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/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/mpage.h>
21 #include <linux/mount.h>
22 #include <linux/uio.h>
23 #include <linux/namei.h>
24 #include <asm/uaccess.h>
27 struct block_device bdev
;
28 struct inode vfs_inode
;
31 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
33 return container_of(inode
, struct bdev_inode
, vfs_inode
);
36 inline struct block_device
*I_BDEV(struct inode
*inode
)
38 return &BDEV_I(inode
)->bdev
;
41 EXPORT_SYMBOL(I_BDEV
);
43 static sector_t
max_block(struct block_device
*bdev
)
45 sector_t retval
= ~((sector_t
)0);
46 loff_t sz
= i_size_read(bdev
->bd_inode
);
49 unsigned int size
= block_size(bdev
);
50 unsigned int sizebits
= blksize_bits(size
);
51 retval
= (sz
>> sizebits
);
56 /* Kill _all_ buffers, dirty or not.. */
57 static void kill_bdev(struct block_device
*bdev
)
59 invalidate_bdev(bdev
, 1);
60 truncate_inode_pages(bdev
->bd_inode
->i_mapping
, 0);
63 int set_blocksize(struct block_device
*bdev
, int size
)
65 /* Size must be a power of two, and between 512 and PAGE_SIZE */
66 if (size
> PAGE_SIZE
|| size
< 512 || (size
& (size
-1)))
69 /* Size cannot be smaller than the size supported by the device */
70 if (size
< bdev_hardsect_size(bdev
))
73 /* Don't change the size if it is same as current */
74 if (bdev
->bd_block_size
!= size
) {
76 bdev
->bd_block_size
= size
;
77 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
83 EXPORT_SYMBOL(set_blocksize
);
85 int sb_set_blocksize(struct super_block
*sb
, int size
)
87 if (set_blocksize(sb
->s_bdev
, size
))
89 /* If we get here, we know size is power of two
90 * and it's value is between 512 and PAGE_SIZE */
91 sb
->s_blocksize
= size
;
92 sb
->s_blocksize_bits
= blksize_bits(size
);
93 return sb
->s_blocksize
;
96 EXPORT_SYMBOL(sb_set_blocksize
);
98 int sb_min_blocksize(struct super_block
*sb
, int size
)
100 int minsize
= bdev_hardsect_size(sb
->s_bdev
);
103 return sb_set_blocksize(sb
, size
);
106 EXPORT_SYMBOL(sb_min_blocksize
);
109 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
110 struct buffer_head
*bh
, int create
)
112 if (iblock
>= max_block(I_BDEV(inode
))) {
117 * for reads, we're just trying to fill a partial page.
118 * return a hole, they will have to call get_block again
119 * before they can fill it, and they will get -EIO at that
124 bh
->b_bdev
= I_BDEV(inode
);
125 bh
->b_blocknr
= iblock
;
126 set_buffer_mapped(bh
);
131 blkdev_get_blocks(struct inode
*inode
, sector_t iblock
,
132 struct buffer_head
*bh
, int create
)
134 sector_t end_block
= max_block(I_BDEV(inode
));
135 unsigned long max_blocks
= bh
->b_size
>> inode
->i_blkbits
;
137 if ((iblock
+ max_blocks
) > end_block
) {
138 max_blocks
= end_block
- iblock
;
139 if ((long)max_blocks
<= 0) {
141 return -EIO
; /* write fully beyond EOF */
143 * It is a read which is fully beyond EOF. We return
144 * a !buffer_mapped buffer
150 bh
->b_bdev
= I_BDEV(inode
);
151 bh
->b_blocknr
= iblock
;
152 bh
->b_size
= max_blocks
<< inode
->i_blkbits
;
154 set_buffer_mapped(bh
);
159 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
160 loff_t offset
, unsigned long nr_segs
)
162 struct file
*file
= iocb
->ki_filp
;
163 struct inode
*inode
= file
->f_mapping
->host
;
165 return blockdev_direct_IO_no_locking(rw
, iocb
, inode
, I_BDEV(inode
),
166 iov
, offset
, nr_segs
, blkdev_get_blocks
, NULL
);
169 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
171 return block_write_full_page(page
, blkdev_get_block
, wbc
);
174 static int blkdev_readpage(struct file
* file
, struct page
* page
)
176 return block_read_full_page(page
, blkdev_get_block
);
179 static int blkdev_prepare_write(struct file
*file
, struct page
*page
, unsigned from
, unsigned to
)
181 return block_prepare_write(page
, from
, to
, blkdev_get_block
);
184 static int blkdev_commit_write(struct file
*file
, struct page
*page
, unsigned from
, unsigned to
)
186 return block_commit_write(page
, from
, to
);
191 * for a block special file file->f_dentry->d_inode->i_size is zero
192 * so we compute the size by hand (just as in block_read/write above)
194 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int origin
)
196 struct inode
*bd_inode
= file
->f_mapping
->host
;
200 mutex_lock(&bd_inode
->i_mutex
);
201 size
= i_size_read(bd_inode
);
208 offset
+= file
->f_pos
;
211 if (offset
>= 0 && offset
<= size
) {
212 if (offset
!= file
->f_pos
) {
213 file
->f_pos
= offset
;
217 mutex_unlock(&bd_inode
->i_mutex
);
222 * Filp is never NULL; the only case when ->fsync() is called with
223 * NULL first argument is nfsd_sync_dir() and that's not a directory.
226 static int block_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
228 return sync_blockdev(I_BDEV(filp
->f_mapping
->host
));
235 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
236 static kmem_cache_t
* bdev_cachep __read_mostly
;
238 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
240 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, SLAB_KERNEL
);
243 return &ei
->vfs_inode
;
246 static void bdev_destroy_inode(struct inode
*inode
)
248 struct bdev_inode
*bdi
= BDEV_I(inode
);
250 bdi
->bdev
.bd_inode_backing_dev_info
= NULL
;
251 kmem_cache_free(bdev_cachep
, bdi
);
254 static void init_once(void * foo
, kmem_cache_t
* cachep
, unsigned long flags
)
256 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
257 struct block_device
*bdev
= &ei
->bdev
;
259 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) ==
260 SLAB_CTOR_CONSTRUCTOR
)
262 memset(bdev
, 0, sizeof(*bdev
));
263 mutex_init(&bdev
->bd_mutex
);
264 mutex_init(&bdev
->bd_mount_mutex
);
265 INIT_LIST_HEAD(&bdev
->bd_inodes
);
266 INIT_LIST_HEAD(&bdev
->bd_list
);
268 INIT_LIST_HEAD(&bdev
->bd_holder_list
);
270 inode_init_once(&ei
->vfs_inode
);
274 static inline void __bd_forget(struct inode
*inode
)
276 list_del_init(&inode
->i_devices
);
277 inode
->i_bdev
= NULL
;
278 inode
->i_mapping
= &inode
->i_data
;
281 static void bdev_clear_inode(struct inode
*inode
)
283 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
285 spin_lock(&bdev_lock
);
286 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
287 __bd_forget(list_entry(p
, struct inode
, i_devices
));
289 list_del_init(&bdev
->bd_list
);
290 spin_unlock(&bdev_lock
);
293 static struct super_operations bdev_sops
= {
294 .statfs
= simple_statfs
,
295 .alloc_inode
= bdev_alloc_inode
,
296 .destroy_inode
= bdev_destroy_inode
,
297 .drop_inode
= generic_delete_inode
,
298 .clear_inode
= bdev_clear_inode
,
301 static int bd_get_sb(struct file_system_type
*fs_type
,
302 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
304 return get_sb_pseudo(fs_type
, "bdev:", &bdev_sops
, 0x62646576, mnt
);
307 static struct file_system_type bd_type
= {
310 .kill_sb
= kill_anon_super
,
313 static struct vfsmount
*bd_mnt __read_mostly
;
314 struct super_block
*blockdev_superblock
;
316 void __init
bdev_cache_init(void)
319 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
320 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
321 SLAB_MEM_SPREAD
|SLAB_PANIC
),
323 err
= register_filesystem(&bd_type
);
325 panic("Cannot register bdev pseudo-fs");
326 bd_mnt
= kern_mount(&bd_type
);
327 err
= PTR_ERR(bd_mnt
);
329 panic("Cannot create bdev pseudo-fs");
330 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
334 * Most likely _very_ bad one - but then it's hardly critical for small
335 * /dev and can be fixed when somebody will need really large one.
336 * Keep in mind that it will be fed through icache hash function too.
338 static inline unsigned long hash(dev_t dev
)
340 return MAJOR(dev
)+MINOR(dev
);
343 static int bdev_test(struct inode
*inode
, void *data
)
345 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
348 static int bdev_set(struct inode
*inode
, void *data
)
350 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
354 static LIST_HEAD(all_bdevs
);
356 struct block_device
*bdget(dev_t dev
)
358 struct block_device
*bdev
;
361 inode
= iget5_locked(bd_mnt
->mnt_sb
, hash(dev
),
362 bdev_test
, bdev_set
, &dev
);
367 bdev
= &BDEV_I(inode
)->bdev
;
369 if (inode
->i_state
& I_NEW
) {
370 bdev
->bd_contains
= NULL
;
371 bdev
->bd_inode
= inode
;
372 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
373 bdev
->bd_part_count
= 0;
374 bdev
->bd_invalidated
= 0;
375 inode
->i_mode
= S_IFBLK
;
377 inode
->i_bdev
= bdev
;
378 inode
->i_data
.a_ops
= &def_blk_aops
;
379 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
380 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
381 spin_lock(&bdev_lock
);
382 list_add(&bdev
->bd_list
, &all_bdevs
);
383 spin_unlock(&bdev_lock
);
384 unlock_new_inode(inode
);
389 EXPORT_SYMBOL(bdget
);
391 long nr_blockdev_pages(void)
395 spin_lock(&bdev_lock
);
396 list_for_each(p
, &all_bdevs
) {
397 struct block_device
*bdev
;
398 bdev
= list_entry(p
, struct block_device
, bd_list
);
399 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
401 spin_unlock(&bdev_lock
);
405 void bdput(struct block_device
*bdev
)
407 iput(bdev
->bd_inode
);
410 EXPORT_SYMBOL(bdput
);
412 static struct block_device
*bd_acquire(struct inode
*inode
)
414 struct block_device
*bdev
;
416 spin_lock(&bdev_lock
);
417 bdev
= inode
->i_bdev
;
419 atomic_inc(&bdev
->bd_inode
->i_count
);
420 spin_unlock(&bdev_lock
);
423 spin_unlock(&bdev_lock
);
425 bdev
= bdget(inode
->i_rdev
);
427 spin_lock(&bdev_lock
);
428 if (!inode
->i_bdev
) {
430 * We take an additional bd_inode->i_count for inode,
431 * and it's released in clear_inode() of inode.
432 * So, we can access it via ->i_mapping always
435 atomic_inc(&bdev
->bd_inode
->i_count
);
436 inode
->i_bdev
= bdev
;
437 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
438 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
440 spin_unlock(&bdev_lock
);
445 /* Call when you free inode */
447 void bd_forget(struct inode
*inode
)
449 struct block_device
*bdev
= NULL
;
451 spin_lock(&bdev_lock
);
453 if (inode
->i_sb
!= blockdev_superblock
)
454 bdev
= inode
->i_bdev
;
457 spin_unlock(&bdev_lock
);
460 iput(bdev
->bd_inode
);
463 int bd_claim(struct block_device
*bdev
, void *holder
)
466 spin_lock(&bdev_lock
);
468 /* first decide result */
469 if (bdev
->bd_holder
== holder
)
470 res
= 0; /* already a holder */
471 else if (bdev
->bd_holder
!= NULL
)
472 res
= -EBUSY
; /* held by someone else */
473 else if (bdev
->bd_contains
== bdev
)
474 res
= 0; /* is a whole device which isn't held */
476 else if (bdev
->bd_contains
->bd_holder
== bd_claim
)
477 res
= 0; /* is a partition of a device that is being partitioned */
478 else if (bdev
->bd_contains
->bd_holder
!= NULL
)
479 res
= -EBUSY
; /* is a partition of a held device */
481 res
= 0; /* is a partition of an un-held device */
483 /* now impose change */
485 /* note that for a whole device bd_holders
486 * will be incremented twice, and bd_holder will
487 * be set to bd_claim before being set to holder
489 bdev
->bd_contains
->bd_holders
++;
490 bdev
->bd_contains
->bd_holder
= bd_claim
;
492 bdev
->bd_holder
= holder
;
494 spin_unlock(&bdev_lock
);
498 EXPORT_SYMBOL(bd_claim
);
500 void bd_release(struct block_device
*bdev
)
502 spin_lock(&bdev_lock
);
503 if (!--bdev
->bd_contains
->bd_holders
)
504 bdev
->bd_contains
->bd_holder
= NULL
;
505 if (!--bdev
->bd_holders
)
506 bdev
->bd_holder
= NULL
;
507 spin_unlock(&bdev_lock
);
510 EXPORT_SYMBOL(bd_release
);
514 * Functions for bd_claim_by_kobject / bd_release_from_kobject
516 * If a kobject is passed to bd_claim_by_kobject()
517 * and the kobject has a parent directory,
518 * following symlinks are created:
519 * o from the kobject to the claimed bdev
520 * o from "holders" directory of the bdev to the parent of the kobject
521 * bd_release_from_kobject() removes these symlinks.
524 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
525 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
526 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
527 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
530 static struct kobject
*bdev_get_kobj(struct block_device
*bdev
)
532 if (bdev
->bd_contains
!= bdev
)
533 return kobject_get(&bdev
->bd_part
->kobj
);
535 return kobject_get(&bdev
->bd_disk
->kobj
);
538 static struct kobject
*bdev_get_holder(struct block_device
*bdev
)
540 if (bdev
->bd_contains
!= bdev
)
541 return kobject_get(bdev
->bd_part
->holder_dir
);
543 return kobject_get(bdev
->bd_disk
->holder_dir
);
546 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
550 return sysfs_create_link(from
, to
, kobject_name(to
));
553 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
557 sysfs_remove_link(from
, kobject_name(to
));
561 * 'struct bd_holder' contains pointers to kobjects symlinked by
562 * bd_claim_by_kobject.
563 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
566 struct list_head list
; /* chain of holders of the bdev */
567 int count
; /* references from the holder */
568 struct kobject
*sdir
; /* holder object, e.g. "/block/dm-0/slaves" */
569 struct kobject
*hdev
; /* e.g. "/block/dm-0" */
570 struct kobject
*hdir
; /* e.g. "/block/sda/holders" */
571 struct kobject
*sdev
; /* e.g. "/block/sda" */
575 * Get references of related kobjects at once.
576 * Returns 1 on success. 0 on failure.
578 * Should call bd_holder_release_dirs() after successful use.
580 static int bd_holder_grab_dirs(struct block_device
*bdev
,
581 struct bd_holder
*bo
)
586 bo
->sdir
= kobject_get(bo
->sdir
);
590 bo
->hdev
= kobject_get(bo
->sdir
->parent
);
594 bo
->sdev
= bdev_get_kobj(bdev
);
598 bo
->hdir
= bdev_get_holder(bdev
);
605 kobject_put(bo
->sdev
);
607 kobject_put(bo
->hdev
);
609 kobject_put(bo
->sdir
);
614 /* Put references of related kobjects at once. */
615 static void bd_holder_release_dirs(struct bd_holder
*bo
)
617 kobject_put(bo
->hdir
);
618 kobject_put(bo
->sdev
);
619 kobject_put(bo
->hdev
);
620 kobject_put(bo
->sdir
);
623 static struct bd_holder
*alloc_bd_holder(struct kobject
*kobj
)
625 struct bd_holder
*bo
;
627 bo
= kzalloc(sizeof(*bo
), GFP_KERNEL
);
637 static void free_bd_holder(struct bd_holder
*bo
)
643 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
645 * @bdev: block device to be bd_claimed
646 * @bo: preallocated and initialized by alloc_bd_holder()
648 * If there is no matching entry with @bo in @bdev->bd_holder_list,
649 * add @bo to the list, create symlinks.
651 * Returns 0 if symlinks are created or already there.
652 * Returns -ve if something fails and @bo can be freed.
654 static int add_bd_holder(struct block_device
*bdev
, struct bd_holder
*bo
)
656 struct bd_holder
*tmp
;
662 list_for_each_entry(tmp
, &bdev
->bd_holder_list
, list
) {
663 if (tmp
->sdir
== bo
->sdir
) {
665 /* We've already done what we need to do here. */
671 if (!bd_holder_grab_dirs(bdev
, bo
))
674 ret
= add_symlink(bo
->sdir
, bo
->sdev
);
676 ret
= add_symlink(bo
->hdir
, bo
->hdev
);
678 del_symlink(bo
->sdir
, bo
->sdev
);
681 list_add_tail(&bo
->list
, &bdev
->bd_holder_list
);
686 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
688 * @bdev: block device to be bd_claimed
689 * @kobj: holder's kobject
691 * If there is matching entry with @kobj in @bdev->bd_holder_list
692 * and no other bd_claim() from the same kobject,
693 * remove the struct bd_holder from the list, delete symlinks for it.
695 * Returns a pointer to the struct bd_holder when it's removed from the list
696 * and ready to be freed.
697 * Returns NULL if matching claim isn't found or there is other bd_claim()
698 * by the same kobject.
700 static struct bd_holder
*del_bd_holder(struct block_device
*bdev
,
701 struct kobject
*kobj
)
703 struct bd_holder
*bo
;
705 list_for_each_entry(bo
, &bdev
->bd_holder_list
, list
) {
706 if (bo
->sdir
== kobj
) {
708 BUG_ON(bo
->count
< 0);
711 del_symlink(bo
->sdir
, bo
->sdev
);
712 del_symlink(bo
->hdir
, bo
->hdev
);
713 bd_holder_release_dirs(bo
);
724 * bd_claim_by_kobject - bd_claim() with additional kobject signature
726 * @bdev: block device to be claimed
727 * @holder: holder's signature
728 * @kobj: holder's kobject
730 * Do bd_claim() and if it succeeds, create sysfs symlinks between
731 * the bdev and the holder's kobject.
732 * Use bd_release_from_kobject() when relesing the claimed bdev.
734 * Returns 0 on success. (same as bd_claim())
735 * Returns errno on failure.
737 static int bd_claim_by_kobject(struct block_device
*bdev
, void *holder
,
738 struct kobject
*kobj
)
741 struct bd_holder
*bo
;
746 bo
= alloc_bd_holder(kobj
);
750 mutex_lock_nested(&bdev
->bd_mutex
, BD_MUTEX_PARTITION
);
751 res
= bd_claim(bdev
, holder
);
753 res
= add_bd_holder(bdev
, bo
);
756 mutex_unlock(&bdev
->bd_mutex
);
762 * bd_release_from_kobject - bd_release() with additional kobject signature
764 * @bdev: block device to be released
765 * @kobj: holder's kobject
767 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
769 static void bd_release_from_kobject(struct block_device
*bdev
,
770 struct kobject
*kobj
)
772 struct bd_holder
*bo
;
777 mutex_lock_nested(&bdev
->bd_mutex
, BD_MUTEX_PARTITION
);
779 if ((bo
= del_bd_holder(bdev
, kobj
)))
781 mutex_unlock(&bdev
->bd_mutex
);
785 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
787 * @bdev: block device to be claimed
788 * @holder: holder's signature
789 * @disk: holder's gendisk
791 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
793 int bd_claim_by_disk(struct block_device
*bdev
, void *holder
,
794 struct gendisk
*disk
)
796 return bd_claim_by_kobject(bdev
, holder
, kobject_get(disk
->slave_dir
));
798 EXPORT_SYMBOL_GPL(bd_claim_by_disk
);
801 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
803 * @bdev: block device to be claimed
804 * @disk: holder's gendisk
806 * Call bd_release_from_kobject() and put @disk->slave_dir.
808 void bd_release_from_disk(struct block_device
*bdev
, struct gendisk
*disk
)
810 bd_release_from_kobject(bdev
, disk
->slave_dir
);
811 kobject_put(disk
->slave_dir
);
813 EXPORT_SYMBOL_GPL(bd_release_from_disk
);
817 * Tries to open block device by device number. Use it ONLY if you
818 * really do not have anything better - i.e. when you are behind a
819 * truly sucky interface and all you are given is a device number. _Never_
820 * to be used for internal purposes. If you ever need it - reconsider
823 struct block_device
*open_by_devnum(dev_t dev
, unsigned mode
)
825 struct block_device
*bdev
= bdget(dev
);
827 int flags
= mode
& FMODE_WRITE
? O_RDWR
: O_RDONLY
;
829 err
= blkdev_get(bdev
, mode
, flags
);
830 return err
? ERR_PTR(err
) : bdev
;
833 EXPORT_SYMBOL(open_by_devnum
);
836 blkdev_get_partition(struct block_device
*bdev
, mode_t mode
, unsigned flags
);
838 struct block_device
*open_partition_by_devnum(dev_t dev
, unsigned mode
)
840 struct block_device
*bdev
= bdget(dev
);
842 int flags
= mode
& FMODE_WRITE
? O_RDWR
: O_RDONLY
;
844 err
= blkdev_get_partition(bdev
, mode
, flags
);
845 return err
? ERR_PTR(err
) : bdev
;
848 EXPORT_SYMBOL(open_partition_by_devnum
);
852 * This routine checks whether a removable media has been changed,
853 * and invalidates all buffer-cache-entries in that case. This
854 * is a relatively slow routine, so we have to try to minimize using
855 * it. Thus it is called only upon a 'mount' or 'open'. This
856 * is the best way of combining speed and utility, I think.
857 * People changing diskettes in the middle of an operation deserve
860 int check_disk_change(struct block_device
*bdev
)
862 struct gendisk
*disk
= bdev
->bd_disk
;
863 struct block_device_operations
* bdops
= disk
->fops
;
865 if (!bdops
->media_changed
)
867 if (!bdops
->media_changed(bdev
->bd_disk
))
870 if (__invalidate_device(bdev
))
871 printk("VFS: busy inodes on changed media.\n");
873 if (bdops
->revalidate_disk
)
874 bdops
->revalidate_disk(bdev
->bd_disk
);
875 if (bdev
->bd_disk
->minors
> 1)
876 bdev
->bd_invalidated
= 1;
880 EXPORT_SYMBOL(check_disk_change
);
882 void bd_set_size(struct block_device
*bdev
, loff_t size
)
884 unsigned bsize
= bdev_hardsect_size(bdev
);
886 bdev
->bd_inode
->i_size
= size
;
887 while (bsize
< PAGE_CACHE_SIZE
) {
892 bdev
->bd_block_size
= bsize
;
893 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
895 EXPORT_SYMBOL(bd_set_size
);
897 static int __blkdev_put(struct block_device
*bdev
, unsigned int subclass
)
900 struct inode
*bd_inode
= bdev
->bd_inode
;
901 struct gendisk
*disk
= bdev
->bd_disk
;
903 mutex_lock_nested(&bdev
->bd_mutex
, subclass
);
905 if (!--bdev
->bd_openers
) {
909 if (bdev
->bd_contains
== bdev
) {
910 if (disk
->fops
->release
)
911 ret
= disk
->fops
->release(bd_inode
, NULL
);
913 mutex_lock_nested(&bdev
->bd_contains
->bd_mutex
,
915 bdev
->bd_contains
->bd_part_count
--;
916 mutex_unlock(&bdev
->bd_contains
->bd_mutex
);
918 if (!bdev
->bd_openers
) {
919 struct module
*owner
= disk
->fops
->owner
;
924 if (bdev
->bd_contains
!= bdev
) {
925 kobject_put(&bdev
->bd_part
->kobj
);
926 bdev
->bd_part
= NULL
;
928 bdev
->bd_disk
= NULL
;
929 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
930 if (bdev
!= bdev
->bd_contains
)
931 __blkdev_put(bdev
->bd_contains
, subclass
+ 1);
932 bdev
->bd_contains
= NULL
;
935 mutex_unlock(&bdev
->bd_mutex
);
940 int blkdev_put(struct block_device
*bdev
)
942 return __blkdev_put(bdev
, BD_MUTEX_NORMAL
);
944 EXPORT_SYMBOL(blkdev_put
);
946 int blkdev_put_partition(struct block_device
*bdev
)
948 return __blkdev_put(bdev
, BD_MUTEX_PARTITION
);
950 EXPORT_SYMBOL(blkdev_put_partition
);
953 blkdev_get_whole(struct block_device
*bdev
, mode_t mode
, unsigned flags
);
956 do_open(struct block_device
*bdev
, struct file
*file
, unsigned int subclass
)
958 struct module
*owner
= NULL
;
959 struct gendisk
*disk
;
963 file
->f_mapping
= bdev
->bd_inode
->i_mapping
;
965 disk
= get_gendisk(bdev
->bd_dev
, &part
);
971 owner
= disk
->fops
->owner
;
973 mutex_lock_nested(&bdev
->bd_mutex
, subclass
);
975 if (!bdev
->bd_openers
) {
976 bdev
->bd_disk
= disk
;
977 bdev
->bd_contains
= bdev
;
979 struct backing_dev_info
*bdi
;
980 if (disk
->fops
->open
) {
981 ret
= disk
->fops
->open(bdev
->bd_inode
, file
);
985 if (!bdev
->bd_openers
) {
986 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
987 bdi
= blk_get_backing_dev_info(bdev
);
989 bdi
= &default_backing_dev_info
;
990 bdev
->bd_inode
->i_data
.backing_dev_info
= bdi
;
992 if (bdev
->bd_invalidated
)
993 rescan_partitions(disk
, bdev
);
996 struct block_device
*whole
;
997 whole
= bdget_disk(disk
, 0);
1001 ret
= blkdev_get_whole(whole
, file
->f_mode
, file
->f_flags
);
1004 bdev
->bd_contains
= whole
;
1005 mutex_lock_nested(&whole
->bd_mutex
, BD_MUTEX_WHOLE
);
1006 whole
->bd_part_count
++;
1007 p
= disk
->part
[part
- 1];
1008 bdev
->bd_inode
->i_data
.backing_dev_info
=
1009 whole
->bd_inode
->i_data
.backing_dev_info
;
1010 if (!(disk
->flags
& GENHD_FL_UP
) || !p
|| !p
->nr_sects
) {
1011 whole
->bd_part_count
--;
1012 mutex_unlock(&whole
->bd_mutex
);
1016 kobject_get(&p
->kobj
);
1018 bd_set_size(bdev
, (loff_t
) p
->nr_sects
<< 9);
1019 mutex_unlock(&whole
->bd_mutex
);
1024 if (bdev
->bd_contains
== bdev
) {
1025 if (bdev
->bd_disk
->fops
->open
) {
1026 ret
= bdev
->bd_disk
->fops
->open(bdev
->bd_inode
, file
);
1030 if (bdev
->bd_invalidated
)
1031 rescan_partitions(bdev
->bd_disk
, bdev
);
1033 mutex_lock_nested(&bdev
->bd_contains
->bd_mutex
,
1034 BD_MUTEX_PARTITION
);
1035 bdev
->bd_contains
->bd_part_count
++;
1036 mutex_unlock(&bdev
->bd_contains
->bd_mutex
);
1040 mutex_unlock(&bdev
->bd_mutex
);
1045 bdev
->bd_disk
= NULL
;
1046 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1047 if (bdev
!= bdev
->bd_contains
)
1048 __blkdev_put(bdev
->bd_contains
, BD_MUTEX_WHOLE
);
1049 bdev
->bd_contains
= NULL
;
1053 mutex_unlock(&bdev
->bd_mutex
);
1060 int blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
)
1063 * This crockload is due to bad choice of ->open() type.
1065 * For now, block device ->open() routine must _not_
1066 * examine anything in 'inode' argument except ->i_rdev.
1068 struct file fake_file
= {};
1069 struct dentry fake_dentry
= {};
1070 fake_file
.f_mode
= mode
;
1071 fake_file
.f_flags
= flags
;
1072 fake_file
.f_dentry
= &fake_dentry
;
1073 fake_dentry
.d_inode
= bdev
->bd_inode
;
1075 return do_open(bdev
, &fake_file
, BD_MUTEX_NORMAL
);
1078 EXPORT_SYMBOL(blkdev_get
);
1081 blkdev_get_whole(struct block_device
*bdev
, mode_t mode
, unsigned flags
)
1084 * This crockload is due to bad choice of ->open() type.
1086 * For now, block device ->open() routine must _not_
1087 * examine anything in 'inode' argument except ->i_rdev.
1089 struct file fake_file
= {};
1090 struct dentry fake_dentry
= {};
1091 fake_file
.f_mode
= mode
;
1092 fake_file
.f_flags
= flags
;
1093 fake_file
.f_dentry
= &fake_dentry
;
1094 fake_dentry
.d_inode
= bdev
->bd_inode
;
1096 return do_open(bdev
, &fake_file
, BD_MUTEX_WHOLE
);
1100 blkdev_get_partition(struct block_device
*bdev
, mode_t mode
, unsigned flags
)
1103 * This crockload is due to bad choice of ->open() type.
1105 * For now, block device ->open() routine must _not_
1106 * examine anything in 'inode' argument except ->i_rdev.
1108 struct file fake_file
= {};
1109 struct dentry fake_dentry
= {};
1110 fake_file
.f_mode
= mode
;
1111 fake_file
.f_flags
= flags
;
1112 fake_file
.f_dentry
= &fake_dentry
;
1113 fake_dentry
.d_inode
= bdev
->bd_inode
;
1115 return do_open(bdev
, &fake_file
, BD_MUTEX_PARTITION
);
1118 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1120 struct block_device
*bdev
;
1124 * Preserve backwards compatibility and allow large file access
1125 * even if userspace doesn't ask for it explicitly. Some mkfs
1126 * binary needs it. We might want to drop this workaround
1127 * during an unstable branch.
1129 filp
->f_flags
|= O_LARGEFILE
;
1131 bdev
= bd_acquire(inode
);
1133 res
= do_open(bdev
, filp
, BD_MUTEX_NORMAL
);
1137 if (!(filp
->f_flags
& O_EXCL
) )
1140 if (!(res
= bd_claim(bdev
, filp
)))
1147 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1149 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1150 if (bdev
->bd_holder
== filp
)
1152 return blkdev_put(bdev
);
1155 static ssize_t
blkdev_file_write(struct file
*file
, const char __user
*buf
,
1156 size_t count
, loff_t
*ppos
)
1158 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= count
};
1160 return generic_file_write_nolock(file
, &local_iov
, 1, ppos
);
1163 static ssize_t
blkdev_file_aio_write(struct kiocb
*iocb
, const char __user
*buf
,
1164 size_t count
, loff_t pos
)
1166 struct iovec local_iov
= { .iov_base
= (void __user
*)buf
, .iov_len
= count
};
1168 return generic_file_aio_write_nolock(iocb
, &local_iov
, 1, &iocb
->ki_pos
);
1171 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1173 return blkdev_ioctl(file
->f_mapping
->host
, file
, cmd
, arg
);
1176 const struct address_space_operations def_blk_aops
= {
1177 .readpage
= blkdev_readpage
,
1178 .writepage
= blkdev_writepage
,
1179 .sync_page
= block_sync_page
,
1180 .prepare_write
= blkdev_prepare_write
,
1181 .commit_write
= blkdev_commit_write
,
1182 .writepages
= generic_writepages
,
1183 .direct_IO
= blkdev_direct_IO
,
1186 const struct file_operations def_blk_fops
= {
1187 .open
= blkdev_open
,
1188 .release
= blkdev_close
,
1189 .llseek
= block_llseek
,
1190 .read
= generic_file_read
,
1191 .write
= blkdev_file_write
,
1192 .aio_read
= generic_file_aio_read
,
1193 .aio_write
= blkdev_file_aio_write
,
1194 .mmap
= generic_file_mmap
,
1195 .fsync
= block_fsync
,
1196 .unlocked_ioctl
= block_ioctl
,
1197 #ifdef CONFIG_COMPAT
1198 .compat_ioctl
= compat_blkdev_ioctl
,
1200 .readv
= generic_file_readv
,
1201 .writev
= generic_file_write_nolock
,
1202 .sendfile
= generic_file_sendfile
,
1203 .splice_read
= generic_file_splice_read
,
1204 .splice_write
= generic_file_splice_write
,
1207 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1210 mm_segment_t old_fs
= get_fs();
1212 res
= blkdev_ioctl(bdev
->bd_inode
, NULL
, cmd
, arg
);
1217 EXPORT_SYMBOL(ioctl_by_bdev
);
1220 * lookup_bdev - lookup a struct block_device by name
1222 * @path: special file representing the block device
1224 * Get a reference to the blockdevice at @path in the current
1225 * namespace if possible and return it. Return ERR_PTR(error)
1228 struct block_device
*lookup_bdev(const char *path
)
1230 struct block_device
*bdev
;
1231 struct inode
*inode
;
1232 struct nameidata nd
;
1235 if (!path
|| !*path
)
1236 return ERR_PTR(-EINVAL
);
1238 error
= path_lookup(path
, LOOKUP_FOLLOW
, &nd
);
1240 return ERR_PTR(error
);
1242 inode
= nd
.dentry
->d_inode
;
1244 if (!S_ISBLK(inode
->i_mode
))
1247 if (nd
.mnt
->mnt_flags
& MNT_NODEV
)
1250 bdev
= bd_acquire(inode
);
1257 bdev
= ERR_PTR(error
);
1262 * open_bdev_excl - open a block device by name and set it up for use
1264 * @path: special file representing the block device
1265 * @flags: %MS_RDONLY for opening read-only
1266 * @holder: owner for exclusion
1268 * Open the blockdevice described by the special file at @path, claim it
1271 struct block_device
*open_bdev_excl(const char *path
, int flags
, void *holder
)
1273 struct block_device
*bdev
;
1274 mode_t mode
= FMODE_READ
;
1277 bdev
= lookup_bdev(path
);
1281 if (!(flags
& MS_RDONLY
))
1282 mode
|= FMODE_WRITE
;
1283 error
= blkdev_get(bdev
, mode
, 0);
1285 return ERR_PTR(error
);
1287 if (!(flags
& MS_RDONLY
) && bdev_read_only(bdev
))
1289 error
= bd_claim(bdev
, holder
);
1297 return ERR_PTR(error
);
1300 EXPORT_SYMBOL(open_bdev_excl
);
1303 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1305 * @bdev: blockdevice to close
1307 * This is the counterpart to open_bdev_excl().
1309 void close_bdev_excl(struct block_device
*bdev
)
1315 EXPORT_SYMBOL(close_bdev_excl
);