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/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 <asm/uaccess.h>
31 struct block_device bdev
;
32 struct inode vfs_inode
;
35 static const struct address_space_operations def_blk_aops
;
37 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
39 return container_of(inode
, struct bdev_inode
, vfs_inode
);
42 inline struct block_device
*I_BDEV(struct inode
*inode
)
44 return &BDEV_I(inode
)->bdev
;
47 EXPORT_SYMBOL(I_BDEV
);
49 static sector_t
max_block(struct block_device
*bdev
)
51 sector_t retval
= ~((sector_t
)0);
52 loff_t sz
= i_size_read(bdev
->bd_inode
);
55 unsigned int size
= block_size(bdev
);
56 unsigned int sizebits
= blksize_bits(size
);
57 retval
= (sz
>> sizebits
);
62 /* Kill _all_ buffers and pagecache , dirty or not.. */
63 static void kill_bdev(struct block_device
*bdev
)
65 if (bdev
->bd_inode
->i_mapping
->nrpages
== 0)
68 truncate_inode_pages(bdev
->bd_inode
->i_mapping
, 0);
71 int set_blocksize(struct block_device
*bdev
, int size
)
73 /* Size must be a power of two, and between 512 and PAGE_SIZE */
74 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
77 /* Size cannot be smaller than the size supported by the device */
78 if (size
< bdev_hardsect_size(bdev
))
81 /* Don't change the size if it is same as current */
82 if (bdev
->bd_block_size
!= size
) {
84 bdev
->bd_block_size
= size
;
85 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
91 EXPORT_SYMBOL(set_blocksize
);
93 int sb_set_blocksize(struct super_block
*sb
, int size
)
95 if (set_blocksize(sb
->s_bdev
, size
))
97 /* If we get here, we know size is power of two
98 * and it's value is between 512 and PAGE_SIZE */
99 sb
->s_blocksize
= size
;
100 sb
->s_blocksize_bits
= blksize_bits(size
);
101 return sb
->s_blocksize
;
104 EXPORT_SYMBOL(sb_set_blocksize
);
106 int sb_min_blocksize(struct super_block
*sb
, int size
)
108 int minsize
= bdev_hardsect_size(sb
->s_bdev
);
111 return sb_set_blocksize(sb
, size
);
114 EXPORT_SYMBOL(sb_min_blocksize
);
117 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
118 struct buffer_head
*bh
, int create
)
120 if (iblock
>= max_block(I_BDEV(inode
))) {
125 * for reads, we're just trying to fill a partial page.
126 * return a hole, they will have to call get_block again
127 * before they can fill it, and they will get -EIO at that
132 bh
->b_bdev
= I_BDEV(inode
);
133 bh
->b_blocknr
= iblock
;
134 set_buffer_mapped(bh
);
139 blkdev_get_blocks(struct inode
*inode
, sector_t iblock
,
140 struct buffer_head
*bh
, int create
)
142 sector_t end_block
= max_block(I_BDEV(inode
));
143 unsigned long max_blocks
= bh
->b_size
>> inode
->i_blkbits
;
145 if ((iblock
+ max_blocks
) > end_block
) {
146 max_blocks
= end_block
- iblock
;
147 if ((long)max_blocks
<= 0) {
149 return -EIO
; /* write fully beyond EOF */
151 * It is a read which is fully beyond EOF. We return
152 * a !buffer_mapped buffer
158 bh
->b_bdev
= I_BDEV(inode
);
159 bh
->b_blocknr
= iblock
;
160 bh
->b_size
= max_blocks
<< inode
->i_blkbits
;
162 set_buffer_mapped(bh
);
167 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
168 loff_t offset
, unsigned long nr_segs
)
170 struct file
*file
= iocb
->ki_filp
;
171 struct inode
*inode
= file
->f_mapping
->host
;
173 return blockdev_direct_IO_no_locking(rw
, iocb
, inode
, I_BDEV(inode
),
174 iov
, offset
, nr_segs
, blkdev_get_blocks
, NULL
);
177 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
179 return block_write_full_page(page
, blkdev_get_block
, wbc
);
182 static int blkdev_readpage(struct file
* file
, struct page
* page
)
184 return block_read_full_page(page
, blkdev_get_block
);
187 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
188 loff_t pos
, unsigned len
, unsigned flags
,
189 struct page
**pagep
, void **fsdata
)
192 return block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
196 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
197 loff_t pos
, unsigned len
, unsigned copied
,
198 struct page
*page
, void *fsdata
)
201 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
204 page_cache_release(page
);
211 * for a block special file file->f_path.dentry->d_inode->i_size is zero
212 * so we compute the size by hand (just as in block_read/write above)
214 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int origin
)
216 struct inode
*bd_inode
= file
->f_mapping
->host
;
220 mutex_lock(&bd_inode
->i_mutex
);
221 size
= i_size_read(bd_inode
);
228 offset
+= file
->f_pos
;
231 if (offset
>= 0 && offset
<= size
) {
232 if (offset
!= file
->f_pos
) {
233 file
->f_pos
= offset
;
237 mutex_unlock(&bd_inode
->i_mutex
);
242 * Filp is never NULL; the only case when ->fsync() is called with
243 * NULL first argument is nfsd_sync_dir() and that's not a directory.
246 static int block_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
248 return sync_blockdev(I_BDEV(filp
->f_mapping
->host
));
255 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
256 static struct kmem_cache
* bdev_cachep __read_mostly
;
258 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
260 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
263 return &ei
->vfs_inode
;
266 static void bdev_destroy_inode(struct inode
*inode
)
268 struct bdev_inode
*bdi
= BDEV_I(inode
);
270 bdi
->bdev
.bd_inode_backing_dev_info
= NULL
;
271 kmem_cache_free(bdev_cachep
, bdi
);
274 static void init_once(void *foo
)
276 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
277 struct block_device
*bdev
= &ei
->bdev
;
279 memset(bdev
, 0, sizeof(*bdev
));
280 mutex_init(&bdev
->bd_mutex
);
281 sema_init(&bdev
->bd_mount_sem
, 1);
282 INIT_LIST_HEAD(&bdev
->bd_inodes
);
283 INIT_LIST_HEAD(&bdev
->bd_list
);
285 INIT_LIST_HEAD(&bdev
->bd_holder_list
);
287 inode_init_once(&ei
->vfs_inode
);
290 static inline void __bd_forget(struct inode
*inode
)
292 list_del_init(&inode
->i_devices
);
293 inode
->i_bdev
= NULL
;
294 inode
->i_mapping
= &inode
->i_data
;
297 static void bdev_clear_inode(struct inode
*inode
)
299 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
301 spin_lock(&bdev_lock
);
302 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
303 __bd_forget(list_entry(p
, struct inode
, i_devices
));
305 list_del_init(&bdev
->bd_list
);
306 spin_unlock(&bdev_lock
);
309 static const struct super_operations bdev_sops
= {
310 .statfs
= simple_statfs
,
311 .alloc_inode
= bdev_alloc_inode
,
312 .destroy_inode
= bdev_destroy_inode
,
313 .drop_inode
= generic_delete_inode
,
314 .clear_inode
= bdev_clear_inode
,
317 static int bd_get_sb(struct file_system_type
*fs_type
,
318 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
320 return get_sb_pseudo(fs_type
, "bdev:", &bdev_sops
, 0x62646576, mnt
);
323 static struct file_system_type bd_type
= {
326 .kill_sb
= kill_anon_super
,
329 static struct vfsmount
*bd_mnt __read_mostly
;
330 struct super_block
*blockdev_superblock
;
332 void __init
bdev_cache_init(void)
335 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
336 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
337 SLAB_MEM_SPREAD
|SLAB_PANIC
),
339 err
= register_filesystem(&bd_type
);
341 panic("Cannot register bdev pseudo-fs");
342 bd_mnt
= kern_mount(&bd_type
);
344 panic("Cannot create bdev pseudo-fs");
345 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
349 * Most likely _very_ bad one - but then it's hardly critical for small
350 * /dev and can be fixed when somebody will need really large one.
351 * Keep in mind that it will be fed through icache hash function too.
353 static inline unsigned long hash(dev_t dev
)
355 return MAJOR(dev
)+MINOR(dev
);
358 static int bdev_test(struct inode
*inode
, void *data
)
360 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
363 static int bdev_set(struct inode
*inode
, void *data
)
365 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
369 static LIST_HEAD(all_bdevs
);
371 struct block_device
*bdget(dev_t dev
)
373 struct block_device
*bdev
;
376 inode
= iget5_locked(bd_mnt
->mnt_sb
, hash(dev
),
377 bdev_test
, bdev_set
, &dev
);
382 bdev
= &BDEV_I(inode
)->bdev
;
384 if (inode
->i_state
& I_NEW
) {
385 bdev
->bd_contains
= NULL
;
386 bdev
->bd_inode
= inode
;
387 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
388 bdev
->bd_part_count
= 0;
389 bdev
->bd_invalidated
= 0;
390 inode
->i_mode
= S_IFBLK
;
392 inode
->i_bdev
= bdev
;
393 inode
->i_data
.a_ops
= &def_blk_aops
;
394 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
395 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
396 spin_lock(&bdev_lock
);
397 list_add(&bdev
->bd_list
, &all_bdevs
);
398 spin_unlock(&bdev_lock
);
399 unlock_new_inode(inode
);
404 EXPORT_SYMBOL(bdget
);
406 long nr_blockdev_pages(void)
408 struct block_device
*bdev
;
410 spin_lock(&bdev_lock
);
411 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
412 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
414 spin_unlock(&bdev_lock
);
418 void bdput(struct block_device
*bdev
)
420 iput(bdev
->bd_inode
);
423 EXPORT_SYMBOL(bdput
);
425 static struct block_device
*bd_acquire(struct inode
*inode
)
427 struct block_device
*bdev
;
429 spin_lock(&bdev_lock
);
430 bdev
= inode
->i_bdev
;
432 atomic_inc(&bdev
->bd_inode
->i_count
);
433 spin_unlock(&bdev_lock
);
436 spin_unlock(&bdev_lock
);
438 bdev
= bdget(inode
->i_rdev
);
440 spin_lock(&bdev_lock
);
441 if (!inode
->i_bdev
) {
443 * We take an additional bd_inode->i_count for inode,
444 * and it's released in clear_inode() of inode.
445 * So, we can access it via ->i_mapping always
448 atomic_inc(&bdev
->bd_inode
->i_count
);
449 inode
->i_bdev
= bdev
;
450 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
451 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
453 spin_unlock(&bdev_lock
);
458 /* Call when you free inode */
460 void bd_forget(struct inode
*inode
)
462 struct block_device
*bdev
= NULL
;
464 spin_lock(&bdev_lock
);
466 if (inode
->i_sb
!= blockdev_superblock
)
467 bdev
= inode
->i_bdev
;
470 spin_unlock(&bdev_lock
);
473 iput(bdev
->bd_inode
);
476 int bd_claim(struct block_device
*bdev
, void *holder
)
479 spin_lock(&bdev_lock
);
481 /* first decide result */
482 if (bdev
->bd_holder
== holder
)
483 res
= 0; /* already a holder */
484 else if (bdev
->bd_holder
!= NULL
)
485 res
= -EBUSY
; /* held by someone else */
486 else if (bdev
->bd_contains
== bdev
)
487 res
= 0; /* is a whole device which isn't held */
489 else if (bdev
->bd_contains
->bd_holder
== bd_claim
)
490 res
= 0; /* is a partition of a device that is being partitioned */
491 else if (bdev
->bd_contains
->bd_holder
!= NULL
)
492 res
= -EBUSY
; /* is a partition of a held device */
494 res
= 0; /* is a partition of an un-held device */
496 /* now impose change */
498 /* note that for a whole device bd_holders
499 * will be incremented twice, and bd_holder will
500 * be set to bd_claim before being set to holder
502 bdev
->bd_contains
->bd_holders
++;
503 bdev
->bd_contains
->bd_holder
= bd_claim
;
505 bdev
->bd_holder
= holder
;
507 spin_unlock(&bdev_lock
);
511 EXPORT_SYMBOL(bd_claim
);
513 void bd_release(struct block_device
*bdev
)
515 spin_lock(&bdev_lock
);
516 if (!--bdev
->bd_contains
->bd_holders
)
517 bdev
->bd_contains
->bd_holder
= NULL
;
518 if (!--bdev
->bd_holders
)
519 bdev
->bd_holder
= NULL
;
520 spin_unlock(&bdev_lock
);
523 EXPORT_SYMBOL(bd_release
);
527 * Functions for bd_claim_by_kobject / bd_release_from_kobject
529 * If a kobject is passed to bd_claim_by_kobject()
530 * and the kobject has a parent directory,
531 * following symlinks are created:
532 * o from the kobject to the claimed bdev
533 * o from "holders" directory of the bdev to the parent of the kobject
534 * bd_release_from_kobject() removes these symlinks.
537 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
538 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
539 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
540 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
543 static struct kobject
*bdev_get_kobj(struct block_device
*bdev
)
545 if (bdev
->bd_contains
!= bdev
)
546 return kobject_get(&part_to_dev(bdev
->bd_part
)->kobj
);
548 return kobject_get(&disk_to_dev(bdev
->bd_disk
)->kobj
);
551 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
555 return sysfs_create_link(from
, to
, kobject_name(to
));
558 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
562 sysfs_remove_link(from
, kobject_name(to
));
566 * 'struct bd_holder' contains pointers to kobjects symlinked by
567 * bd_claim_by_kobject.
568 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
571 struct list_head list
; /* chain of holders of the bdev */
572 int count
; /* references from the holder */
573 struct kobject
*sdir
; /* holder object, e.g. "/block/dm-0/slaves" */
574 struct kobject
*hdev
; /* e.g. "/block/dm-0" */
575 struct kobject
*hdir
; /* e.g. "/block/sda/holders" */
576 struct kobject
*sdev
; /* e.g. "/block/sda" */
580 * Get references of related kobjects at once.
581 * Returns 1 on success. 0 on failure.
583 * Should call bd_holder_release_dirs() after successful use.
585 static int bd_holder_grab_dirs(struct block_device
*bdev
,
586 struct bd_holder
*bo
)
591 bo
->sdir
= kobject_get(bo
->sdir
);
595 bo
->hdev
= kobject_get(bo
->sdir
->parent
);
599 bo
->sdev
= bdev_get_kobj(bdev
);
603 bo
->hdir
= kobject_get(bdev
->bd_part
->holder_dir
);
610 kobject_put(bo
->sdev
);
612 kobject_put(bo
->hdev
);
614 kobject_put(bo
->sdir
);
619 /* Put references of related kobjects at once. */
620 static void bd_holder_release_dirs(struct bd_holder
*bo
)
622 kobject_put(bo
->hdir
);
623 kobject_put(bo
->sdev
);
624 kobject_put(bo
->hdev
);
625 kobject_put(bo
->sdir
);
628 static struct bd_holder
*alloc_bd_holder(struct kobject
*kobj
)
630 struct bd_holder
*bo
;
632 bo
= kzalloc(sizeof(*bo
), GFP_KERNEL
);
642 static void free_bd_holder(struct bd_holder
*bo
)
648 * find_bd_holder - find matching struct bd_holder from the block device
650 * @bdev: struct block device to be searched
651 * @bo: target struct bd_holder
653 * Returns matching entry with @bo in @bdev->bd_holder_list.
654 * If found, increment the reference count and return the pointer.
655 * If not found, returns NULL.
657 static struct bd_holder
*find_bd_holder(struct block_device
*bdev
,
658 struct bd_holder
*bo
)
660 struct bd_holder
*tmp
;
662 list_for_each_entry(tmp
, &bdev
->bd_holder_list
, list
)
663 if (tmp
->sdir
== bo
->sdir
) {
672 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
674 * @bdev: block device to be bd_claimed
675 * @bo: preallocated and initialized by alloc_bd_holder()
677 * Add @bo to @bdev->bd_holder_list, create symlinks.
679 * Returns 0 if symlinks are created.
680 * Returns -ve if something fails.
682 static int add_bd_holder(struct block_device
*bdev
, struct bd_holder
*bo
)
689 if (!bd_holder_grab_dirs(bdev
, bo
))
692 err
= add_symlink(bo
->sdir
, bo
->sdev
);
696 err
= add_symlink(bo
->hdir
, bo
->hdev
);
698 del_symlink(bo
->sdir
, bo
->sdev
);
702 list_add_tail(&bo
->list
, &bdev
->bd_holder_list
);
707 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
709 * @bdev: block device to be bd_claimed
710 * @kobj: holder's kobject
712 * If there is matching entry with @kobj in @bdev->bd_holder_list
713 * and no other bd_claim() from the same kobject,
714 * remove the struct bd_holder from the list, delete symlinks for it.
716 * Returns a pointer to the struct bd_holder when it's removed from the list
717 * and ready to be freed.
718 * Returns NULL if matching claim isn't found or there is other bd_claim()
719 * by the same kobject.
721 static struct bd_holder
*del_bd_holder(struct block_device
*bdev
,
722 struct kobject
*kobj
)
724 struct bd_holder
*bo
;
726 list_for_each_entry(bo
, &bdev
->bd_holder_list
, list
) {
727 if (bo
->sdir
== kobj
) {
729 BUG_ON(bo
->count
< 0);
732 del_symlink(bo
->sdir
, bo
->sdev
);
733 del_symlink(bo
->hdir
, bo
->hdev
);
734 bd_holder_release_dirs(bo
);
745 * bd_claim_by_kobject - bd_claim() with additional kobject signature
747 * @bdev: block device to be claimed
748 * @holder: holder's signature
749 * @kobj: holder's kobject
751 * Do bd_claim() and if it succeeds, create sysfs symlinks between
752 * the bdev and the holder's kobject.
753 * Use bd_release_from_kobject() when relesing the claimed bdev.
755 * Returns 0 on success. (same as bd_claim())
756 * Returns errno on failure.
758 static int bd_claim_by_kobject(struct block_device
*bdev
, void *holder
,
759 struct kobject
*kobj
)
762 struct bd_holder
*bo
, *found
;
767 bo
= alloc_bd_holder(kobj
);
771 mutex_lock(&bdev
->bd_mutex
);
773 err
= bd_claim(bdev
, holder
);
777 found
= find_bd_holder(bdev
, bo
);
781 err
= add_bd_holder(bdev
, bo
);
787 mutex_unlock(&bdev
->bd_mutex
);
793 * bd_release_from_kobject - bd_release() with additional kobject signature
795 * @bdev: block device to be released
796 * @kobj: holder's kobject
798 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
800 static void bd_release_from_kobject(struct block_device
*bdev
,
801 struct kobject
*kobj
)
806 mutex_lock(&bdev
->bd_mutex
);
808 free_bd_holder(del_bd_holder(bdev
, kobj
));
809 mutex_unlock(&bdev
->bd_mutex
);
813 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
815 * @bdev: block device to be claimed
816 * @holder: holder's signature
817 * @disk: holder's gendisk
819 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
821 int bd_claim_by_disk(struct block_device
*bdev
, void *holder
,
822 struct gendisk
*disk
)
824 return bd_claim_by_kobject(bdev
, holder
, kobject_get(disk
->slave_dir
));
826 EXPORT_SYMBOL_GPL(bd_claim_by_disk
);
829 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
831 * @bdev: block device to be claimed
832 * @disk: holder's gendisk
834 * Call bd_release_from_kobject() and put @disk->slave_dir.
836 void bd_release_from_disk(struct block_device
*bdev
, struct gendisk
*disk
)
838 bd_release_from_kobject(bdev
, disk
->slave_dir
);
839 kobject_put(disk
->slave_dir
);
841 EXPORT_SYMBOL_GPL(bd_release_from_disk
);
845 * Tries to open block device by device number. Use it ONLY if you
846 * really do not have anything better - i.e. when you are behind a
847 * truly sucky interface and all you are given is a device number. _Never_
848 * to be used for internal purposes. If you ever need it - reconsider
851 struct block_device
*open_by_devnum(dev_t dev
, unsigned mode
)
853 struct block_device
*bdev
= bdget(dev
);
855 int flags
= mode
& FMODE_WRITE
? O_RDWR
: O_RDONLY
;
857 err
= blkdev_get(bdev
, mode
, flags
);
858 return err
? ERR_PTR(err
) : bdev
;
861 EXPORT_SYMBOL(open_by_devnum
);
864 * This routine checks whether a removable media has been changed,
865 * and invalidates all buffer-cache-entries in that case. This
866 * is a relatively slow routine, so we have to try to minimize using
867 * it. Thus it is called only upon a 'mount' or 'open'. This
868 * is the best way of combining speed and utility, I think.
869 * People changing diskettes in the middle of an operation deserve
872 int check_disk_change(struct block_device
*bdev
)
874 struct gendisk
*disk
= bdev
->bd_disk
;
875 struct block_device_operations
* bdops
= disk
->fops
;
877 if (!bdops
->media_changed
)
879 if (!bdops
->media_changed(bdev
->bd_disk
))
882 if (__invalidate_device(bdev
))
883 printk("VFS: busy inodes on changed media.\n");
885 if (bdops
->revalidate_disk
)
886 bdops
->revalidate_disk(bdev
->bd_disk
);
887 if (disk_partitionable(bdev
->bd_disk
))
888 bdev
->bd_invalidated
= 1;
892 EXPORT_SYMBOL(check_disk_change
);
894 void bd_set_size(struct block_device
*bdev
, loff_t size
)
896 unsigned bsize
= bdev_hardsect_size(bdev
);
898 bdev
->bd_inode
->i_size
= size
;
899 while (bsize
< PAGE_CACHE_SIZE
) {
904 bdev
->bd_block_size
= bsize
;
905 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
907 EXPORT_SYMBOL(bd_set_size
);
909 static int __blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
,
911 static int __blkdev_put(struct block_device
*bdev
, int for_part
);
916 * mutex_lock(part->bd_mutex)
917 * mutex_lock_nested(whole->bd_mutex, 1)
920 static int do_open(struct block_device
*bdev
, struct file
*file
, int for_part
)
922 struct module
*owner
= NULL
;
923 struct gendisk
*disk
;
924 struct hd_struct
*part
= NULL
;
929 if (file
->f_mode
& FMODE_READ
)
931 if (file
->f_mode
& FMODE_WRITE
)
934 * hooks: /n/, see "layering violations".
936 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
943 file
->f_mapping
= bdev
->bd_inode
->i_mapping
;
945 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
951 owner
= disk
->fops
->owner
;
953 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
954 if (!bdev
->bd_openers
) {
955 bdev
->bd_disk
= disk
;
956 bdev
->bd_contains
= bdev
;
958 struct backing_dev_info
*bdi
;
959 if (disk
->fops
->open
) {
960 ret
= disk
->fops
->open(bdev
->bd_inode
, file
);
964 if (!bdev
->bd_openers
) {
965 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
966 bdi
= blk_get_backing_dev_info(bdev
);
968 bdi
= &default_backing_dev_info
;
969 bdev
->bd_inode
->i_data
.backing_dev_info
= bdi
;
971 if (bdev
->bd_invalidated
)
972 rescan_partitions(disk
, bdev
);
974 struct block_device
*whole
;
975 whole
= bdget_disk(disk
, 0);
980 ret
= __blkdev_get(whole
, file
->f_mode
, file
->f_flags
, 1);
983 bdev
->bd_contains
= whole
;
984 part
= disk_get_part(disk
, partno
);
985 bdev
->bd_inode
->i_data
.backing_dev_info
=
986 whole
->bd_inode
->i_data
.backing_dev_info
;
987 if (!(disk
->flags
& GENHD_FL_UP
) ||
988 !part
|| !part
->nr_sects
) {
992 bdev
->bd_part
= part
;
993 bd_set_size(bdev
, (loff_t
)part
->nr_sects
<< 9);
998 if (bdev
->bd_contains
== bdev
) {
999 if (bdev
->bd_disk
->fops
->open
) {
1000 ret
= bdev
->bd_disk
->fops
->open(bdev
->bd_inode
, file
);
1004 if (bdev
->bd_invalidated
)
1005 rescan_partitions(bdev
->bd_disk
, bdev
);
1010 bdev
->bd_part_count
++;
1011 mutex_unlock(&bdev
->bd_mutex
);
1016 bdev
->bd_disk
= NULL
;
1017 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1018 if (bdev
!= bdev
->bd_contains
)
1019 __blkdev_put(bdev
->bd_contains
, 1);
1020 bdev
->bd_contains
= NULL
;
1022 disk_put_part(part
);
1025 mutex_unlock(&bdev
->bd_mutex
);
1032 static int __blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
,
1036 * This crockload is due to bad choice of ->open() type.
1038 * For now, block device ->open() routine must _not_
1039 * examine anything in 'inode' argument except ->i_rdev.
1041 struct file fake_file
= {};
1042 struct dentry fake_dentry
= {};
1043 fake_file
.f_mode
= mode
;
1044 fake_file
.f_flags
= flags
;
1045 fake_file
.f_path
.dentry
= &fake_dentry
;
1046 fake_dentry
.d_inode
= bdev
->bd_inode
;
1048 return do_open(bdev
, &fake_file
, for_part
);
1051 int blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
)
1053 return __blkdev_get(bdev
, mode
, flags
, 0);
1055 EXPORT_SYMBOL(blkdev_get
);
1057 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1059 struct block_device
*bdev
;
1063 * Preserve backwards compatibility and allow large file access
1064 * even if userspace doesn't ask for it explicitly. Some mkfs
1065 * binary needs it. We might want to drop this workaround
1066 * during an unstable branch.
1068 filp
->f_flags
|= O_LARGEFILE
;
1070 bdev
= bd_acquire(inode
);
1074 res
= do_open(bdev
, filp
, 0);
1078 if (!(filp
->f_flags
& O_EXCL
) )
1081 if (!(res
= bd_claim(bdev
, filp
)))
1088 static int __blkdev_put(struct block_device
*bdev
, int for_part
)
1091 struct inode
*bd_inode
= bdev
->bd_inode
;
1092 struct gendisk
*disk
= bdev
->bd_disk
;
1093 struct block_device
*victim
= NULL
;
1095 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1098 bdev
->bd_part_count
--;
1100 if (!--bdev
->bd_openers
) {
1101 sync_blockdev(bdev
);
1104 if (bdev
->bd_contains
== bdev
) {
1105 if (disk
->fops
->release
)
1106 ret
= disk
->fops
->release(bd_inode
, NULL
);
1108 if (!bdev
->bd_openers
) {
1109 struct module
*owner
= disk
->fops
->owner
;
1114 if (bdev
->bd_contains
!= bdev
) {
1115 disk_put_part(bdev
->bd_part
);
1116 bdev
->bd_part
= NULL
;
1118 bdev
->bd_disk
= NULL
;
1119 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1120 if (bdev
!= bdev
->bd_contains
)
1121 victim
= bdev
->bd_contains
;
1122 bdev
->bd_contains
= NULL
;
1125 mutex_unlock(&bdev
->bd_mutex
);
1128 __blkdev_put(victim
, 1);
1132 int blkdev_put(struct block_device
*bdev
)
1134 return __blkdev_put(bdev
, 0);
1136 EXPORT_SYMBOL(blkdev_put
);
1138 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1140 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1141 if (bdev
->bd_holder
== filp
)
1143 return blkdev_put(bdev
);
1146 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1148 return blkdev_ioctl(file
->f_mapping
->host
, file
, cmd
, arg
);
1151 static const struct address_space_operations def_blk_aops
= {
1152 .readpage
= blkdev_readpage
,
1153 .writepage
= blkdev_writepage
,
1154 .sync_page
= block_sync_page
,
1155 .write_begin
= blkdev_write_begin
,
1156 .write_end
= blkdev_write_end
,
1157 .writepages
= generic_writepages
,
1158 .direct_IO
= blkdev_direct_IO
,
1161 const struct file_operations def_blk_fops
= {
1162 .open
= blkdev_open
,
1163 .release
= blkdev_close
,
1164 .llseek
= block_llseek
,
1165 .read
= do_sync_read
,
1166 .write
= do_sync_write
,
1167 .aio_read
= generic_file_aio_read
,
1168 .aio_write
= generic_file_aio_write_nolock
,
1169 .mmap
= generic_file_mmap
,
1170 .fsync
= block_fsync
,
1171 .unlocked_ioctl
= block_ioctl
,
1172 #ifdef CONFIG_COMPAT
1173 .compat_ioctl
= compat_blkdev_ioctl
,
1175 .splice_read
= generic_file_splice_read
,
1176 .splice_write
= generic_file_splice_write
,
1179 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1182 mm_segment_t old_fs
= get_fs();
1184 res
= blkdev_ioctl(bdev
->bd_inode
, NULL
, cmd
, arg
);
1189 EXPORT_SYMBOL(ioctl_by_bdev
);
1192 * lookup_bdev - lookup a struct block_device by name
1194 * @path: special file representing the block device
1196 * Get a reference to the blockdevice at @path in the current
1197 * namespace if possible and return it. Return ERR_PTR(error)
1200 struct block_device
*lookup_bdev(const char *path
)
1202 struct block_device
*bdev
;
1203 struct inode
*inode
;
1204 struct nameidata nd
;
1207 if (!path
|| !*path
)
1208 return ERR_PTR(-EINVAL
);
1210 error
= path_lookup(path
, LOOKUP_FOLLOW
, &nd
);
1212 return ERR_PTR(error
);
1214 inode
= nd
.path
.dentry
->d_inode
;
1216 if (!S_ISBLK(inode
->i_mode
))
1219 if (nd
.path
.mnt
->mnt_flags
& MNT_NODEV
)
1222 bdev
= bd_acquire(inode
);
1229 bdev
= ERR_PTR(error
);
1232 EXPORT_SYMBOL(lookup_bdev
);
1235 * open_bdev_excl - open a block device by name and set it up for use
1237 * @path: special file representing the block device
1238 * @flags: %MS_RDONLY for opening read-only
1239 * @holder: owner for exclusion
1241 * Open the blockdevice described by the special file at @path, claim it
1244 struct block_device
*open_bdev_excl(const char *path
, int flags
, void *holder
)
1246 struct block_device
*bdev
;
1247 mode_t mode
= FMODE_READ
;
1250 bdev
= lookup_bdev(path
);
1254 if (!(flags
& MS_RDONLY
))
1255 mode
|= FMODE_WRITE
;
1256 error
= blkdev_get(bdev
, mode
, 0);
1258 return ERR_PTR(error
);
1260 if (!(flags
& MS_RDONLY
) && bdev_read_only(bdev
))
1262 error
= bd_claim(bdev
, holder
);
1270 return ERR_PTR(error
);
1273 EXPORT_SYMBOL(open_bdev_excl
);
1276 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1278 * @bdev: blockdevice to close
1280 * This is the counterpart to open_bdev_excl().
1282 void close_bdev_excl(struct block_device
*bdev
)
1288 EXPORT_SYMBOL(close_bdev_excl
);
1290 int __invalidate_device(struct block_device
*bdev
)
1292 struct super_block
*sb
= get_super(bdev
);
1297 * no need to lock the super, get_super holds the
1298 * read mutex so the filesystem cannot go away
1299 * under us (->put_super runs with the write lock
1302 shrink_dcache_sb(sb
);
1303 res
= invalidate_inodes(sb
);
1306 invalidate_bdev(bdev
);
1309 EXPORT_SYMBOL(__invalidate_device
);