4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <linux/badblocks.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <asm/uaccess.h>
39 struct block_device bdev
;
40 struct inode vfs_inode
;
43 static const struct address_space_operations def_blk_aops
;
45 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
47 return container_of(inode
, struct bdev_inode
, vfs_inode
);
50 struct block_device
*I_BDEV(struct inode
*inode
)
52 return &BDEV_I(inode
)->bdev
;
54 EXPORT_SYMBOL(I_BDEV
);
56 void __vfs_msg(struct super_block
*sb
, const char *prefix
, const char *fmt
, ...)
64 printk_ratelimited("%sVFS (%s): %pV\n", prefix
, sb
->s_id
, &vaf
);
68 static void bdev_write_inode(struct block_device
*bdev
)
70 struct inode
*inode
= bdev
->bd_inode
;
73 spin_lock(&inode
->i_lock
);
74 while (inode
->i_state
& I_DIRTY
) {
75 spin_unlock(&inode
->i_lock
);
76 ret
= write_inode_now(inode
, true);
78 char name
[BDEVNAME_SIZE
];
79 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
80 "for block device %s (err=%d).\n",
81 bdevname(bdev
, name
), ret
);
83 spin_lock(&inode
->i_lock
);
85 spin_unlock(&inode
->i_lock
);
88 /* Kill _all_ buffers and pagecache , dirty or not.. */
89 void kill_bdev(struct block_device
*bdev
)
91 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
93 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
97 truncate_inode_pages(mapping
, 0);
99 EXPORT_SYMBOL(kill_bdev
);
101 /* Invalidate clean unused buffers and pagecache. */
102 void invalidate_bdev(struct block_device
*bdev
)
104 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
106 if (mapping
->nrpages
== 0)
109 invalidate_bh_lrus();
110 lru_add_drain_all(); /* make sure all lru add caches are flushed */
111 invalidate_mapping_pages(mapping
, 0, -1);
112 /* 99% of the time, we don't need to flush the cleancache on the bdev.
113 * But, for the strange corners, lets be cautious
115 cleancache_invalidate_inode(mapping
);
117 EXPORT_SYMBOL(invalidate_bdev
);
119 int set_blocksize(struct block_device
*bdev
, int size
)
121 /* Size must be a power of two, and between 512 and PAGE_SIZE */
122 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
125 /* Size cannot be smaller than the size supported by the device */
126 if (size
< bdev_logical_block_size(bdev
))
129 /* Don't change the size if it is same as current */
130 if (bdev
->bd_block_size
!= size
) {
132 bdev
->bd_block_size
= size
;
133 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
139 EXPORT_SYMBOL(set_blocksize
);
141 int sb_set_blocksize(struct super_block
*sb
, int size
)
143 if (set_blocksize(sb
->s_bdev
, size
))
145 /* If we get here, we know size is power of two
146 * and it's value is between 512 and PAGE_SIZE */
147 sb
->s_blocksize
= size
;
148 sb
->s_blocksize_bits
= blksize_bits(size
);
149 return sb
->s_blocksize
;
152 EXPORT_SYMBOL(sb_set_blocksize
);
154 int sb_min_blocksize(struct super_block
*sb
, int size
)
156 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
159 return sb_set_blocksize(sb
, size
);
162 EXPORT_SYMBOL(sb_min_blocksize
);
165 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
166 struct buffer_head
*bh
, int create
)
168 bh
->b_bdev
= I_BDEV(inode
);
169 bh
->b_blocknr
= iblock
;
170 set_buffer_mapped(bh
);
174 static struct inode
*bdev_file_inode(struct file
*file
)
176 return file
->f_mapping
->host
;
179 #define DIO_INLINE_BIO_VECS 4
181 static void blkdev_bio_end_io_simple(struct bio
*bio
)
183 struct task_struct
*waiter
= bio
->bi_private
;
185 WRITE_ONCE(bio
->bi_private
, NULL
);
186 wake_up_process(waiter
);
190 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
193 struct file
*file
= iocb
->ki_filp
;
194 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
195 unsigned blkbits
= blksize_bits(bdev_logical_block_size(bdev
));
196 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
197 loff_t pos
= iocb
->ki_pos
;
198 bool should_dirty
= false;
204 if ((pos
| iov_iter_alignment(iter
)) & ((1 << blkbits
) - 1))
207 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
210 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
216 bio
.bi_max_vecs
= nr_pages
;
217 bio
.bi_io_vec
= vecs
;
219 bio
.bi_iter
.bi_sector
= pos
>> blkbits
;
220 bio
.bi_private
= current
;
221 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
223 ret
= bio_iov_iter_get_pages(&bio
, iter
);
226 ret
= bio
.bi_iter
.bi_size
;
228 if (iov_iter_rw(iter
) == READ
) {
229 bio_set_op_attrs(&bio
, REQ_OP_READ
, 0);
230 if (iter_is_iovec(iter
))
233 bio_set_op_attrs(&bio
, REQ_OP_WRITE
, REQ_SYNC
| REQ_IDLE
);
234 task_io_account_write(ret
);
237 qc
= submit_bio(&bio
);
239 set_current_state(TASK_UNINTERRUPTIBLE
);
240 if (!READ_ONCE(bio
.bi_private
))
242 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
243 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
246 __set_current_state(TASK_RUNNING
);
248 bio_for_each_segment_all(bvec
, &bio
, i
) {
249 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
250 set_page_dirty_lock(bvec
->bv_page
);
251 put_page(bvec
->bv_page
);
254 if (vecs
!= inline_vecs
)
257 if (unlikely(bio
.bi_error
))
264 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
266 struct file
*file
= iocb
->ki_filp
;
267 struct inode
*inode
= bdev_file_inode(file
);
270 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
273 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
274 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
275 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
,
276 blkdev_get_block
, NULL
, NULL
,
280 int __sync_blockdev(struct block_device
*bdev
, int wait
)
285 return filemap_flush(bdev
->bd_inode
->i_mapping
);
286 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
290 * Write out and wait upon all the dirty data associated with a block
291 * device via its mapping. Does not take the superblock lock.
293 int sync_blockdev(struct block_device
*bdev
)
295 return __sync_blockdev(bdev
, 1);
297 EXPORT_SYMBOL(sync_blockdev
);
300 * Write out and wait upon all dirty data associated with this
301 * device. Filesystem data as well as the underlying block
302 * device. Takes the superblock lock.
304 int fsync_bdev(struct block_device
*bdev
)
306 struct super_block
*sb
= get_super(bdev
);
308 int res
= sync_filesystem(sb
);
312 return sync_blockdev(bdev
);
314 EXPORT_SYMBOL(fsync_bdev
);
317 * freeze_bdev -- lock a filesystem and force it into a consistent state
318 * @bdev: blockdevice to lock
320 * If a superblock is found on this device, we take the s_umount semaphore
321 * on it to make sure nobody unmounts until the snapshot creation is done.
322 * The reference counter (bd_fsfreeze_count) guarantees that only the last
323 * unfreeze process can unfreeze the frozen filesystem actually when multiple
324 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
325 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
328 struct super_block
*freeze_bdev(struct block_device
*bdev
)
330 struct super_block
*sb
;
333 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
334 if (++bdev
->bd_fsfreeze_count
> 1) {
336 * We don't even need to grab a reference - the first call
337 * to freeze_bdev grab an active reference and only the last
338 * thaw_bdev drops it.
340 sb
= get_super(bdev
);
343 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
347 sb
= get_active_super(bdev
);
350 if (sb
->s_op
->freeze_super
)
351 error
= sb
->s_op
->freeze_super(sb
);
353 error
= freeze_super(sb
);
355 deactivate_super(sb
);
356 bdev
->bd_fsfreeze_count
--;
357 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
358 return ERR_PTR(error
);
360 deactivate_super(sb
);
363 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
364 return sb
; /* thaw_bdev releases s->s_umount */
366 EXPORT_SYMBOL(freeze_bdev
);
369 * thaw_bdev -- unlock filesystem
370 * @bdev: blockdevice to unlock
371 * @sb: associated superblock
373 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
375 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
379 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
380 if (!bdev
->bd_fsfreeze_count
)
384 if (--bdev
->bd_fsfreeze_count
> 0)
390 if (sb
->s_op
->thaw_super
)
391 error
= sb
->s_op
->thaw_super(sb
);
393 error
= thaw_super(sb
);
395 bdev
->bd_fsfreeze_count
++;
397 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
400 EXPORT_SYMBOL(thaw_bdev
);
402 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
404 return block_write_full_page(page
, blkdev_get_block
, wbc
);
407 static int blkdev_readpage(struct file
* file
, struct page
* page
)
409 return block_read_full_page(page
, blkdev_get_block
);
412 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
413 struct list_head
*pages
, unsigned nr_pages
)
415 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
418 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
419 loff_t pos
, unsigned len
, unsigned flags
,
420 struct page
**pagep
, void **fsdata
)
422 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
426 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
427 loff_t pos
, unsigned len
, unsigned copied
,
428 struct page
*page
, void *fsdata
)
431 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
441 * for a block special file file_inode(file)->i_size is zero
442 * so we compute the size by hand (just as in block_read/write above)
444 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
446 struct inode
*bd_inode
= bdev_file_inode(file
);
449 inode_lock(bd_inode
);
450 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
451 inode_unlock(bd_inode
);
455 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
457 struct inode
*bd_inode
= bdev_file_inode(filp
);
458 struct block_device
*bdev
= I_BDEV(bd_inode
);
461 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
466 * There is no need to serialise calls to blkdev_issue_flush with
467 * i_mutex and doing so causes performance issues with concurrent
468 * O_SYNC writers to a block device.
470 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
471 if (error
== -EOPNOTSUPP
)
476 EXPORT_SYMBOL(blkdev_fsync
);
479 * bdev_read_page() - Start reading a page from a block device
480 * @bdev: The device to read the page from
481 * @sector: The offset on the device to read the page to (need not be aligned)
482 * @page: The page to read
484 * On entry, the page should be locked. It will be unlocked when the page
485 * has been read. If the block driver implements rw_page synchronously,
486 * that will be true on exit from this function, but it need not be.
488 * Errors returned by this function are usually "soft", eg out of memory, or
489 * queue full; callers should try a different route to read this page rather
490 * than propagate an error back up the stack.
492 * Return: negative errno if an error occurs, 0 if submission was successful.
494 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
497 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
498 int result
= -EOPNOTSUPP
;
500 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
503 result
= blk_queue_enter(bdev
->bd_queue
, false);
506 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
507 blk_queue_exit(bdev
->bd_queue
);
510 EXPORT_SYMBOL_GPL(bdev_read_page
);
513 * bdev_write_page() - Start writing a page to a block device
514 * @bdev: The device to write the page to
515 * @sector: The offset on the device to write the page to (need not be aligned)
516 * @page: The page to write
517 * @wbc: The writeback_control for the write
519 * On entry, the page should be locked and not currently under writeback.
520 * On exit, if the write started successfully, the page will be unlocked and
521 * under writeback. If the write failed already (eg the driver failed to
522 * queue the page to the device), the page will still be locked. If the
523 * caller is a ->writepage implementation, it will need to unlock the page.
525 * Errors returned by this function are usually "soft", eg out of memory, or
526 * queue full; callers should try a different route to write this page rather
527 * than propagate an error back up the stack.
529 * Return: negative errno if an error occurs, 0 if submission was successful.
531 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
532 struct page
*page
, struct writeback_control
*wbc
)
535 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
537 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
539 result
= blk_queue_enter(bdev
->bd_queue
, false);
543 set_page_writeback(page
);
544 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
546 end_page_writeback(page
);
549 blk_queue_exit(bdev
->bd_queue
);
552 EXPORT_SYMBOL_GPL(bdev_write_page
);
555 * bdev_direct_access() - Get the address for directly-accessibly memory
556 * @bdev: The device containing the memory
557 * @dax: control and output parameters for ->direct_access
559 * If a block device is made up of directly addressable memory, this function
560 * will tell the caller the PFN and the address of the memory. The address
561 * may be directly dereferenced within the kernel without the need to call
562 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
565 * Return: negative errno if an error occurs, otherwise the number of bytes
566 * accessible at this address.
568 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
570 sector_t sector
= dax
->sector
;
571 long avail
, size
= dax
->size
;
572 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
575 * The device driver is allowed to sleep, in order to make the
576 * memory directly accessible.
582 if (!blk_queue_dax(bdev_get_queue(bdev
)) || !ops
->direct_access
)
584 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
585 part_nr_sects_read(bdev
->bd_part
))
587 sector
+= get_start_sect(bdev
);
588 if (sector
% (PAGE_SIZE
/ 512))
590 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
, size
);
593 if (avail
> 0 && avail
& ~PAGE_MASK
)
595 return min(avail
, size
);
597 EXPORT_SYMBOL_GPL(bdev_direct_access
);
600 * bdev_dax_supported() - Check if the device supports dax for filesystem
601 * @sb: The superblock of the device
602 * @blocksize: The block size of the device
604 * This is a library function for filesystems to check if the block device
605 * can be mounted with dax option.
607 * Return: negative errno if unsupported, 0 if supported.
609 int bdev_dax_supported(struct super_block
*sb
, int blocksize
)
611 struct blk_dax_ctl dax
= {
617 if (blocksize
!= PAGE_SIZE
) {
618 vfs_msg(sb
, KERN_ERR
, "error: unsupported blocksize for dax");
622 err
= bdev_direct_access(sb
->s_bdev
, &dax
);
626 vfs_msg(sb
, KERN_ERR
,
627 "error: device does not support dax");
630 vfs_msg(sb
, KERN_ERR
,
631 "error: unaligned partition for dax");
634 vfs_msg(sb
, KERN_ERR
,
635 "error: dax access failed (%d)", err
);
642 EXPORT_SYMBOL_GPL(bdev_dax_supported
);
645 * bdev_dax_capable() - Return if the raw device is capable for dax
646 * @bdev: The device for raw block device access
648 bool bdev_dax_capable(struct block_device
*bdev
)
650 struct blk_dax_ctl dax
= {
654 if (!IS_ENABLED(CONFIG_FS_DAX
))
658 if (bdev_direct_access(bdev
, &dax
) < 0)
661 dax
.sector
= bdev
->bd_part
->nr_sects
- (PAGE_SIZE
/ 512);
662 if (bdev_direct_access(bdev
, &dax
) < 0)
672 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
673 static struct kmem_cache
* bdev_cachep __read_mostly
;
675 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
677 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
680 return &ei
->vfs_inode
;
683 static void bdev_i_callback(struct rcu_head
*head
)
685 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
686 struct bdev_inode
*bdi
= BDEV_I(inode
);
688 kmem_cache_free(bdev_cachep
, bdi
);
691 static void bdev_destroy_inode(struct inode
*inode
)
693 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
696 static void init_once(void *foo
)
698 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
699 struct block_device
*bdev
= &ei
->bdev
;
701 memset(bdev
, 0, sizeof(*bdev
));
702 mutex_init(&bdev
->bd_mutex
);
703 INIT_LIST_HEAD(&bdev
->bd_list
);
705 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
707 inode_init_once(&ei
->vfs_inode
);
708 /* Initialize mutex for freeze. */
709 mutex_init(&bdev
->bd_fsfreeze_mutex
);
712 static void bdev_evict_inode(struct inode
*inode
)
714 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
715 truncate_inode_pages_final(&inode
->i_data
);
716 invalidate_inode_buffers(inode
); /* is it needed here? */
718 spin_lock(&bdev_lock
);
719 list_del_init(&bdev
->bd_list
);
720 spin_unlock(&bdev_lock
);
723 static const struct super_operations bdev_sops
= {
724 .statfs
= simple_statfs
,
725 .alloc_inode
= bdev_alloc_inode
,
726 .destroy_inode
= bdev_destroy_inode
,
727 .drop_inode
= generic_delete_inode
,
728 .evict_inode
= bdev_evict_inode
,
731 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
732 int flags
, const char *dev_name
, void *data
)
735 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
737 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
741 static struct file_system_type bd_type
= {
744 .kill_sb
= kill_anon_super
,
747 struct super_block
*blockdev_superblock __read_mostly
;
748 EXPORT_SYMBOL_GPL(blockdev_superblock
);
750 void __init
bdev_cache_init(void)
753 static struct vfsmount
*bd_mnt
;
755 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
756 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
757 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
759 err
= register_filesystem(&bd_type
);
761 panic("Cannot register bdev pseudo-fs");
762 bd_mnt
= kern_mount(&bd_type
);
764 panic("Cannot create bdev pseudo-fs");
765 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
769 * Most likely _very_ bad one - but then it's hardly critical for small
770 * /dev and can be fixed when somebody will need really large one.
771 * Keep in mind that it will be fed through icache hash function too.
773 static inline unsigned long hash(dev_t dev
)
775 return MAJOR(dev
)+MINOR(dev
);
778 static int bdev_test(struct inode
*inode
, void *data
)
780 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
783 static int bdev_set(struct inode
*inode
, void *data
)
785 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
789 static LIST_HEAD(all_bdevs
);
791 struct block_device
*bdget(dev_t dev
)
793 struct block_device
*bdev
;
796 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
797 bdev_test
, bdev_set
, &dev
);
802 bdev
= &BDEV_I(inode
)->bdev
;
804 if (inode
->i_state
& I_NEW
) {
805 bdev
->bd_contains
= NULL
;
806 bdev
->bd_super
= NULL
;
807 bdev
->bd_inode
= inode
;
808 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
809 bdev
->bd_part_count
= 0;
810 bdev
->bd_invalidated
= 0;
811 inode
->i_mode
= S_IFBLK
;
813 inode
->i_bdev
= bdev
;
814 inode
->i_data
.a_ops
= &def_blk_aops
;
815 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
816 spin_lock(&bdev_lock
);
817 list_add(&bdev
->bd_list
, &all_bdevs
);
818 spin_unlock(&bdev_lock
);
819 unlock_new_inode(inode
);
824 EXPORT_SYMBOL(bdget
);
827 * bdgrab -- Grab a reference to an already referenced block device
828 * @bdev: Block device to grab a reference to.
830 struct block_device
*bdgrab(struct block_device
*bdev
)
832 ihold(bdev
->bd_inode
);
835 EXPORT_SYMBOL(bdgrab
);
837 long nr_blockdev_pages(void)
839 struct block_device
*bdev
;
841 spin_lock(&bdev_lock
);
842 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
843 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
845 spin_unlock(&bdev_lock
);
849 void bdput(struct block_device
*bdev
)
851 iput(bdev
->bd_inode
);
854 EXPORT_SYMBOL(bdput
);
856 static struct block_device
*bd_acquire(struct inode
*inode
)
858 struct block_device
*bdev
;
860 spin_lock(&bdev_lock
);
861 bdev
= inode
->i_bdev
;
864 spin_unlock(&bdev_lock
);
867 spin_unlock(&bdev_lock
);
869 bdev
= bdget(inode
->i_rdev
);
871 spin_lock(&bdev_lock
);
872 if (!inode
->i_bdev
) {
874 * We take an additional reference to bd_inode,
875 * and it's released in clear_inode() of inode.
876 * So, we can access it via ->i_mapping always
880 inode
->i_bdev
= bdev
;
881 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
883 spin_unlock(&bdev_lock
);
888 /* Call when you free inode */
890 void bd_forget(struct inode
*inode
)
892 struct block_device
*bdev
= NULL
;
894 spin_lock(&bdev_lock
);
895 if (!sb_is_blkdev_sb(inode
->i_sb
))
896 bdev
= inode
->i_bdev
;
897 inode
->i_bdev
= NULL
;
898 inode
->i_mapping
= &inode
->i_data
;
899 spin_unlock(&bdev_lock
);
906 * bd_may_claim - test whether a block device can be claimed
907 * @bdev: block device of interest
908 * @whole: whole block device containing @bdev, may equal @bdev
909 * @holder: holder trying to claim @bdev
911 * Test whether @bdev can be claimed by @holder.
914 * spin_lock(&bdev_lock).
917 * %true if @bdev can be claimed, %false otherwise.
919 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
922 if (bdev
->bd_holder
== holder
)
923 return true; /* already a holder */
924 else if (bdev
->bd_holder
!= NULL
)
925 return false; /* held by someone else */
926 else if (bdev
->bd_contains
== bdev
)
927 return true; /* is a whole device which isn't held */
929 else if (whole
->bd_holder
== bd_may_claim
)
930 return true; /* is a partition of a device that is being partitioned */
931 else if (whole
->bd_holder
!= NULL
)
932 return false; /* is a partition of a held device */
934 return true; /* is a partition of an un-held device */
938 * bd_prepare_to_claim - prepare to claim a block device
939 * @bdev: block device of interest
940 * @whole: the whole device containing @bdev, may equal @bdev
941 * @holder: holder trying to claim @bdev
943 * Prepare to claim @bdev. This function fails if @bdev is already
944 * claimed by another holder and waits if another claiming is in
945 * progress. This function doesn't actually claim. On successful
946 * return, the caller has ownership of bd_claiming and bd_holder[s].
949 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
953 * 0 if @bdev can be claimed, -EBUSY otherwise.
955 static int bd_prepare_to_claim(struct block_device
*bdev
,
956 struct block_device
*whole
, void *holder
)
959 /* if someone else claimed, fail */
960 if (!bd_may_claim(bdev
, whole
, holder
))
963 /* if claiming is already in progress, wait for it to finish */
964 if (whole
->bd_claiming
) {
965 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
968 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
969 spin_unlock(&bdev_lock
);
971 finish_wait(wq
, &wait
);
972 spin_lock(&bdev_lock
);
981 * bd_start_claiming - start claiming a block device
982 * @bdev: block device of interest
983 * @holder: holder trying to claim @bdev
985 * @bdev is about to be opened exclusively. Check @bdev can be opened
986 * exclusively and mark that an exclusive open is in progress. Each
987 * successful call to this function must be matched with a call to
988 * either bd_finish_claiming() or bd_abort_claiming() (which do not
991 * This function is used to gain exclusive access to the block device
992 * without actually causing other exclusive open attempts to fail. It
993 * should be used when the open sequence itself requires exclusive
994 * access but may subsequently fail.
1000 * Pointer to the block device containing @bdev on success, ERR_PTR()
1003 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1006 struct gendisk
*disk
;
1007 struct block_device
*whole
;
1013 * @bdev might not have been initialized properly yet, look up
1014 * and grab the outer block device the hard way.
1016 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1018 return ERR_PTR(-ENXIO
);
1021 * Normally, @bdev should equal what's returned from bdget_disk()
1022 * if partno is 0; however, some drivers (floppy) use multiple
1023 * bdev's for the same physical device and @bdev may be one of the
1024 * aliases. Keep @bdev if partno is 0. This means claimer
1025 * tracking is broken for those devices but it has always been that
1029 whole
= bdget_disk(disk
, 0);
1031 whole
= bdgrab(bdev
);
1033 module_put(disk
->fops
->owner
);
1036 return ERR_PTR(-ENOMEM
);
1038 /* prepare to claim, if successful, mark claiming in progress */
1039 spin_lock(&bdev_lock
);
1041 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1043 whole
->bd_claiming
= holder
;
1044 spin_unlock(&bdev_lock
);
1047 spin_unlock(&bdev_lock
);
1049 return ERR_PTR(err
);
1054 struct bd_holder_disk
{
1055 struct list_head list
;
1056 struct gendisk
*disk
;
1060 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1061 struct gendisk
*disk
)
1063 struct bd_holder_disk
*holder
;
1065 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1066 if (holder
->disk
== disk
)
1071 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1073 return sysfs_create_link(from
, to
, kobject_name(to
));
1076 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1078 sysfs_remove_link(from
, kobject_name(to
));
1082 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1083 * @bdev: the claimed slave bdev
1084 * @disk: the holding disk
1086 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1088 * This functions creates the following sysfs symlinks.
1090 * - from "slaves" directory of the holder @disk to the claimed @bdev
1091 * - from "holders" directory of the @bdev to the holder @disk
1093 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1094 * passed to bd_link_disk_holder(), then:
1096 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1097 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1099 * The caller must have claimed @bdev before calling this function and
1100 * ensure that both @bdev and @disk are valid during the creation and
1101 * lifetime of these symlinks.
1107 * 0 on success, -errno on failure.
1109 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1111 struct bd_holder_disk
*holder
;
1114 mutex_lock(&bdev
->bd_mutex
);
1116 WARN_ON_ONCE(!bdev
->bd_holder
);
1118 /* FIXME: remove the following once add_disk() handles errors */
1119 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1122 holder
= bd_find_holder_disk(bdev
, disk
);
1128 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1134 INIT_LIST_HEAD(&holder
->list
);
1135 holder
->disk
= disk
;
1138 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1142 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1146 * bdev could be deleted beneath us which would implicitly destroy
1147 * the holder directory. Hold on to it.
1149 kobject_get(bdev
->bd_part
->holder_dir
);
1151 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1155 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1159 mutex_unlock(&bdev
->bd_mutex
);
1162 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1165 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1166 * @bdev: the calimed slave bdev
1167 * @disk: the holding disk
1169 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1174 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1176 struct bd_holder_disk
*holder
;
1178 mutex_lock(&bdev
->bd_mutex
);
1180 holder
= bd_find_holder_disk(bdev
, disk
);
1182 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1183 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1184 del_symlink(bdev
->bd_part
->holder_dir
,
1185 &disk_to_dev(disk
)->kobj
);
1186 kobject_put(bdev
->bd_part
->holder_dir
);
1187 list_del_init(&holder
->list
);
1191 mutex_unlock(&bdev
->bd_mutex
);
1193 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1197 * flush_disk - invalidates all buffer-cache entries on a disk
1199 * @bdev: struct block device to be flushed
1200 * @kill_dirty: flag to guide handling of dirty inodes
1202 * Invalidates all buffer-cache entries on a disk. It should be called
1203 * when a disk has been changed -- either by a media change or online
1206 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1208 if (__invalidate_device(bdev
, kill_dirty
)) {
1209 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1210 "resized disk %s\n",
1211 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1216 if (disk_part_scan_enabled(bdev
->bd_disk
))
1217 bdev
->bd_invalidated
= 1;
1221 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1222 * @disk: struct gendisk to check
1223 * @bdev: struct bdev to adjust.
1225 * This routine checks to see if the bdev size does not match the disk size
1226 * and adjusts it if it differs.
1228 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1230 loff_t disk_size
, bdev_size
;
1232 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1233 bdev_size
= i_size_read(bdev
->bd_inode
);
1234 if (disk_size
!= bdev_size
) {
1236 "%s: detected capacity change from %lld to %lld\n",
1237 disk
->disk_name
, bdev_size
, disk_size
);
1238 i_size_write(bdev
->bd_inode
, disk_size
);
1239 flush_disk(bdev
, false);
1242 EXPORT_SYMBOL(check_disk_size_change
);
1245 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1246 * @disk: struct gendisk to be revalidated
1248 * This routine is a wrapper for lower-level driver's revalidate_disk
1249 * call-backs. It is used to do common pre and post operations needed
1250 * for all revalidate_disk operations.
1252 int revalidate_disk(struct gendisk
*disk
)
1254 struct block_device
*bdev
;
1257 if (disk
->fops
->revalidate_disk
)
1258 ret
= disk
->fops
->revalidate_disk(disk
);
1259 blk_integrity_revalidate(disk
);
1260 bdev
= bdget_disk(disk
, 0);
1264 mutex_lock(&bdev
->bd_mutex
);
1265 check_disk_size_change(disk
, bdev
);
1266 bdev
->bd_invalidated
= 0;
1267 mutex_unlock(&bdev
->bd_mutex
);
1271 EXPORT_SYMBOL(revalidate_disk
);
1274 * This routine checks whether a removable media has been changed,
1275 * and invalidates all buffer-cache-entries in that case. This
1276 * is a relatively slow routine, so we have to try to minimize using
1277 * it. Thus it is called only upon a 'mount' or 'open'. This
1278 * is the best way of combining speed and utility, I think.
1279 * People changing diskettes in the middle of an operation deserve
1282 int check_disk_change(struct block_device
*bdev
)
1284 struct gendisk
*disk
= bdev
->bd_disk
;
1285 const struct block_device_operations
*bdops
= disk
->fops
;
1286 unsigned int events
;
1288 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1289 DISK_EVENT_EJECT_REQUEST
);
1290 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1293 flush_disk(bdev
, true);
1294 if (bdops
->revalidate_disk
)
1295 bdops
->revalidate_disk(bdev
->bd_disk
);
1299 EXPORT_SYMBOL(check_disk_change
);
1301 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1303 unsigned bsize
= bdev_logical_block_size(bdev
);
1305 inode_lock(bdev
->bd_inode
);
1306 i_size_write(bdev
->bd_inode
, size
);
1307 inode_unlock(bdev
->bd_inode
);
1308 while (bsize
< PAGE_SIZE
) {
1313 bdev
->bd_block_size
= bsize
;
1314 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1316 EXPORT_SYMBOL(bd_set_size
);
1318 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1323 * mutex_lock(part->bd_mutex)
1324 * mutex_lock_nested(whole->bd_mutex, 1)
1327 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1329 struct gendisk
*disk
;
1330 struct module
*owner
;
1335 if (mode
& FMODE_READ
)
1337 if (mode
& FMODE_WRITE
)
1340 * hooks: /n/, see "layering violations".
1343 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1353 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1356 owner
= disk
->fops
->owner
;
1358 disk_block_events(disk
);
1359 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1360 if (!bdev
->bd_openers
) {
1361 bdev
->bd_disk
= disk
;
1362 bdev
->bd_queue
= disk
->queue
;
1363 bdev
->bd_contains
= bdev
;
1367 bdev
->bd_part
= disk_get_part(disk
, partno
);
1372 if (disk
->fops
->open
) {
1373 ret
= disk
->fops
->open(bdev
, mode
);
1374 if (ret
== -ERESTARTSYS
) {
1375 /* Lost a race with 'disk' being
1376 * deleted, try again.
1379 disk_put_part(bdev
->bd_part
);
1380 bdev
->bd_part
= NULL
;
1381 bdev
->bd_disk
= NULL
;
1382 bdev
->bd_queue
= NULL
;
1383 mutex_unlock(&bdev
->bd_mutex
);
1384 disk_unblock_events(disk
);
1392 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1395 * If the device is invalidated, rescan partition
1396 * if open succeeded or failed with -ENOMEDIUM.
1397 * The latter is necessary to prevent ghost
1398 * partitions on a removed medium.
1400 if (bdev
->bd_invalidated
) {
1402 rescan_partitions(disk
, bdev
);
1403 else if (ret
== -ENOMEDIUM
)
1404 invalidate_partitions(disk
, bdev
);
1410 struct block_device
*whole
;
1411 whole
= bdget_disk(disk
, 0);
1416 ret
= __blkdev_get(whole
, mode
, 1);
1419 bdev
->bd_contains
= whole
;
1420 bdev
->bd_part
= disk_get_part(disk
, partno
);
1421 if (!(disk
->flags
& GENHD_FL_UP
) ||
1422 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1426 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1429 if (bdev
->bd_contains
== bdev
) {
1431 if (bdev
->bd_disk
->fops
->open
)
1432 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1433 /* the same as first opener case, read comment there */
1434 if (bdev
->bd_invalidated
) {
1436 rescan_partitions(bdev
->bd_disk
, bdev
);
1437 else if (ret
== -ENOMEDIUM
)
1438 invalidate_partitions(bdev
->bd_disk
, bdev
);
1441 goto out_unlock_bdev
;
1443 /* only one opener holds refs to the module and disk */
1449 bdev
->bd_part_count
++;
1450 mutex_unlock(&bdev
->bd_mutex
);
1451 disk_unblock_events(disk
);
1455 disk_put_part(bdev
->bd_part
);
1456 bdev
->bd_disk
= NULL
;
1457 bdev
->bd_part
= NULL
;
1458 bdev
->bd_queue
= NULL
;
1459 if (bdev
!= bdev
->bd_contains
)
1460 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1461 bdev
->bd_contains
= NULL
;
1463 mutex_unlock(&bdev
->bd_mutex
);
1464 disk_unblock_events(disk
);
1474 * blkdev_get - open a block device
1475 * @bdev: block_device to open
1476 * @mode: FMODE_* mask
1477 * @holder: exclusive holder identifier
1479 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1480 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1481 * @holder is invalid. Exclusive opens may nest for the same @holder.
1483 * On success, the reference count of @bdev is unchanged. On failure,
1490 * 0 on success, -errno on failure.
1492 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1494 struct block_device
*whole
= NULL
;
1497 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1499 if ((mode
& FMODE_EXCL
) && holder
) {
1500 whole
= bd_start_claiming(bdev
, holder
);
1501 if (IS_ERR(whole
)) {
1503 return PTR_ERR(whole
);
1507 res
= __blkdev_get(bdev
, mode
, 0);
1510 struct gendisk
*disk
= whole
->bd_disk
;
1512 /* finish claiming */
1513 mutex_lock(&bdev
->bd_mutex
);
1514 spin_lock(&bdev_lock
);
1517 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1519 * Note that for a whole device bd_holders
1520 * will be incremented twice, and bd_holder
1521 * will be set to bd_may_claim before being
1524 whole
->bd_holders
++;
1525 whole
->bd_holder
= bd_may_claim
;
1527 bdev
->bd_holder
= holder
;
1530 /* tell others that we're done */
1531 BUG_ON(whole
->bd_claiming
!= holder
);
1532 whole
->bd_claiming
= NULL
;
1533 wake_up_bit(&whole
->bd_claiming
, 0);
1535 spin_unlock(&bdev_lock
);
1538 * Block event polling for write claims if requested. Any
1539 * write holder makes the write_holder state stick until
1540 * all are released. This is good enough and tracking
1541 * individual writeable reference is too fragile given the
1542 * way @mode is used in blkdev_get/put().
1544 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1545 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1546 bdev
->bd_write_holder
= true;
1547 disk_block_events(disk
);
1550 mutex_unlock(&bdev
->bd_mutex
);
1556 EXPORT_SYMBOL(blkdev_get
);
1559 * blkdev_get_by_path - open a block device by name
1560 * @path: path to the block device to open
1561 * @mode: FMODE_* mask
1562 * @holder: exclusive holder identifier
1564 * Open the blockdevice described by the device file at @path. @mode
1565 * and @holder are identical to blkdev_get().
1567 * On success, the returned block_device has reference count of one.
1573 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1575 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1578 struct block_device
*bdev
;
1581 bdev
= lookup_bdev(path
);
1585 err
= blkdev_get(bdev
, mode
, holder
);
1587 return ERR_PTR(err
);
1589 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1590 blkdev_put(bdev
, mode
);
1591 return ERR_PTR(-EACCES
);
1596 EXPORT_SYMBOL(blkdev_get_by_path
);
1599 * blkdev_get_by_dev - open a block device by device number
1600 * @dev: device number of block device to open
1601 * @mode: FMODE_* mask
1602 * @holder: exclusive holder identifier
1604 * Open the blockdevice described by device number @dev. @mode and
1605 * @holder are identical to blkdev_get().
1607 * Use it ONLY if you really do not have anything better - i.e. when
1608 * you are behind a truly sucky interface and all you are given is a
1609 * device number. _Never_ to be used for internal purposes. If you
1610 * ever need it - reconsider your API.
1612 * On success, the returned block_device has reference count of one.
1618 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1620 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1622 struct block_device
*bdev
;
1627 return ERR_PTR(-ENOMEM
);
1629 err
= blkdev_get(bdev
, mode
, holder
);
1631 return ERR_PTR(err
);
1635 EXPORT_SYMBOL(blkdev_get_by_dev
);
1637 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1639 struct block_device
*bdev
;
1642 * Preserve backwards compatibility and allow large file access
1643 * even if userspace doesn't ask for it explicitly. Some mkfs
1644 * binary needs it. We might want to drop this workaround
1645 * during an unstable branch.
1647 filp
->f_flags
|= O_LARGEFILE
;
1649 if (filp
->f_flags
& O_NDELAY
)
1650 filp
->f_mode
|= FMODE_NDELAY
;
1651 if (filp
->f_flags
& O_EXCL
)
1652 filp
->f_mode
|= FMODE_EXCL
;
1653 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1654 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1656 bdev
= bd_acquire(inode
);
1660 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1662 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1665 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1667 struct gendisk
*disk
= bdev
->bd_disk
;
1668 struct block_device
*victim
= NULL
;
1670 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1672 bdev
->bd_part_count
--;
1674 if (!--bdev
->bd_openers
) {
1675 WARN_ON_ONCE(bdev
->bd_holders
);
1676 sync_blockdev(bdev
);
1679 bdev_write_inode(bdev
);
1681 * Detaching bdev inode from its wb in __destroy_inode()
1682 * is too late: the queue which embeds its bdi (along with
1683 * root wb) can be gone as soon as we put_disk() below.
1685 inode_detach_wb(bdev
->bd_inode
);
1687 if (bdev
->bd_contains
== bdev
) {
1688 if (disk
->fops
->release
)
1689 disk
->fops
->release(disk
, mode
);
1691 if (!bdev
->bd_openers
) {
1692 struct module
*owner
= disk
->fops
->owner
;
1694 disk_put_part(bdev
->bd_part
);
1695 bdev
->bd_part
= NULL
;
1696 bdev
->bd_disk
= NULL
;
1697 if (bdev
!= bdev
->bd_contains
)
1698 victim
= bdev
->bd_contains
;
1699 bdev
->bd_contains
= NULL
;
1704 mutex_unlock(&bdev
->bd_mutex
);
1707 __blkdev_put(victim
, mode
, 1);
1710 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1712 mutex_lock(&bdev
->bd_mutex
);
1714 if (mode
& FMODE_EXCL
) {
1718 * Release a claim on the device. The holder fields
1719 * are protected with bdev_lock. bd_mutex is to
1720 * synchronize disk_holder unlinking.
1722 spin_lock(&bdev_lock
);
1724 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1725 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1727 /* bd_contains might point to self, check in a separate step */
1728 if ((bdev_free
= !bdev
->bd_holders
))
1729 bdev
->bd_holder
= NULL
;
1730 if (!bdev
->bd_contains
->bd_holders
)
1731 bdev
->bd_contains
->bd_holder
= NULL
;
1733 spin_unlock(&bdev_lock
);
1736 * If this was the last claim, remove holder link and
1737 * unblock evpoll if it was a write holder.
1739 if (bdev_free
&& bdev
->bd_write_holder
) {
1740 disk_unblock_events(bdev
->bd_disk
);
1741 bdev
->bd_write_holder
= false;
1746 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1747 * event. This is to ensure detection of media removal commanded
1748 * from userland - e.g. eject(1).
1750 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1752 mutex_unlock(&bdev
->bd_mutex
);
1754 __blkdev_put(bdev
, mode
, 0);
1756 EXPORT_SYMBOL(blkdev_put
);
1758 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1760 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1761 blkdev_put(bdev
, filp
->f_mode
);
1765 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1767 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1768 fmode_t mode
= file
->f_mode
;
1771 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1772 * to updated it before every ioctl.
1774 if (file
->f_flags
& O_NDELAY
)
1775 mode
|= FMODE_NDELAY
;
1777 mode
&= ~FMODE_NDELAY
;
1779 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1783 * Write data to the block device. Only intended for the block device itself
1784 * and the raw driver which basically is a fake block device.
1786 * Does not take i_mutex for the write and thus is not for general purpose
1789 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1791 struct file
*file
= iocb
->ki_filp
;
1792 struct inode
*bd_inode
= bdev_file_inode(file
);
1793 loff_t size
= i_size_read(bd_inode
);
1794 struct blk_plug plug
;
1797 if (bdev_read_only(I_BDEV(bd_inode
)))
1800 if (!iov_iter_count(from
))
1803 if (iocb
->ki_pos
>= size
)
1806 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1808 blk_start_plug(&plug
);
1809 ret
= __generic_file_write_iter(iocb
, from
);
1811 ret
= generic_write_sync(iocb
, ret
);
1812 blk_finish_plug(&plug
);
1815 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1817 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1819 struct file
*file
= iocb
->ki_filp
;
1820 struct inode
*bd_inode
= bdev_file_inode(file
);
1821 loff_t size
= i_size_read(bd_inode
);
1822 loff_t pos
= iocb
->ki_pos
;
1828 iov_iter_truncate(to
, size
);
1829 return generic_file_read_iter(iocb
, to
);
1831 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1834 * Try to release a page associated with block device when the system
1835 * is under memory pressure.
1837 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1839 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1841 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1842 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1844 return try_to_free_buffers(page
);
1847 static int blkdev_writepages(struct address_space
*mapping
,
1848 struct writeback_control
*wbc
)
1850 if (dax_mapping(mapping
)) {
1851 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1853 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1855 return generic_writepages(mapping
, wbc
);
1858 static const struct address_space_operations def_blk_aops
= {
1859 .readpage
= blkdev_readpage
,
1860 .readpages
= blkdev_readpages
,
1861 .writepage
= blkdev_writepage
,
1862 .write_begin
= blkdev_write_begin
,
1863 .write_end
= blkdev_write_end
,
1864 .writepages
= blkdev_writepages
,
1865 .releasepage
= blkdev_releasepage
,
1866 .direct_IO
= blkdev_direct_IO
,
1867 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1870 #define BLKDEV_FALLOC_FL_SUPPORTED \
1871 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1872 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1874 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
1877 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1878 struct request_queue
*q
= bdev_get_queue(bdev
);
1879 struct address_space
*mapping
;
1880 loff_t end
= start
+ len
- 1;
1884 /* Fail if we don't recognize the flags. */
1885 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
1888 /* Don't go off the end of the device. */
1889 isize
= i_size_read(bdev
->bd_inode
);
1893 if (mode
& FALLOC_FL_KEEP_SIZE
) {
1894 len
= isize
- start
;
1895 end
= start
+ len
- 1;
1901 * Don't allow IO that isn't aligned to logical block size.
1903 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
1906 /* Invalidate the page cache, including dirty pages. */
1907 mapping
= bdev
->bd_inode
->i_mapping
;
1908 truncate_inode_pages_range(mapping
, start
, end
);
1911 case FALLOC_FL_ZERO_RANGE
:
1912 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
1913 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
1916 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
1917 /* Only punch if the device can do zeroing discard. */
1918 if (!blk_queue_discard(q
) || !q
->limits
.discard_zeroes_data
)
1920 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
1923 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
1924 if (!blk_queue_discard(q
))
1926 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
1936 * Invalidate again; if someone wandered in and dirtied a page,
1937 * the caller will be given -EBUSY. The third argument is
1938 * inclusive, so the rounding here is safe.
1940 return invalidate_inode_pages2_range(mapping
,
1941 start
>> PAGE_SHIFT
,
1945 const struct file_operations def_blk_fops
= {
1946 .open
= blkdev_open
,
1947 .release
= blkdev_close
,
1948 .llseek
= block_llseek
,
1949 .read_iter
= blkdev_read_iter
,
1950 .write_iter
= blkdev_write_iter
,
1951 .mmap
= generic_file_mmap
,
1952 .fsync
= blkdev_fsync
,
1953 .unlocked_ioctl
= block_ioctl
,
1954 #ifdef CONFIG_COMPAT
1955 .compat_ioctl
= compat_blkdev_ioctl
,
1957 .splice_read
= generic_file_splice_read
,
1958 .splice_write
= iter_file_splice_write
,
1959 .fallocate
= blkdev_fallocate
,
1962 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1965 mm_segment_t old_fs
= get_fs();
1967 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1972 EXPORT_SYMBOL(ioctl_by_bdev
);
1975 * lookup_bdev - lookup a struct block_device by name
1976 * @pathname: special file representing the block device
1978 * Get a reference to the blockdevice at @pathname in the current
1979 * namespace if possible and return it. Return ERR_PTR(error)
1982 struct block_device
*lookup_bdev(const char *pathname
)
1984 struct block_device
*bdev
;
1985 struct inode
*inode
;
1989 if (!pathname
|| !*pathname
)
1990 return ERR_PTR(-EINVAL
);
1992 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1994 return ERR_PTR(error
);
1996 inode
= d_backing_inode(path
.dentry
);
1998 if (!S_ISBLK(inode
->i_mode
))
2001 if (!may_open_dev(&path
))
2004 bdev
= bd_acquire(inode
);
2011 bdev
= ERR_PTR(error
);
2014 EXPORT_SYMBOL(lookup_bdev
);
2016 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2018 struct super_block
*sb
= get_super(bdev
);
2023 * no need to lock the super, get_super holds the
2024 * read mutex so the filesystem cannot go away
2025 * under us (->put_super runs with the write lock
2028 shrink_dcache_sb(sb
);
2029 res
= invalidate_inodes(sb
, kill_dirty
);
2032 invalidate_bdev(bdev
);
2035 EXPORT_SYMBOL(__invalidate_device
);
2037 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2039 struct inode
*inode
, *old_inode
= NULL
;
2041 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2042 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2043 struct address_space
*mapping
= inode
->i_mapping
;
2045 spin_lock(&inode
->i_lock
);
2046 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2047 mapping
->nrpages
== 0) {
2048 spin_unlock(&inode
->i_lock
);
2052 spin_unlock(&inode
->i_lock
);
2053 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2055 * We hold a reference to 'inode' so it couldn't have been
2056 * removed from s_inodes list while we dropped the
2057 * s_inode_list_lock We cannot iput the inode now as we can
2058 * be holding the last reference and we cannot iput it under
2059 * s_inode_list_lock. So we keep the reference and iput it
2065 func(I_BDEV(inode
), arg
);
2067 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2069 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);