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/writeback.h>
21 #include <linux/mpage.h>
22 #include <linux/mount.h>
23 #include <linux/uio.h>
24 #include <linux/namei.h>
25 #include <asm/uaccess.h>
29 struct block_device bdev
;
30 struct inode vfs_inode
;
33 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
35 return container_of(inode
, struct bdev_inode
, vfs_inode
);
38 inline struct block_device
*I_BDEV(struct inode
*inode
)
40 return &BDEV_I(inode
)->bdev
;
43 EXPORT_SYMBOL(I_BDEV
);
45 static sector_t
max_block(struct block_device
*bdev
)
47 sector_t retval
= ~((sector_t
)0);
48 loff_t sz
= i_size_read(bdev
->bd_inode
);
51 unsigned int size
= block_size(bdev
);
52 unsigned int sizebits
= blksize_bits(size
);
53 retval
= (sz
>> sizebits
);
58 /* Kill _all_ buffers and pagecache , dirty or not.. */
59 static void kill_bdev(struct block_device
*bdev
)
61 if (bdev
->bd_inode
->i_mapping
->nrpages
== 0)
64 truncate_inode_pages(bdev
->bd_inode
->i_mapping
, 0);
67 int set_blocksize(struct block_device
*bdev
, int size
)
69 /* Size must be a power of two, and between 512 and PAGE_SIZE */
70 if (size
> PAGE_SIZE
|| size
< 512 || (size
& (size
-1)))
73 /* Size cannot be smaller than the size supported by the device */
74 if (size
< bdev_hardsect_size(bdev
))
77 /* Don't change the size if it is same as current */
78 if (bdev
->bd_block_size
!= size
) {
80 bdev
->bd_block_size
= size
;
81 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
87 EXPORT_SYMBOL(set_blocksize
);
89 int sb_set_blocksize(struct super_block
*sb
, int size
)
91 if (set_blocksize(sb
->s_bdev
, size
))
93 /* If we get here, we know size is power of two
94 * and it's value is between 512 and PAGE_SIZE */
95 sb
->s_blocksize
= size
;
96 sb
->s_blocksize_bits
= blksize_bits(size
);
97 return sb
->s_blocksize
;
100 EXPORT_SYMBOL(sb_set_blocksize
);
102 int sb_min_blocksize(struct super_block
*sb
, int size
)
104 int minsize
= bdev_hardsect_size(sb
->s_bdev
);
107 return sb_set_blocksize(sb
, size
);
110 EXPORT_SYMBOL(sb_min_blocksize
);
113 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
114 struct buffer_head
*bh
, int create
)
116 if (iblock
>= max_block(I_BDEV(inode
))) {
121 * for reads, we're just trying to fill a partial page.
122 * return a hole, they will have to call get_block again
123 * before they can fill it, and they will get -EIO at that
128 bh
->b_bdev
= I_BDEV(inode
);
129 bh
->b_blocknr
= iblock
;
130 set_buffer_mapped(bh
);
135 blkdev_get_blocks(struct inode
*inode
, sector_t iblock
,
136 struct buffer_head
*bh
, int create
)
138 sector_t end_block
= max_block(I_BDEV(inode
));
139 unsigned long max_blocks
= bh
->b_size
>> inode
->i_blkbits
;
141 if ((iblock
+ max_blocks
) > end_block
) {
142 max_blocks
= end_block
- iblock
;
143 if ((long)max_blocks
<= 0) {
145 return -EIO
; /* write fully beyond EOF */
147 * It is a read which is fully beyond EOF. We return
148 * a !buffer_mapped buffer
154 bh
->b_bdev
= I_BDEV(inode
);
155 bh
->b_blocknr
= iblock
;
156 bh
->b_size
= max_blocks
<< inode
->i_blkbits
;
158 set_buffer_mapped(bh
);
163 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
164 loff_t offset
, unsigned long nr_segs
)
166 struct file
*file
= iocb
->ki_filp
;
167 struct inode
*inode
= file
->f_mapping
->host
;
169 return blockdev_direct_IO_no_locking(rw
, iocb
, inode
, I_BDEV(inode
),
170 iov
, offset
, nr_segs
, blkdev_get_blocks
, NULL
);
174 static int blk_end_aio(struct bio
*bio
, unsigned int bytes_done
, int error
)
176 struct kiocb
*iocb
= bio
->bi_private
;
177 atomic_t
*bio_count
= &iocb
->ki_bio_count
;
179 if (bio_data_dir(bio
) == READ
)
180 bio_check_pages_dirty(bio
);
182 bio_release_pages(bio
);
186 /* iocb->ki_nbytes stores error code from LLDD */
188 iocb
->ki_nbytes
= -EIO
;
190 if (atomic_dec_and_test(bio_count
)) {
191 if ((long)iocb
->ki_nbytes
< 0)
192 aio_complete(iocb
, iocb
->ki_nbytes
, 0);
194 aio_complete(iocb
, iocb
->ki_left
, 0);
204 struct page
*page
[VEC_SIZE
];
207 #define PAGES_SPANNED(addr, len) \
208 (DIV_ROUND_UP((addr) + (len), PAGE_SIZE) - (addr) / PAGE_SIZE);
211 * get page pointer for user addr, we internally cache struct page array for
212 * (addr, count) range in pvec to avoid frequent call to get_user_pages. If
213 * internal page list is exhausted, a batch count of up to VEC_SIZE is used
214 * to get next set of page struct.
216 static struct page
*blk_get_page(unsigned long addr
, size_t count
, int rw
,
220 if (pvec
->idx
== pvec
->nr
) {
221 nr_pages
= PAGES_SPANNED(addr
, count
);
222 nr_pages
= min(nr_pages
, VEC_SIZE
);
223 down_read(¤t
->mm
->mmap_sem
);
224 ret
= get_user_pages(current
, current
->mm
, addr
, nr_pages
,
225 rw
== READ
, 0, pvec
->page
, NULL
);
226 up_read(¤t
->mm
->mmap_sem
);
232 return pvec
->page
[pvec
->idx
++];
235 /* return a page back to pvec array */
236 static void blk_unget_page(struct page
*page
, struct pvec
*pvec
)
238 pvec
->page
[--pvec
->idx
] = page
;
242 blkdev_direct_IO(int rw
, struct kiocb
*iocb
, const struct iovec
*iov
,
243 loff_t pos
, unsigned long nr_segs
)
245 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
246 unsigned blkbits
= blksize_bits(bdev_hardsect_size(I_BDEV(inode
)));
247 unsigned blocksize_mask
= (1 << blkbits
) - 1;
248 unsigned long seg
= 0; /* iov segment iterator */
249 unsigned long nvec
; /* number of bio vec needed */
250 unsigned long cur_off
; /* offset into current page */
251 unsigned long cur_len
; /* I/O len of current page, up to PAGE_SIZE */
253 unsigned long addr
; /* user iovec address */
254 size_t count
; /* user iovec len */
255 size_t nbytes
= iocb
->ki_nbytes
= iocb
->ki_left
; /* total xfer size */
256 loff_t size
; /* size of block device */
258 atomic_t
*bio_count
= &iocb
->ki_bio_count
;
265 if (pos
& blocksize_mask
)
268 size
= i_size_read(inode
);
269 if (pos
+ nbytes
> size
) {
271 iocb
->ki_left
= nbytes
;
275 * check first non-zero iov alignment, the remaining
276 * iov alignment is checked inside bio loop below.
279 addr
= (unsigned long) iov
[seg
].iov_base
;
280 count
= min(iov
[seg
].iov_len
, nbytes
);
281 if (addr
& blocksize_mask
|| count
& blocksize_mask
)
283 } while (!count
&& ++seg
< nr_segs
);
284 atomic_set(bio_count
, 1);
287 /* roughly estimate number of bio vec needed */
288 nvec
= (nbytes
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
289 nvec
= max(nvec
, nr_segs
- seg
);
290 nvec
= min(nvec
, (unsigned long) BIO_MAX_PAGES
);
292 /* bio_alloc should not fail with GFP_KERNEL flag */
293 bio
= bio_alloc(GFP_KERNEL
, nvec
);
294 bio
->bi_bdev
= I_BDEV(inode
);
295 bio
->bi_end_io
= blk_end_aio
;
296 bio
->bi_private
= iocb
;
297 bio
->bi_sector
= pos
>> blkbits
;
299 cur_off
= addr
& ~PAGE_MASK
;
300 cur_len
= PAGE_SIZE
- cur_off
;
304 page
= blk_get_page(addr
, count
, rw
, &pvec
);
305 if (unlikely(IS_ERR(page
)))
308 if (bio_add_page(bio
, page
, cur_len
, cur_off
)) {
316 while (++seg
< nr_segs
) {
317 addr
= (unsigned long) iov
[seg
].iov_base
;
318 count
= iov
[seg
].iov_len
;
321 if (unlikely(addr
& blocksize_mask
||
322 count
& blocksize_mask
)) {
323 page
= ERR_PTR(-EINVAL
);
326 count
= min(count
, nbytes
);
330 blk_unget_page(page
, &pvec
);
333 /* bio is ready, submit it */
335 bio_set_pages_dirty(bio
);
336 atomic_inc(bio_count
);
341 iocb
->ki_left
-= nbytes
;
342 nbytes
= iocb
->ki_left
;
343 iocb
->ki_pos
+= nbytes
;
345 blk_run_address_space(inode
->i_mapping
);
346 if (atomic_dec_and_test(bio_count
))
347 aio_complete(iocb
, nbytes
, 0);
353 * back out nbytes count constructed so far for this bio,
354 * we will throw away current bio.
356 nbytes
+= bio
->bi_size
;
357 bio_release_pages(bio
);
361 * if no bio was submmitted, return the error code.
362 * otherwise, proceed with pending I/O completion.
364 if (atomic_read(bio_count
) == 1)
365 return PTR_ERR(page
);
370 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
372 return block_write_full_page(page
, blkdev_get_block
, wbc
);
375 static int blkdev_readpage(struct file
* file
, struct page
* page
)
377 return block_read_full_page(page
, blkdev_get_block
);
380 static int blkdev_prepare_write(struct file
*file
, struct page
*page
, unsigned from
, unsigned to
)
382 return block_prepare_write(page
, from
, to
, blkdev_get_block
);
385 static int blkdev_commit_write(struct file
*file
, struct page
*page
, unsigned from
, unsigned to
)
387 return block_commit_write(page
, from
, to
);
392 * for a block special file file->f_path.dentry->d_inode->i_size is zero
393 * so we compute the size by hand (just as in block_read/write above)
395 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int origin
)
397 struct inode
*bd_inode
= file
->f_mapping
->host
;
401 mutex_lock(&bd_inode
->i_mutex
);
402 size
= i_size_read(bd_inode
);
409 offset
+= file
->f_pos
;
412 if (offset
>= 0 && offset
<= size
) {
413 if (offset
!= file
->f_pos
) {
414 file
->f_pos
= offset
;
418 mutex_unlock(&bd_inode
->i_mutex
);
423 * Filp is never NULL; the only case when ->fsync() is called with
424 * NULL first argument is nfsd_sync_dir() and that's not a directory.
427 static int block_fsync(struct file
*filp
, struct dentry
*dentry
, int datasync
)
429 return sync_blockdev(I_BDEV(filp
->f_mapping
->host
));
436 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
437 static struct kmem_cache
* bdev_cachep __read_mostly
;
439 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
441 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
444 return &ei
->vfs_inode
;
447 static void bdev_destroy_inode(struct inode
*inode
)
449 struct bdev_inode
*bdi
= BDEV_I(inode
);
451 bdi
->bdev
.bd_inode_backing_dev_info
= NULL
;
452 kmem_cache_free(bdev_cachep
, bdi
);
455 static void init_once(void * foo
, struct kmem_cache
* cachep
, unsigned long flags
)
457 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
458 struct block_device
*bdev
= &ei
->bdev
;
460 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) ==
461 SLAB_CTOR_CONSTRUCTOR
)
463 memset(bdev
, 0, sizeof(*bdev
));
464 mutex_init(&bdev
->bd_mutex
);
465 sema_init(&bdev
->bd_mount_sem
, 1);
466 INIT_LIST_HEAD(&bdev
->bd_inodes
);
467 INIT_LIST_HEAD(&bdev
->bd_list
);
469 INIT_LIST_HEAD(&bdev
->bd_holder_list
);
471 inode_init_once(&ei
->vfs_inode
);
475 static inline void __bd_forget(struct inode
*inode
)
477 list_del_init(&inode
->i_devices
);
478 inode
->i_bdev
= NULL
;
479 inode
->i_mapping
= &inode
->i_data
;
482 static void bdev_clear_inode(struct inode
*inode
)
484 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
486 spin_lock(&bdev_lock
);
487 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
488 __bd_forget(list_entry(p
, struct inode
, i_devices
));
490 list_del_init(&bdev
->bd_list
);
491 spin_unlock(&bdev_lock
);
494 static const struct super_operations bdev_sops
= {
495 .statfs
= simple_statfs
,
496 .alloc_inode
= bdev_alloc_inode
,
497 .destroy_inode
= bdev_destroy_inode
,
498 .drop_inode
= generic_delete_inode
,
499 .clear_inode
= bdev_clear_inode
,
502 static int bd_get_sb(struct file_system_type
*fs_type
,
503 int flags
, const char *dev_name
, void *data
, struct vfsmount
*mnt
)
505 return get_sb_pseudo(fs_type
, "bdev:", &bdev_sops
, 0x62646576, mnt
);
508 static struct file_system_type bd_type
= {
511 .kill_sb
= kill_anon_super
,
514 static struct vfsmount
*bd_mnt __read_mostly
;
515 struct super_block
*blockdev_superblock
;
517 void __init
bdev_cache_init(void)
520 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
521 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
522 SLAB_MEM_SPREAD
|SLAB_PANIC
),
524 err
= register_filesystem(&bd_type
);
526 panic("Cannot register bdev pseudo-fs");
527 bd_mnt
= kern_mount(&bd_type
);
528 err
= PTR_ERR(bd_mnt
);
530 panic("Cannot create bdev pseudo-fs");
531 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
535 * Most likely _very_ bad one - but then it's hardly critical for small
536 * /dev and can be fixed when somebody will need really large one.
537 * Keep in mind that it will be fed through icache hash function too.
539 static inline unsigned long hash(dev_t dev
)
541 return MAJOR(dev
)+MINOR(dev
);
544 static int bdev_test(struct inode
*inode
, void *data
)
546 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
549 static int bdev_set(struct inode
*inode
, void *data
)
551 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
555 static LIST_HEAD(all_bdevs
);
557 struct block_device
*bdget(dev_t dev
)
559 struct block_device
*bdev
;
562 inode
= iget5_locked(bd_mnt
->mnt_sb
, hash(dev
),
563 bdev_test
, bdev_set
, &dev
);
568 bdev
= &BDEV_I(inode
)->bdev
;
570 if (inode
->i_state
& I_NEW
) {
571 bdev
->bd_contains
= NULL
;
572 bdev
->bd_inode
= inode
;
573 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
574 bdev
->bd_part_count
= 0;
575 bdev
->bd_invalidated
= 0;
576 inode
->i_mode
= S_IFBLK
;
578 inode
->i_bdev
= bdev
;
579 inode
->i_data
.a_ops
= &def_blk_aops
;
580 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
581 inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
582 spin_lock(&bdev_lock
);
583 list_add(&bdev
->bd_list
, &all_bdevs
);
584 spin_unlock(&bdev_lock
);
585 unlock_new_inode(inode
);
590 EXPORT_SYMBOL(bdget
);
592 long nr_blockdev_pages(void)
596 spin_lock(&bdev_lock
);
597 list_for_each(p
, &all_bdevs
) {
598 struct block_device
*bdev
;
599 bdev
= list_entry(p
, struct block_device
, bd_list
);
600 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
602 spin_unlock(&bdev_lock
);
606 void bdput(struct block_device
*bdev
)
608 iput(bdev
->bd_inode
);
611 EXPORT_SYMBOL(bdput
);
613 static struct block_device
*bd_acquire(struct inode
*inode
)
615 struct block_device
*bdev
;
617 spin_lock(&bdev_lock
);
618 bdev
= inode
->i_bdev
;
620 atomic_inc(&bdev
->bd_inode
->i_count
);
621 spin_unlock(&bdev_lock
);
624 spin_unlock(&bdev_lock
);
626 bdev
= bdget(inode
->i_rdev
);
628 spin_lock(&bdev_lock
);
629 if (!inode
->i_bdev
) {
631 * We take an additional bd_inode->i_count for inode,
632 * and it's released in clear_inode() of inode.
633 * So, we can access it via ->i_mapping always
636 atomic_inc(&bdev
->bd_inode
->i_count
);
637 inode
->i_bdev
= bdev
;
638 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
639 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
641 spin_unlock(&bdev_lock
);
646 /* Call when you free inode */
648 void bd_forget(struct inode
*inode
)
650 struct block_device
*bdev
= NULL
;
652 spin_lock(&bdev_lock
);
654 if (inode
->i_sb
!= blockdev_superblock
)
655 bdev
= inode
->i_bdev
;
658 spin_unlock(&bdev_lock
);
661 iput(bdev
->bd_inode
);
664 int bd_claim(struct block_device
*bdev
, void *holder
)
667 spin_lock(&bdev_lock
);
669 /* first decide result */
670 if (bdev
->bd_holder
== holder
)
671 res
= 0; /* already a holder */
672 else if (bdev
->bd_holder
!= NULL
)
673 res
= -EBUSY
; /* held by someone else */
674 else if (bdev
->bd_contains
== bdev
)
675 res
= 0; /* is a whole device which isn't held */
677 else if (bdev
->bd_contains
->bd_holder
== bd_claim
)
678 res
= 0; /* is a partition of a device that is being partitioned */
679 else if (bdev
->bd_contains
->bd_holder
!= NULL
)
680 res
= -EBUSY
; /* is a partition of a held device */
682 res
= 0; /* is a partition of an un-held device */
684 /* now impose change */
686 /* note that for a whole device bd_holders
687 * will be incremented twice, and bd_holder will
688 * be set to bd_claim before being set to holder
690 bdev
->bd_contains
->bd_holders
++;
691 bdev
->bd_contains
->bd_holder
= bd_claim
;
693 bdev
->bd_holder
= holder
;
695 spin_unlock(&bdev_lock
);
699 EXPORT_SYMBOL(bd_claim
);
701 void bd_release(struct block_device
*bdev
)
703 spin_lock(&bdev_lock
);
704 if (!--bdev
->bd_contains
->bd_holders
)
705 bdev
->bd_contains
->bd_holder
= NULL
;
706 if (!--bdev
->bd_holders
)
707 bdev
->bd_holder
= NULL
;
708 spin_unlock(&bdev_lock
);
711 EXPORT_SYMBOL(bd_release
);
715 * Functions for bd_claim_by_kobject / bd_release_from_kobject
717 * If a kobject is passed to bd_claim_by_kobject()
718 * and the kobject has a parent directory,
719 * following symlinks are created:
720 * o from the kobject to the claimed bdev
721 * o from "holders" directory of the bdev to the parent of the kobject
722 * bd_release_from_kobject() removes these symlinks.
725 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
726 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
727 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
728 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
731 static struct kobject
*bdev_get_kobj(struct block_device
*bdev
)
733 if (bdev
->bd_contains
!= bdev
)
734 return kobject_get(&bdev
->bd_part
->kobj
);
736 return kobject_get(&bdev
->bd_disk
->kobj
);
739 static struct kobject
*bdev_get_holder(struct block_device
*bdev
)
741 if (bdev
->bd_contains
!= bdev
)
742 return kobject_get(bdev
->bd_part
->holder_dir
);
744 return kobject_get(bdev
->bd_disk
->holder_dir
);
747 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
751 return sysfs_create_link(from
, to
, kobject_name(to
));
754 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
758 sysfs_remove_link(from
, kobject_name(to
));
762 * 'struct bd_holder' contains pointers to kobjects symlinked by
763 * bd_claim_by_kobject.
764 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
767 struct list_head list
; /* chain of holders of the bdev */
768 int count
; /* references from the holder */
769 struct kobject
*sdir
; /* holder object, e.g. "/block/dm-0/slaves" */
770 struct kobject
*hdev
; /* e.g. "/block/dm-0" */
771 struct kobject
*hdir
; /* e.g. "/block/sda/holders" */
772 struct kobject
*sdev
; /* e.g. "/block/sda" */
776 * Get references of related kobjects at once.
777 * Returns 1 on success. 0 on failure.
779 * Should call bd_holder_release_dirs() after successful use.
781 static int bd_holder_grab_dirs(struct block_device
*bdev
,
782 struct bd_holder
*bo
)
787 bo
->sdir
= kobject_get(bo
->sdir
);
791 bo
->hdev
= kobject_get(bo
->sdir
->parent
);
795 bo
->sdev
= bdev_get_kobj(bdev
);
799 bo
->hdir
= bdev_get_holder(bdev
);
806 kobject_put(bo
->sdev
);
808 kobject_put(bo
->hdev
);
810 kobject_put(bo
->sdir
);
815 /* Put references of related kobjects at once. */
816 static void bd_holder_release_dirs(struct bd_holder
*bo
)
818 kobject_put(bo
->hdir
);
819 kobject_put(bo
->sdev
);
820 kobject_put(bo
->hdev
);
821 kobject_put(bo
->sdir
);
824 static struct bd_holder
*alloc_bd_holder(struct kobject
*kobj
)
826 struct bd_holder
*bo
;
828 bo
= kzalloc(sizeof(*bo
), GFP_KERNEL
);
838 static void free_bd_holder(struct bd_holder
*bo
)
844 * find_bd_holder - find matching struct bd_holder from the block device
846 * @bdev: struct block device to be searched
847 * @bo: target struct bd_holder
849 * Returns matching entry with @bo in @bdev->bd_holder_list.
850 * If found, increment the reference count and return the pointer.
851 * If not found, returns NULL.
853 static struct bd_holder
*find_bd_holder(struct block_device
*bdev
,
854 struct bd_holder
*bo
)
856 struct bd_holder
*tmp
;
858 list_for_each_entry(tmp
, &bdev
->bd_holder_list
, list
)
859 if (tmp
->sdir
== bo
->sdir
) {
868 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
870 * @bdev: block device to be bd_claimed
871 * @bo: preallocated and initialized by alloc_bd_holder()
873 * Add @bo to @bdev->bd_holder_list, create symlinks.
875 * Returns 0 if symlinks are created.
876 * Returns -ve if something fails.
878 static int add_bd_holder(struct block_device
*bdev
, struct bd_holder
*bo
)
885 if (!bd_holder_grab_dirs(bdev
, bo
))
888 ret
= add_symlink(bo
->sdir
, bo
->sdev
);
890 ret
= add_symlink(bo
->hdir
, bo
->hdev
);
892 del_symlink(bo
->sdir
, bo
->sdev
);
895 list_add_tail(&bo
->list
, &bdev
->bd_holder_list
);
900 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
902 * @bdev: block device to be bd_claimed
903 * @kobj: holder's kobject
905 * If there is matching entry with @kobj in @bdev->bd_holder_list
906 * and no other bd_claim() from the same kobject,
907 * remove the struct bd_holder from the list, delete symlinks for it.
909 * Returns a pointer to the struct bd_holder when it's removed from the list
910 * and ready to be freed.
911 * Returns NULL if matching claim isn't found or there is other bd_claim()
912 * by the same kobject.
914 static struct bd_holder
*del_bd_holder(struct block_device
*bdev
,
915 struct kobject
*kobj
)
917 struct bd_holder
*bo
;
919 list_for_each_entry(bo
, &bdev
->bd_holder_list
, list
) {
920 if (bo
->sdir
== kobj
) {
922 BUG_ON(bo
->count
< 0);
925 del_symlink(bo
->sdir
, bo
->sdev
);
926 del_symlink(bo
->hdir
, bo
->hdev
);
927 bd_holder_release_dirs(bo
);
938 * bd_claim_by_kobject - bd_claim() with additional kobject signature
940 * @bdev: block device to be claimed
941 * @holder: holder's signature
942 * @kobj: holder's kobject
944 * Do bd_claim() and if it succeeds, create sysfs symlinks between
945 * the bdev and the holder's kobject.
946 * Use bd_release_from_kobject() when relesing the claimed bdev.
948 * Returns 0 on success. (same as bd_claim())
949 * Returns errno on failure.
951 static int bd_claim_by_kobject(struct block_device
*bdev
, void *holder
,
952 struct kobject
*kobj
)
955 struct bd_holder
*bo
, *found
;
960 bo
= alloc_bd_holder(kobj
);
964 mutex_lock(&bdev
->bd_mutex
);
965 res
= bd_claim(bdev
, holder
);
967 found
= find_bd_holder(bdev
, bo
);
969 res
= add_bd_holder(bdev
, bo
);
977 mutex_unlock(&bdev
->bd_mutex
);
983 * bd_release_from_kobject - bd_release() with additional kobject signature
985 * @bdev: block device to be released
986 * @kobj: holder's kobject
988 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
990 static void bd_release_from_kobject(struct block_device
*bdev
,
991 struct kobject
*kobj
)
993 struct bd_holder
*bo
;
998 mutex_lock(&bdev
->bd_mutex
);
1000 if ((bo
= del_bd_holder(bdev
, kobj
)))
1002 mutex_unlock(&bdev
->bd_mutex
);
1006 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1008 * @bdev: block device to be claimed
1009 * @holder: holder's signature
1010 * @disk: holder's gendisk
1012 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1014 int bd_claim_by_disk(struct block_device
*bdev
, void *holder
,
1015 struct gendisk
*disk
)
1017 return bd_claim_by_kobject(bdev
, holder
, kobject_get(disk
->slave_dir
));
1019 EXPORT_SYMBOL_GPL(bd_claim_by_disk
);
1022 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1024 * @bdev: block device to be claimed
1025 * @disk: holder's gendisk
1027 * Call bd_release_from_kobject() and put @disk->slave_dir.
1029 void bd_release_from_disk(struct block_device
*bdev
, struct gendisk
*disk
)
1031 bd_release_from_kobject(bdev
, disk
->slave_dir
);
1032 kobject_put(disk
->slave_dir
);
1034 EXPORT_SYMBOL_GPL(bd_release_from_disk
);
1038 * Tries to open block device by device number. Use it ONLY if you
1039 * really do not have anything better - i.e. when you are behind a
1040 * truly sucky interface and all you are given is a device number. _Never_
1041 * to be used for internal purposes. If you ever need it - reconsider
1044 struct block_device
*open_by_devnum(dev_t dev
, unsigned mode
)
1046 struct block_device
*bdev
= bdget(dev
);
1048 int flags
= mode
& FMODE_WRITE
? O_RDWR
: O_RDONLY
;
1050 err
= blkdev_get(bdev
, mode
, flags
);
1051 return err
? ERR_PTR(err
) : bdev
;
1054 EXPORT_SYMBOL(open_by_devnum
);
1057 * This routine checks whether a removable media has been changed,
1058 * and invalidates all buffer-cache-entries in that case. This
1059 * is a relatively slow routine, so we have to try to minimize using
1060 * it. Thus it is called only upon a 'mount' or 'open'. This
1061 * is the best way of combining speed and utility, I think.
1062 * People changing diskettes in the middle of an operation deserve
1065 int check_disk_change(struct block_device
*bdev
)
1067 struct gendisk
*disk
= bdev
->bd_disk
;
1068 struct block_device_operations
* bdops
= disk
->fops
;
1070 if (!bdops
->media_changed
)
1072 if (!bdops
->media_changed(bdev
->bd_disk
))
1075 if (__invalidate_device(bdev
))
1076 printk("VFS: busy inodes on changed media.\n");
1078 if (bdops
->revalidate_disk
)
1079 bdops
->revalidate_disk(bdev
->bd_disk
);
1080 if (bdev
->bd_disk
->minors
> 1)
1081 bdev
->bd_invalidated
= 1;
1085 EXPORT_SYMBOL(check_disk_change
);
1087 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1089 unsigned bsize
= bdev_hardsect_size(bdev
);
1091 bdev
->bd_inode
->i_size
= size
;
1092 while (bsize
< PAGE_CACHE_SIZE
) {
1097 bdev
->bd_block_size
= bsize
;
1098 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1100 EXPORT_SYMBOL(bd_set_size
);
1102 static int __blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
,
1104 static int __blkdev_put(struct block_device
*bdev
, int for_part
);
1109 * mutex_lock(part->bd_mutex)
1110 * mutex_lock_nested(whole->bd_mutex, 1)
1113 static int do_open(struct block_device
*bdev
, struct file
*file
, int for_part
)
1115 struct module
*owner
= NULL
;
1116 struct gendisk
*disk
;
1120 file
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1122 disk
= get_gendisk(bdev
->bd_dev
, &part
);
1128 owner
= disk
->fops
->owner
;
1130 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1131 if (!bdev
->bd_openers
) {
1132 bdev
->bd_disk
= disk
;
1133 bdev
->bd_contains
= bdev
;
1135 struct backing_dev_info
*bdi
;
1136 if (disk
->fops
->open
) {
1137 ret
= disk
->fops
->open(bdev
->bd_inode
, file
);
1141 if (!bdev
->bd_openers
) {
1142 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1143 bdi
= blk_get_backing_dev_info(bdev
);
1145 bdi
= &default_backing_dev_info
;
1146 bdev
->bd_inode
->i_data
.backing_dev_info
= bdi
;
1148 if (bdev
->bd_invalidated
)
1149 rescan_partitions(disk
, bdev
);
1151 struct hd_struct
*p
;
1152 struct block_device
*whole
;
1153 whole
= bdget_disk(disk
, 0);
1158 ret
= __blkdev_get(whole
, file
->f_mode
, file
->f_flags
, 1);
1161 bdev
->bd_contains
= whole
;
1162 p
= disk
->part
[part
- 1];
1163 bdev
->bd_inode
->i_data
.backing_dev_info
=
1164 whole
->bd_inode
->i_data
.backing_dev_info
;
1165 if (!(disk
->flags
& GENHD_FL_UP
) || !p
|| !p
->nr_sects
) {
1169 kobject_get(&p
->kobj
);
1171 bd_set_size(bdev
, (loff_t
) p
->nr_sects
<< 9);
1176 if (bdev
->bd_contains
== bdev
) {
1177 if (bdev
->bd_disk
->fops
->open
) {
1178 ret
= bdev
->bd_disk
->fops
->open(bdev
->bd_inode
, file
);
1182 if (bdev
->bd_invalidated
)
1183 rescan_partitions(bdev
->bd_disk
, bdev
);
1188 bdev
->bd_part_count
++;
1189 mutex_unlock(&bdev
->bd_mutex
);
1194 bdev
->bd_disk
= NULL
;
1195 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1196 if (bdev
!= bdev
->bd_contains
)
1197 __blkdev_put(bdev
->bd_contains
, 1);
1198 bdev
->bd_contains
= NULL
;
1202 mutex_unlock(&bdev
->bd_mutex
);
1209 static int __blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
,
1213 * This crockload is due to bad choice of ->open() type.
1215 * For now, block device ->open() routine must _not_
1216 * examine anything in 'inode' argument except ->i_rdev.
1218 struct file fake_file
= {};
1219 struct dentry fake_dentry
= {};
1220 fake_file
.f_mode
= mode
;
1221 fake_file
.f_flags
= flags
;
1222 fake_file
.f_path
.dentry
= &fake_dentry
;
1223 fake_dentry
.d_inode
= bdev
->bd_inode
;
1225 return do_open(bdev
, &fake_file
, for_part
);
1228 int blkdev_get(struct block_device
*bdev
, mode_t mode
, unsigned flags
)
1230 return __blkdev_get(bdev
, mode
, flags
, 0);
1232 EXPORT_SYMBOL(blkdev_get
);
1234 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1236 struct block_device
*bdev
;
1240 * Preserve backwards compatibility and allow large file access
1241 * even if userspace doesn't ask for it explicitly. Some mkfs
1242 * binary needs it. We might want to drop this workaround
1243 * during an unstable branch.
1245 filp
->f_flags
|= O_LARGEFILE
;
1247 bdev
= bd_acquire(inode
);
1251 res
= do_open(bdev
, filp
, 0);
1255 if (!(filp
->f_flags
& O_EXCL
) )
1258 if (!(res
= bd_claim(bdev
, filp
)))
1265 static int __blkdev_put(struct block_device
*bdev
, int for_part
)
1268 struct inode
*bd_inode
= bdev
->bd_inode
;
1269 struct gendisk
*disk
= bdev
->bd_disk
;
1270 struct block_device
*victim
= NULL
;
1272 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1275 bdev
->bd_part_count
--;
1277 if (!--bdev
->bd_openers
) {
1278 sync_blockdev(bdev
);
1281 if (bdev
->bd_contains
== bdev
) {
1282 if (disk
->fops
->release
)
1283 ret
= disk
->fops
->release(bd_inode
, NULL
);
1285 if (!bdev
->bd_openers
) {
1286 struct module
*owner
= disk
->fops
->owner
;
1291 if (bdev
->bd_contains
!= bdev
) {
1292 kobject_put(&bdev
->bd_part
->kobj
);
1293 bdev
->bd_part
= NULL
;
1295 bdev
->bd_disk
= NULL
;
1296 bdev
->bd_inode
->i_data
.backing_dev_info
= &default_backing_dev_info
;
1297 if (bdev
!= bdev
->bd_contains
)
1298 victim
= bdev
->bd_contains
;
1299 bdev
->bd_contains
= NULL
;
1302 mutex_unlock(&bdev
->bd_mutex
);
1305 __blkdev_put(victim
, 1);
1309 int blkdev_put(struct block_device
*bdev
)
1311 return __blkdev_put(bdev
, 0);
1313 EXPORT_SYMBOL(blkdev_put
);
1315 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1317 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1318 if (bdev
->bd_holder
== filp
)
1320 return blkdev_put(bdev
);
1323 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1325 return blkdev_ioctl(file
->f_mapping
->host
, file
, cmd
, arg
);
1328 const struct address_space_operations def_blk_aops
= {
1329 .readpage
= blkdev_readpage
,
1330 .writepage
= blkdev_writepage
,
1331 .sync_page
= block_sync_page
,
1332 .prepare_write
= blkdev_prepare_write
,
1333 .commit_write
= blkdev_commit_write
,
1334 .writepages
= generic_writepages
,
1335 .direct_IO
= blkdev_direct_IO
,
1338 const struct file_operations def_blk_fops
= {
1339 .open
= blkdev_open
,
1340 .release
= blkdev_close
,
1341 .llseek
= block_llseek
,
1342 .read
= do_sync_read
,
1343 .write
= do_sync_write
,
1344 .aio_read
= generic_file_aio_read
,
1345 .aio_write
= generic_file_aio_write_nolock
,
1346 .mmap
= generic_file_mmap
,
1347 .fsync
= block_fsync
,
1348 .unlocked_ioctl
= block_ioctl
,
1349 #ifdef CONFIG_COMPAT
1350 .compat_ioctl
= compat_blkdev_ioctl
,
1352 .sendfile
= generic_file_sendfile
,
1353 .splice_read
= generic_file_splice_read
,
1354 .splice_write
= generic_file_splice_write
,
1357 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1360 mm_segment_t old_fs
= get_fs();
1362 res
= blkdev_ioctl(bdev
->bd_inode
, NULL
, cmd
, arg
);
1367 EXPORT_SYMBOL(ioctl_by_bdev
);
1370 * lookup_bdev - lookup a struct block_device by name
1372 * @path: special file representing the block device
1374 * Get a reference to the blockdevice at @path in the current
1375 * namespace if possible and return it. Return ERR_PTR(error)
1378 struct block_device
*lookup_bdev(const char *path
)
1380 struct block_device
*bdev
;
1381 struct inode
*inode
;
1382 struct nameidata nd
;
1385 if (!path
|| !*path
)
1386 return ERR_PTR(-EINVAL
);
1388 error
= path_lookup(path
, LOOKUP_FOLLOW
, &nd
);
1390 return ERR_PTR(error
);
1392 inode
= nd
.dentry
->d_inode
;
1394 if (!S_ISBLK(inode
->i_mode
))
1397 if (nd
.mnt
->mnt_flags
& MNT_NODEV
)
1400 bdev
= bd_acquire(inode
);
1407 bdev
= ERR_PTR(error
);
1412 * open_bdev_excl - open a block device by name and set it up for use
1414 * @path: special file representing the block device
1415 * @flags: %MS_RDONLY for opening read-only
1416 * @holder: owner for exclusion
1418 * Open the blockdevice described by the special file at @path, claim it
1421 struct block_device
*open_bdev_excl(const char *path
, int flags
, void *holder
)
1423 struct block_device
*bdev
;
1424 mode_t mode
= FMODE_READ
;
1427 bdev
= lookup_bdev(path
);
1431 if (!(flags
& MS_RDONLY
))
1432 mode
|= FMODE_WRITE
;
1433 error
= blkdev_get(bdev
, mode
, 0);
1435 return ERR_PTR(error
);
1437 if (!(flags
& MS_RDONLY
) && bdev_read_only(bdev
))
1439 error
= bd_claim(bdev
, holder
);
1447 return ERR_PTR(error
);
1450 EXPORT_SYMBOL(open_bdev_excl
);
1453 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1455 * @bdev: blockdevice to close
1457 * This is the counterpart to open_bdev_excl().
1459 void close_bdev_excl(struct block_device
*bdev
)
1465 EXPORT_SYMBOL(close_bdev_excl
);
1467 int __invalidate_device(struct block_device
*bdev
)
1469 struct super_block
*sb
= get_super(bdev
);
1474 * no need to lock the super, get_super holds the
1475 * read mutex so the filesystem cannot go away
1476 * under us (->put_super runs with the write lock
1479 shrink_dcache_sb(sb
);
1480 res
= invalidate_inodes(sb
);
1483 invalidate_bdev(bdev
);
1486 EXPORT_SYMBOL(__invalidate_device
);