deleted, try again.
1210 disk_put_part(bdev
->bd_part
);
1211 bdev
->bd_part
= NULL
;
1212 module_put(disk
->fops
->owner
);
1214 bdev
->bd_disk
= NULL
;
1215 mutex_unlock(&bdev
->bd_mutex
);
1221 if (!bdev
->bd_openers
) {
1222 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1223 bdi
= blk_get_backing_dev_info(bdev
);
1225 bdi
= &default_backing_dev_info
;
1226 bdev
->bd_inode
->i_data
.backing_dev_info
= bdi
;
1228 if (bdev
->bd_invalidated
)
1229 rescan_partitions(disk
, bdev
);
1231 struct block_device
*whole
;
1232 whole
= bdget_disk(disk
, 0);
1237 ret
= __blkdev_get(whole
, mode
, 1);
1240 bdev
->bd_contains
= whole
;
1241 bdev
->bd_inode
->i_data
.backing_dev_info
=
1242 whole
->bd_inode
->i_data
.backing_dev_info
;
1243 bdev
->bd_part
= disk_get_part(disk
, partno
);
1244 if (!(disk
->flags
& GENHD_FL_UP
) ||
1245 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1249 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1252 module_put(disk
->fops
->owner
);
1255 if (bdev
->bd_contains
== bdev
) {
1256 if (bdev
->bd_disk
->fops
->open
) {
1257 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1259 goto out_unlock_bdev
;
1261 if (bdev
->bd_invalidated
)
1262 rescan_partitions(bdev
->bd_disk
, bdev
);
1267 bdev
->bd_part_count
++;
1268 mutex_unlock(&bdev
->bd_mutex
);
1273 disk_put_part(bdev
->bd_part
);
1274 bdev
->bd_disk
= NULL
;
1275 bdev
->bd_part
= NULL
;
1276 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1277 if (bdev
!= bdev
->bd_contains
)
1278 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1279 bdev
->bd_contains
= NULL
;
1281 mutex_unlock(&bdev
->bd_mutex
);
1286 module_put(disk
->fops
->owner
);
1293 int blkdev_get(struct block_device
*bdev
, fmode_t mode
)
1295 return __blkdev_get(bdev
, mode
, 0);
1297 EXPORT_SYMBOL(blkdev_get
);
1299 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1301 struct block_device
*bdev
;
1305 * Preserve backwards compatibility and allow large file access
1306 * even if userspace doesn't ask for it explicitly. Some mkfs
1307 * binary needs it. We might want to drop this workaround
1308 * during an unstable branch.
1310 filp
->f_flags
|= O_LARGEFILE
;
1312 if (filp
->f_flags
& O_NDELAY
)
1313 filp
->f_mode
|= FMODE_NDELAY
;
1314 if (filp
->f_flags
& O_EXCL
)
1315 filp
->f_mode
|= FMODE_EXCL
;
1316 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1317 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1319 bdev
= bd_acquire(inode
);
1323 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1325 res
= blkdev_get(bdev
, filp
->f_mode
);
1329 if (filp
->f_mode
& FMODE_EXCL
) {
1330 res
= bd_claim(bdev
, filp
);
1332 goto out_blkdev_put
;
1338 blkdev_put(bdev
, filp
->f_mode
);
1342 static int __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1345 struct gendisk
*disk
= bdev
->bd_disk
;
1346 struct block_device
*victim
= NULL
;
1348 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1351 bdev
->bd_part_count
--;
1353 if (!--bdev
->bd_openers
) {
1354 sync_blockdev(bdev
);
1357 if (bdev
->bd_contains
== bdev
) {
1358 if (disk
->fops
->release
)
1359 ret
= disk
->fops
->release(disk
, mode
);
1361 if (!bdev
->bd_openers
) {
1362 struct module
*owner
= disk
->fops
->owner
;
1366 disk_put_part(bdev
->bd_part
);
1367 bdev
->bd_part
= NULL
;
1368 bdev
->bd_disk
= NULL
;
1369 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1370 if (bdev
!= bdev
->bd_contains
)
1371 victim
= bdev
->bd_contains
;
1372 bdev
->bd_contains
= NULL
;
1375 mutex_unlock(&bdev
->bd_mutex
);
1378 __blkdev_put(victim
, mode
, 1);
1382 int blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1384 return __blkdev_put(bdev
, mode
, 0);
1386 EXPORT_SYMBOL(blkdev_put
);
1388 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1390 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1391 if (bdev
->bd_holder
== filp
)
1393 return blkdev_put(bdev
, filp
->f_mode
);
1396 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1398 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1399 fmode_t mode
= file
->f_mode
;
1402 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1403 * to updated it before every ioctl.
1405 if (file
->f_flags
& O_NDELAY
)
1406 mode
|= FMODE_NDELAY
;
1408 mode
&= ~FMODE_NDELAY
;
1410 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1414 * Write data to the block device. Only intended for the block device itself
1415 * and the raw driver which basically is a fake block device.
1417 * Does not take i_mutex for the write and thus is not for general purpose
1420 ssize_t
blkdev_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
1421 unsigned long nr_segs
, loff_t pos
)
1423 struct file
*file
= iocb
->ki_filp
;
1426 BUG_ON(iocb
->ki_pos
!= pos
);
1428 ret
= __generic_file_aio_write(iocb
, iov
, nr_segs
, &iocb
->ki_pos
);
1429 if (ret
> 0 || ret
== -EIOCBQUEUED
) {
1432 err
= generic_write_sync(file
, pos
, ret
);
1433 if (err
< 0 && ret
> 0)
1438 EXPORT_SYMBOL_GPL(blkdev_aio_write
);
1441 * Try to release a page associated with block device when the system
1442 * is under memory pressure.
1444 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1446 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1448 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1449 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1451 return try_to_free_buffers(page
);
1454 static const struct address_space_operations def_blk_aops
= {
1455 .readpage
= blkdev_readpage
,
1456 .writepage
= blkdev_writepage
,
1457 .sync_page
= block_sync_page
,
1458 .write_begin
= blkdev_write_begin
,
1459 .write_end
= blkdev_write_end
,
1460 .writepages
= generic_writepages
,
1461 .releasepage
= blkdev_releasepage
,
1462 .direct_IO
= blkdev_direct_IO
,
1465 const struct file_operations def_blk_fops
= {
1466 .open
= blkdev_open
,
1467 .release
= blkdev_close
,
1468 .llseek
= block_llseek
,
1469 .read
= do_sync_read
,
1470 .write
= do_sync_write
,
1471 .aio_read
= generic_file_aio_read
,
1472 .aio_write
= blkdev_aio_write
,
1473 .mmap
= generic_file_mmap
,
1474 .fsync
= block_fsync
,
1475 .unlocked_ioctl
= block_ioctl
,
1476 #ifdef CONFIG_COMPAT
1477 .compat_ioctl
= compat_blkdev_ioctl
,
1479 .splice_read
= generic_file_splice_read
,
1480 .splice_write
= generic_file_splice_write
,
1483 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1486 mm_segment_t old_fs
= get_fs();
1488 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1493 EXPORT_SYMBOL(ioctl_by_bdev
);
1496 * lookup_bdev - lookup a struct block_device by name
1497 * @pathname: special file representing the block device
1499 * Get a reference to the blockdevice at @pathname in the current
1500 * namespace if possible and return it. Return ERR_PTR(error)
1503 struct block_device
*lookup_bdev(const char *pathname
)
1505 struct block_device
*bdev
;
1506 struct inode
*inode
;
1510 if (!pathname
|| !*pathname
)
1511 return ERR_PTR(-EINVAL
);
1513 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1515 return ERR_PTR(error
);
1517 inode
= path
.dentry
->d_inode
;
1519 if (!S_ISBLK(inode
->i_mode
))
1522 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1525 bdev
= bd_acquire(inode
);
1532 bdev
= ERR_PTR(error
);
1535 EXPORT_SYMBOL(lookup_bdev
);
1538 * open_bdev_exclusive - open a block device by name and set it up for use
1540 * @path: special file representing the block device
1541 * @mode: FMODE_... combination to pass be used
1542 * @holder: owner for exclusion
1544 * Open the blockdevice described by the special file at @path, claim it
1547 struct block_device
*open_bdev_exclusive(const char *path
, fmode_t mode
, void *holder
)
1549 struct block_device
*bdev
;
1552 bdev
= lookup_bdev(path
);
1556 error
= blkdev_get(bdev
, mode
);
1558 return ERR_PTR(error
);
1560 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
))
1562 error
= bd_claim(bdev
, holder
);
1569 blkdev_put(bdev
, mode
);
1570 return ERR_PTR(error
);
1573 EXPORT_SYMBOL(open_bdev_exclusive
);
1576 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1578 * @bdev: blockdevice to close
1579 * @mode: mode, must match that used to open.
1581 * This is the counterpart to open_bdev_exclusive().
1583 void close_bdev_exclusive(struct block_device
*bdev
, fmode_t mode
)
1586 blkdev_put(bdev
, mode
);
1589 EXPORT_SYMBOL(close_bdev_exclusive
);
1591 int __invalidate_device(struct block_device
*bdev
)
1593 struct super_block
*sb
= get_super(bdev
);
1598 * no need to lock the super, get_super holds the
1599 * read mutex so the filesystem cannot go away
1600 * under us (->put_super runs with the write lock
1603 shrink_dcache_sb(sb
);
1604 res
= invalidate_inodes(sb
);
1607 invalidate_bdev(bdev
);
1610 EXPORT_SYMBOL(__invalidate_device
);