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 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
181 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
183 /* avoid the need for a I/O completion work item */
184 if (iocb
->ki_flags
& IOCB_DSYNC
)
189 #define DIO_INLINE_BIO_VECS 4
191 static void blkdev_bio_end_io_simple(struct bio
*bio
)
193 struct task_struct
*waiter
= bio
->bi_private
;
195 WRITE_ONCE(bio
->bi_private
, NULL
);
196 wake_up_process(waiter
);
200 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
203 struct file
*file
= iocb
->ki_filp
;
204 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
205 unsigned blkbits
= blksize_bits(bdev_logical_block_size(bdev
));
206 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
207 loff_t pos
= iocb
->ki_pos
;
208 bool should_dirty
= false;
214 if ((pos
| iov_iter_alignment(iter
)) & ((1 << blkbits
) - 1))
217 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
220 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
226 bio
.bi_max_vecs
= nr_pages
;
227 bio
.bi_io_vec
= vecs
;
229 bio
.bi_iter
.bi_sector
= pos
>> blkbits
;
230 bio
.bi_private
= current
;
231 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
233 ret
= bio_iov_iter_get_pages(&bio
, iter
);
236 ret
= bio
.bi_iter
.bi_size
;
238 if (iov_iter_rw(iter
) == READ
) {
239 bio
.bi_opf
= REQ_OP_READ
;
240 if (iter_is_iovec(iter
))
243 bio
.bi_opf
= dio_bio_write_op(iocb
);
244 task_io_account_write(ret
);
247 qc
= submit_bio(&bio
);
249 set_current_state(TASK_UNINTERRUPTIBLE
);
250 if (!READ_ONCE(bio
.bi_private
))
252 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
253 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
256 __set_current_state(TASK_RUNNING
);
258 bio_for_each_segment_all(bvec
, &bio
, i
) {
259 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
260 set_page_dirty_lock(bvec
->bv_page
);
261 put_page(bvec
->bv_page
);
264 if (vecs
!= inline_vecs
)
267 if (unlikely(bio
.bi_error
))
274 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
276 struct file
*file
= iocb
->ki_filp
;
277 struct inode
*inode
= bdev_file_inode(file
);
280 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
283 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
284 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
285 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
,
286 blkdev_get_block
, NULL
, NULL
,
290 int __sync_blockdev(struct block_device
*bdev
, int wait
)
295 return filemap_flush(bdev
->bd_inode
->i_mapping
);
296 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
300 * Write out and wait upon all the dirty data associated with a block
301 * device via its mapping. Does not take the superblock lock.
303 int sync_blockdev(struct block_device
*bdev
)
305 return __sync_blockdev(bdev
, 1);
307 EXPORT_SYMBOL(sync_blockdev
);
310 * Write out and wait upon all dirty data associated with this
311 * device. Filesystem data as well as the underlying block
312 * device. Takes the superblock lock.
314 int fsync_bdev(struct block_device
*bdev
)
316 struct super_block
*sb
= get_super(bdev
);
318 int res
= sync_filesystem(sb
);
322 return sync_blockdev(bdev
);
324 EXPORT_SYMBOL(fsync_bdev
);
327 * freeze_bdev -- lock a filesystem and force it into a consistent state
328 * @bdev: blockdevice to lock
330 * If a superblock is found on this device, we take the s_umount semaphore
331 * on it to make sure nobody unmounts until the snapshot creation is done.
332 * The reference counter (bd_fsfreeze_count) guarantees that only the last
333 * unfreeze process can unfreeze the frozen filesystem actually when multiple
334 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
335 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
338 struct super_block
*freeze_bdev(struct block_device
*bdev
)
340 struct super_block
*sb
;
343 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
344 if (++bdev
->bd_fsfreeze_count
> 1) {
346 * We don't even need to grab a reference - the first call
347 * to freeze_bdev grab an active reference and only the last
348 * thaw_bdev drops it.
350 sb
= get_super(bdev
);
353 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
357 sb
= get_active_super(bdev
);
360 if (sb
->s_op
->freeze_super
)
361 error
= sb
->s_op
->freeze_super(sb
);
363 error
= freeze_super(sb
);
365 deactivate_super(sb
);
366 bdev
->bd_fsfreeze_count
--;
367 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
368 return ERR_PTR(error
);
370 deactivate_super(sb
);
373 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
374 return sb
; /* thaw_bdev releases s->s_umount */
376 EXPORT_SYMBOL(freeze_bdev
);
379 * thaw_bdev -- unlock filesystem
380 * @bdev: blockdevice to unlock
381 * @sb: associated superblock
383 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
385 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
389 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
390 if (!bdev
->bd_fsfreeze_count
)
394 if (--bdev
->bd_fsfreeze_count
> 0)
400 if (sb
->s_op
->thaw_super
)
401 error
= sb
->s_op
->thaw_super(sb
);
403 error
= thaw_super(sb
);
405 bdev
->bd_fsfreeze_count
++;
407 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
410 EXPORT_SYMBOL(thaw_bdev
);
412 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
414 return block_write_full_page(page
, blkdev_get_block
, wbc
);
417 static int blkdev_readpage(struct file
* file
, struct page
* page
)
419 return block_read_full_page(page
, blkdev_get_block
);
422 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
423 struct list_head
*pages
, unsigned nr_pages
)
425 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
428 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
429 loff_t pos
, unsigned len
, unsigned flags
,
430 struct page
**pagep
, void **fsdata
)
432 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
436 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
437 loff_t pos
, unsigned len
, unsigned copied
,
438 struct page
*page
, void *fsdata
)
441 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
451 * for a block special file file_inode(file)->i_size is zero
452 * so we compute the size by hand (just as in block_read/write above)
454 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
456 struct inode
*bd_inode
= bdev_file_inode(file
);
459 inode_lock(bd_inode
);
460 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
461 inode_unlock(bd_inode
);
465 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
467 struct inode
*bd_inode
= bdev_file_inode(filp
);
468 struct block_device
*bdev
= I_BDEV(bd_inode
);
471 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
476 * There is no need to serialise calls to blkdev_issue_flush with
477 * i_mutex and doing so causes performance issues with concurrent
478 * O_SYNC writers to a block device.
480 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
481 if (error
== -EOPNOTSUPP
)
486 EXPORT_SYMBOL(blkdev_fsync
);
489 * bdev_read_page() - Start reading a page from a block device
490 * @bdev: The device to read the page from
491 * @sector: The offset on the device to read the page to (need not be aligned)
492 * @page: The page to read
494 * On entry, the page should be locked. It will be unlocked when the page
495 * has been read. If the block driver implements rw_page synchronously,
496 * that will be true on exit from this function, but it need not be.
498 * Errors returned by this function are usually "soft", eg out of memory, or
499 * queue full; callers should try a different route to read this page rather
500 * than propagate an error back up the stack.
502 * Return: negative errno if an error occurs, 0 if submission was successful.
504 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
507 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
508 int result
= -EOPNOTSUPP
;
510 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
513 result
= blk_queue_enter(bdev
->bd_queue
, false);
516 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
517 blk_queue_exit(bdev
->bd_queue
);
520 EXPORT_SYMBOL_GPL(bdev_read_page
);
523 * bdev_write_page() - Start writing a page to a block device
524 * @bdev: The device to write the page to
525 * @sector: The offset on the device to write the page to (need not be aligned)
526 * @page: The page to write
527 * @wbc: The writeback_control for the write
529 * On entry, the page should be locked and not currently under writeback.
530 * On exit, if the write started successfully, the page will be unlocked and
531 * under writeback. If the write failed already (eg the driver failed to
532 * queue the page to the device), the page will still be locked. If the
533 * caller is a ->writepage implementation, it will need to unlock the page.
535 * Errors returned by this function are usually "soft", eg out of memory, or
536 * queue full; callers should try a different route to write this page rather
537 * than propagate an error back up the stack.
539 * Return: negative errno if an error occurs, 0 if submission was successful.
541 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
542 struct page
*page
, struct writeback_control
*wbc
)
545 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
547 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
549 result
= blk_queue_enter(bdev
->bd_queue
, false);
553 set_page_writeback(page
);
554 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
556 end_page_writeback(page
);
559 blk_queue_exit(bdev
->bd_queue
);
562 EXPORT_SYMBOL_GPL(bdev_write_page
);
565 * bdev_direct_access() - Get the address for directly-accessibly memory
566 * @bdev: The device containing the memory
567 * @dax: control and output parameters for ->direct_access
569 * If a block device is made up of directly addressable memory, this function
570 * will tell the caller the PFN and the address of the memory. The address
571 * may be directly dereferenced within the kernel without the need to call
572 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
575 * Return: negative errno if an error occurs, otherwise the number of bytes
576 * accessible at this address.
578 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
580 sector_t sector
= dax
->sector
;
581 long avail
, size
= dax
->size
;
582 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
585 * The device driver is allowed to sleep, in order to make the
586 * memory directly accessible.
592 if (!blk_queue_dax(bdev_get_queue(bdev
)) || !ops
->direct_access
)
594 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
595 part_nr_sects_read(bdev
->bd_part
))
597 sector
+= get_start_sect(bdev
);
598 if (sector
% (PAGE_SIZE
/ 512))
600 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
, size
);
603 if (avail
> 0 && avail
& ~PAGE_MASK
)
605 return min(avail
, size
);
607 EXPORT_SYMBOL_GPL(bdev_direct_access
);
610 * bdev_dax_supported() - Check if the device supports dax for filesystem
611 * @sb: The superblock of the device
612 * @blocksize: The block size of the device
614 * This is a library function for filesystems to check if the block device
615 * can be mounted with dax option.
617 * Return: negative errno if unsupported, 0 if supported.
619 int bdev_dax_supported(struct super_block
*sb
, int blocksize
)
621 struct blk_dax_ctl dax
= {
627 if (blocksize
!= PAGE_SIZE
) {
628 vfs_msg(sb
, KERN_ERR
, "error: unsupported blocksize for dax");
632 err
= bdev_direct_access(sb
->s_bdev
, &dax
);
636 vfs_msg(sb
, KERN_ERR
,
637 "error: device does not support dax");
640 vfs_msg(sb
, KERN_ERR
,
641 "error: unaligned partition for dax");
644 vfs_msg(sb
, KERN_ERR
,
645 "error: dax access failed (%d)", err
);
652 EXPORT_SYMBOL_GPL(bdev_dax_supported
);
655 * bdev_dax_capable() - Return if the raw device is capable for dax
656 * @bdev: The device for raw block device access
658 bool bdev_dax_capable(struct block_device
*bdev
)
660 struct blk_dax_ctl dax
= {
664 if (!IS_ENABLED(CONFIG_FS_DAX
))
668 if (bdev_direct_access(bdev
, &dax
) < 0)
671 dax
.sector
= bdev
->bd_part
->nr_sects
- (PAGE_SIZE
/ 512);
672 if (bdev_direct_access(bdev
, &dax
) < 0)
682 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
683 static struct kmem_cache
* bdev_cachep __read_mostly
;
685 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
687 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
690 return &ei
->vfs_inode
;
693 static void bdev_i_callback(struct rcu_head
*head
)
695 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
696 struct bdev_inode
*bdi
= BDEV_I(inode
);
698 kmem_cache_free(bdev_cachep
, bdi
);
701 static void bdev_destroy_inode(struct inode
*inode
)
703 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
706 static void init_once(void *foo
)
708 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
709 struct block_device
*bdev
= &ei
->bdev
;
711 memset(bdev
, 0, sizeof(*bdev
));
712 mutex_init(&bdev
->bd_mutex
);
713 INIT_LIST_HEAD(&bdev
->bd_list
);
715 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
717 inode_init_once(&ei
->vfs_inode
);
718 /* Initialize mutex for freeze. */
719 mutex_init(&bdev
->bd_fsfreeze_mutex
);
722 static void bdev_evict_inode(struct inode
*inode
)
724 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
725 truncate_inode_pages_final(&inode
->i_data
);
726 invalidate_inode_buffers(inode
); /* is it needed here? */
728 spin_lock(&bdev_lock
);
729 list_del_init(&bdev
->bd_list
);
730 spin_unlock(&bdev_lock
);
733 static const struct super_operations bdev_sops
= {
734 .statfs
= simple_statfs
,
735 .alloc_inode
= bdev_alloc_inode
,
736 .destroy_inode
= bdev_destroy_inode
,
737 .drop_inode
= generic_delete_inode
,
738 .evict_inode
= bdev_evict_inode
,
741 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
742 int flags
, const char *dev_name
, void *data
)
745 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
747 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
751 static struct file_system_type bd_type
= {
754 .kill_sb
= kill_anon_super
,
757 struct super_block
*blockdev_superblock __read_mostly
;
758 EXPORT_SYMBOL_GPL(blockdev_superblock
);
760 void __init
bdev_cache_init(void)
763 static struct vfsmount
*bd_mnt
;
765 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
766 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
767 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
769 err
= register_filesystem(&bd_type
);
771 panic("Cannot register bdev pseudo-fs");
772 bd_mnt
= kern_mount(&bd_type
);
774 panic("Cannot create bdev pseudo-fs");
775 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
779 * Most likely _very_ bad one - but then it's hardly critical for small
780 * /dev and can be fixed when somebody will need really large one.
781 * Keep in mind that it will be fed through icache hash function too.
783 static inline unsigned long hash(dev_t dev
)
785 return MAJOR(dev
)+MINOR(dev
);
788 static int bdev_test(struct inode
*inode
, void *data
)
790 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
793 static int bdev_set(struct inode
*inode
, void *data
)
795 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
799 static LIST_HEAD(all_bdevs
);
801 struct block_device
*bdget(dev_t dev
)
803 struct block_device
*bdev
;
806 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
807 bdev_test
, bdev_set
, &dev
);
812 bdev
= &BDEV_I(inode
)->bdev
;
814 if (inode
->i_state
& I_NEW
) {
815 bdev
->bd_contains
= NULL
;
816 bdev
->bd_super
= NULL
;
817 bdev
->bd_inode
= inode
;
818 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
819 bdev
->bd_part_count
= 0;
820 bdev
->bd_invalidated
= 0;
821 inode
->i_mode
= S_IFBLK
;
823 inode
->i_bdev
= bdev
;
824 inode
->i_data
.a_ops
= &def_blk_aops
;
825 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
826 spin_lock(&bdev_lock
);
827 list_add(&bdev
->bd_list
, &all_bdevs
);
828 spin_unlock(&bdev_lock
);
829 unlock_new_inode(inode
);
834 EXPORT_SYMBOL(bdget
);
837 * bdgrab -- Grab a reference to an already referenced block device
838 * @bdev: Block device to grab a reference to.
840 struct block_device
*bdgrab(struct block_device
*bdev
)
842 ihold(bdev
->bd_inode
);
845 EXPORT_SYMBOL(bdgrab
);
847 long nr_blockdev_pages(void)
849 struct block_device
*bdev
;
851 spin_lock(&bdev_lock
);
852 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
853 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
855 spin_unlock(&bdev_lock
);
859 void bdput(struct block_device
*bdev
)
861 iput(bdev
->bd_inode
);
864 EXPORT_SYMBOL(bdput
);
866 static struct block_device
*bd_acquire(struct inode
*inode
)
868 struct block_device
*bdev
;
870 spin_lock(&bdev_lock
);
871 bdev
= inode
->i_bdev
;
874 spin_unlock(&bdev_lock
);
877 spin_unlock(&bdev_lock
);
879 bdev
= bdget(inode
->i_rdev
);
881 spin_lock(&bdev_lock
);
882 if (!inode
->i_bdev
) {
884 * We take an additional reference to bd_inode,
885 * and it's released in clear_inode() of inode.
886 * So, we can access it via ->i_mapping always
890 inode
->i_bdev
= bdev
;
891 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
893 spin_unlock(&bdev_lock
);
898 /* Call when you free inode */
900 void bd_forget(struct inode
*inode
)
902 struct block_device
*bdev
= NULL
;
904 spin_lock(&bdev_lock
);
905 if (!sb_is_blkdev_sb(inode
->i_sb
))
906 bdev
= inode
->i_bdev
;
907 inode
->i_bdev
= NULL
;
908 inode
->i_mapping
= &inode
->i_data
;
909 spin_unlock(&bdev_lock
);
916 * bd_may_claim - test whether a block device can be claimed
917 * @bdev: block device of interest
918 * @whole: whole block device containing @bdev, may equal @bdev
919 * @holder: holder trying to claim @bdev
921 * Test whether @bdev can be claimed by @holder.
924 * spin_lock(&bdev_lock).
927 * %true if @bdev can be claimed, %false otherwise.
929 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
932 if (bdev
->bd_holder
== holder
)
933 return true; /* already a holder */
934 else if (bdev
->bd_holder
!= NULL
)
935 return false; /* held by someone else */
936 else if (bdev
->bd_contains
== bdev
)
937 return true; /* is a whole device which isn't held */
939 else if (whole
->bd_holder
== bd_may_claim
)
940 return true; /* is a partition of a device that is being partitioned */
941 else if (whole
->bd_holder
!= NULL
)
942 return false; /* is a partition of a held device */
944 return true; /* is a partition of an un-held device */
948 * bd_prepare_to_claim - prepare to claim a block device
949 * @bdev: block device of interest
950 * @whole: the whole device containing @bdev, may equal @bdev
951 * @holder: holder trying to claim @bdev
953 * Prepare to claim @bdev. This function fails if @bdev is already
954 * claimed by another holder and waits if another claiming is in
955 * progress. This function doesn't actually claim. On successful
956 * return, the caller has ownership of bd_claiming and bd_holder[s].
959 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
963 * 0 if @bdev can be claimed, -EBUSY otherwise.
965 static int bd_prepare_to_claim(struct block_device
*bdev
,
966 struct block_device
*whole
, void *holder
)
969 /* if someone else claimed, fail */
970 if (!bd_may_claim(bdev
, whole
, holder
))
973 /* if claiming is already in progress, wait for it to finish */
974 if (whole
->bd_claiming
) {
975 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
978 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
979 spin_unlock(&bdev_lock
);
981 finish_wait(wq
, &wait
);
982 spin_lock(&bdev_lock
);
991 * bd_start_claiming - start claiming a block device
992 * @bdev: block device of interest
993 * @holder: holder trying to claim @bdev
995 * @bdev is about to be opened exclusively. Check @bdev can be opened
996 * exclusively and mark that an exclusive open is in progress. Each
997 * successful call to this function must be matched with a call to
998 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1001 * This function is used to gain exclusive access to the block device
1002 * without actually causing other exclusive open attempts to fail. It
1003 * should be used when the open sequence itself requires exclusive
1004 * access but may subsequently fail.
1010 * Pointer to the block device containing @bdev on success, ERR_PTR()
1013 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1016 struct gendisk
*disk
;
1017 struct block_device
*whole
;
1023 * @bdev might not have been initialized properly yet, look up
1024 * and grab the outer block device the hard way.
1026 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1028 return ERR_PTR(-ENXIO
);
1031 * Normally, @bdev should equal what's returned from bdget_disk()
1032 * if partno is 0; however, some drivers (floppy) use multiple
1033 * bdev's for the same physical device and @bdev may be one of the
1034 * aliases. Keep @bdev if partno is 0. This means claimer
1035 * tracking is broken for those devices but it has always been that
1039 whole
= bdget_disk(disk
, 0);
1041 whole
= bdgrab(bdev
);
1043 module_put(disk
->fops
->owner
);
1046 return ERR_PTR(-ENOMEM
);
1048 /* prepare to claim, if successful, mark claiming in progress */
1049 spin_lock(&bdev_lock
);
1051 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1053 whole
->bd_claiming
= holder
;
1054 spin_unlock(&bdev_lock
);
1057 spin_unlock(&bdev_lock
);
1059 return ERR_PTR(err
);
1064 struct bd_holder_disk
{
1065 struct list_head list
;
1066 struct gendisk
*disk
;
1070 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1071 struct gendisk
*disk
)
1073 struct bd_holder_disk
*holder
;
1075 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1076 if (holder
->disk
== disk
)
1081 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1083 return sysfs_create_link(from
, to
, kobject_name(to
));
1086 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1088 sysfs_remove_link(from
, kobject_name(to
));
1092 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1093 * @bdev: the claimed slave bdev
1094 * @disk: the holding disk
1096 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1098 * This functions creates the following sysfs symlinks.
1100 * - from "slaves" directory of the holder @disk to the claimed @bdev
1101 * - from "holders" directory of the @bdev to the holder @disk
1103 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1104 * passed to bd_link_disk_holder(), then:
1106 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1107 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1109 * The caller must have claimed @bdev before calling this function and
1110 * ensure that both @bdev and @disk are valid during the creation and
1111 * lifetime of these symlinks.
1117 * 0 on success, -errno on failure.
1119 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1121 struct bd_holder_disk
*holder
;
1124 mutex_lock(&bdev
->bd_mutex
);
1126 WARN_ON_ONCE(!bdev
->bd_holder
);
1128 /* FIXME: remove the following once add_disk() handles errors */
1129 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1132 holder
= bd_find_holder_disk(bdev
, disk
);
1138 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1144 INIT_LIST_HEAD(&holder
->list
);
1145 holder
->disk
= disk
;
1148 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1152 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1156 * bdev could be deleted beneath us which would implicitly destroy
1157 * the holder directory. Hold on to it.
1159 kobject_get(bdev
->bd_part
->holder_dir
);
1161 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1165 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1169 mutex_unlock(&bdev
->bd_mutex
);
1172 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1175 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1176 * @bdev: the calimed slave bdev
1177 * @disk: the holding disk
1179 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1184 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1186 struct bd_holder_disk
*holder
;
1188 mutex_lock(&bdev
->bd_mutex
);
1190 holder
= bd_find_holder_disk(bdev
, disk
);
1192 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1193 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1194 del_symlink(bdev
->bd_part
->holder_dir
,
1195 &disk_to_dev(disk
)->kobj
);
1196 kobject_put(bdev
->bd_part
->holder_dir
);
1197 list_del_init(&holder
->list
);
1201 mutex_unlock(&bdev
->bd_mutex
);
1203 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1207 * flush_disk - invalidates all buffer-cache entries on a disk
1209 * @bdev: struct block device to be flushed
1210 * @kill_dirty: flag to guide handling of dirty inodes
1212 * Invalidates all buffer-cache entries on a disk. It should be called
1213 * when a disk has been changed -- either by a media change or online
1216 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1218 if (__invalidate_device(bdev
, kill_dirty
)) {
1219 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1220 "resized disk %s\n",
1221 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1226 if (disk_part_scan_enabled(bdev
->bd_disk
))
1227 bdev
->bd_invalidated
= 1;
1231 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1232 * @disk: struct gendisk to check
1233 * @bdev: struct bdev to adjust.
1235 * This routine checks to see if the bdev size does not match the disk size
1236 * and adjusts it if it differs.
1238 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1240 loff_t disk_size
, bdev_size
;
1242 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1243 bdev_size
= i_size_read(bdev
->bd_inode
);
1244 if (disk_size
!= bdev_size
) {
1246 "%s: detected capacity change from %lld to %lld\n",
1247 disk
->disk_name
, bdev_size
, disk_size
);
1248 i_size_write(bdev
->bd_inode
, disk_size
);
1249 flush_disk(bdev
, false);
1252 EXPORT_SYMBOL(check_disk_size_change
);
1255 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1256 * @disk: struct gendisk to be revalidated
1258 * This routine is a wrapper for lower-level driver's revalidate_disk
1259 * call-backs. It is used to do common pre and post operations needed
1260 * for all revalidate_disk operations.
1262 int revalidate_disk(struct gendisk
*disk
)
1264 struct block_device
*bdev
;
1267 if (disk
->fops
->revalidate_disk
)
1268 ret
= disk
->fops
->revalidate_disk(disk
);
1269 blk_integrity_revalidate(disk
);
1270 bdev
= bdget_disk(disk
, 0);
1274 mutex_lock(&bdev
->bd_mutex
);
1275 check_disk_size_change(disk
, bdev
);
1276 bdev
->bd_invalidated
= 0;
1277 mutex_unlock(&bdev
->bd_mutex
);
1281 EXPORT_SYMBOL(revalidate_disk
);
1284 * This routine checks whether a removable media has been changed,
1285 * and invalidates all buffer-cache-entries in that case. This
1286 * is a relatively slow routine, so we have to try to minimize using
1287 * it. Thus it is called only upon a 'mount' or 'open'. This
1288 * is the best way of combining speed and utility, I think.
1289 * People changing diskettes in the middle of an operation deserve
1292 int check_disk_change(struct block_device
*bdev
)
1294 struct gendisk
*disk
= bdev
->bd_disk
;
1295 const struct block_device_operations
*bdops
= disk
->fops
;
1296 unsigned int events
;
1298 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1299 DISK_EVENT_EJECT_REQUEST
);
1300 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1303 flush_disk(bdev
, true);
1304 if (bdops
->revalidate_disk
)
1305 bdops
->revalidate_disk(bdev
->bd_disk
);
1309 EXPORT_SYMBOL(check_disk_change
);
1311 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1313 unsigned bsize
= bdev_logical_block_size(bdev
);
1315 inode_lock(bdev
->bd_inode
);
1316 i_size_write(bdev
->bd_inode
, size
);
1317 inode_unlock(bdev
->bd_inode
);
1318 while (bsize
< PAGE_SIZE
) {
1323 bdev
->bd_block_size
= bsize
;
1324 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1326 EXPORT_SYMBOL(bd_set_size
);
1328 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1333 * mutex_lock(part->bd_mutex)
1334 * mutex_lock_nested(whole->bd_mutex, 1)
1337 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1339 struct gendisk
*disk
;
1340 struct module
*owner
;
1345 if (mode
& FMODE_READ
)
1347 if (mode
& FMODE_WRITE
)
1350 * hooks: /n/, see "layering violations".
1353 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1363 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1366 owner
= disk
->fops
->owner
;
1368 disk_block_events(disk
);
1369 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1370 if (!bdev
->bd_openers
) {
1371 bdev
->bd_disk
= disk
;
1372 bdev
->bd_queue
= disk
->queue
;
1373 bdev
->bd_contains
= bdev
;
1377 bdev
->bd_part
= disk_get_part(disk
, partno
);
1382 if (disk
->fops
->open
) {
1383 ret
= disk
->fops
->open(bdev
, mode
);
1384 if (ret
== -ERESTARTSYS
) {
1385 /* Lost a race with 'disk' being
1386 * deleted, try again.
1389 disk_put_part(bdev
->bd_part
);
1390 bdev
->bd_part
= NULL
;
1391 bdev
->bd_disk
= NULL
;
1392 bdev
->bd_queue
= NULL
;
1393 mutex_unlock(&bdev
->bd_mutex
);
1394 disk_unblock_events(disk
);
1402 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1405 * If the device is invalidated, rescan partition
1406 * if open succeeded or failed with -ENOMEDIUM.
1407 * The latter is necessary to prevent ghost
1408 * partitions on a removed medium.
1410 if (bdev
->bd_invalidated
) {
1412 rescan_partitions(disk
, bdev
);
1413 else if (ret
== -ENOMEDIUM
)
1414 invalidate_partitions(disk
, bdev
);
1420 struct block_device
*whole
;
1421 whole
= bdget_disk(disk
, 0);
1426 ret
= __blkdev_get(whole
, mode
, 1);
1429 bdev
->bd_contains
= whole
;
1430 bdev
->bd_part
= disk_get_part(disk
, partno
);
1431 if (!(disk
->flags
& GENHD_FL_UP
) ||
1432 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1436 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1439 if (bdev
->bd_contains
== bdev
) {
1441 if (bdev
->bd_disk
->fops
->open
)
1442 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1443 /* the same as first opener case, read comment there */
1444 if (bdev
->bd_invalidated
) {
1446 rescan_partitions(bdev
->bd_disk
, bdev
);
1447 else if (ret
== -ENOMEDIUM
)
1448 invalidate_partitions(bdev
->bd_disk
, bdev
);
1451 goto out_unlock_bdev
;
1453 /* only one opener holds refs to the module and disk */
1459 bdev
->bd_part_count
++;
1460 mutex_unlock(&bdev
->bd_mutex
);
1461 disk_unblock_events(disk
);
1465 disk_put_part(bdev
->bd_part
);
1466 bdev
->bd_disk
= NULL
;
1467 bdev
->bd_part
= NULL
;
1468 bdev
->bd_queue
= NULL
;
1469 if (bdev
!= bdev
->bd_contains
)
1470 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1471 bdev
->bd_contains
= NULL
;
1473 mutex_unlock(&bdev
->bd_mutex
);
1474 disk_unblock_events(disk
);
1484 * blkdev_get - open a block device
1485 * @bdev: block_device to open
1486 * @mode: FMODE_* mask
1487 * @holder: exclusive holder identifier
1489 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1490 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1491 * @holder is invalid. Exclusive opens may nest for the same @holder.
1493 * On success, the reference count of @bdev is unchanged. On failure,
1500 * 0 on success, -errno on failure.
1502 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1504 struct block_device
*whole
= NULL
;
1507 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1509 if ((mode
& FMODE_EXCL
) && holder
) {
1510 whole
= bd_start_claiming(bdev
, holder
);
1511 if (IS_ERR(whole
)) {
1513 return PTR_ERR(whole
);
1517 res
= __blkdev_get(bdev
, mode
, 0);
1520 struct gendisk
*disk
= whole
->bd_disk
;
1522 /* finish claiming */
1523 mutex_lock(&bdev
->bd_mutex
);
1524 spin_lock(&bdev_lock
);
1527 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1529 * Note that for a whole device bd_holders
1530 * will be incremented twice, and bd_holder
1531 * will be set to bd_may_claim before being
1534 whole
->bd_holders
++;
1535 whole
->bd_holder
= bd_may_claim
;
1537 bdev
->bd_holder
= holder
;
1540 /* tell others that we're done */
1541 BUG_ON(whole
->bd_claiming
!= holder
);
1542 whole
->bd_claiming
= NULL
;
1543 wake_up_bit(&whole
->bd_claiming
, 0);
1545 spin_unlock(&bdev_lock
);
1548 * Block event polling for write claims if requested. Any
1549 * write holder makes the write_holder state stick until
1550 * all are released. This is good enough and tracking
1551 * individual writeable reference is too fragile given the
1552 * way @mode is used in blkdev_get/put().
1554 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1555 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1556 bdev
->bd_write_holder
= true;
1557 disk_block_events(disk
);
1560 mutex_unlock(&bdev
->bd_mutex
);
1566 EXPORT_SYMBOL(blkdev_get
);
1569 * blkdev_get_by_path - open a block device by name
1570 * @path: path to the block device to open
1571 * @mode: FMODE_* mask
1572 * @holder: exclusive holder identifier
1574 * Open the blockdevice described by the device file at @path. @mode
1575 * and @holder are identical to blkdev_get().
1577 * On success, the returned block_device has reference count of one.
1583 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1585 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1588 struct block_device
*bdev
;
1591 bdev
= lookup_bdev(path
);
1595 err
= blkdev_get(bdev
, mode
, holder
);
1597 return ERR_PTR(err
);
1599 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1600 blkdev_put(bdev
, mode
);
1601 return ERR_PTR(-EACCES
);
1606 EXPORT_SYMBOL(blkdev_get_by_path
);
1609 * blkdev_get_by_dev - open a block device by device number
1610 * @dev: device number of block device to open
1611 * @mode: FMODE_* mask
1612 * @holder: exclusive holder identifier
1614 * Open the blockdevice described by device number @dev. @mode and
1615 * @holder are identical to blkdev_get().
1617 * Use it ONLY if you really do not have anything better - i.e. when
1618 * you are behind a truly sucky interface and all you are given is a
1619 * device number. _Never_ to be used for internal purposes. If you
1620 * ever need it - reconsider your API.
1622 * On success, the returned block_device has reference count of one.
1628 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1630 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1632 struct block_device
*bdev
;
1637 return ERR_PTR(-ENOMEM
);
1639 err
= blkdev_get(bdev
, mode
, holder
);
1641 return ERR_PTR(err
);
1645 EXPORT_SYMBOL(blkdev_get_by_dev
);
1647 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1649 struct block_device
*bdev
;
1652 * Preserve backwards compatibility and allow large file access
1653 * even if userspace doesn't ask for it explicitly. Some mkfs
1654 * binary needs it. We might want to drop this workaround
1655 * during an unstable branch.
1657 filp
->f_flags
|= O_LARGEFILE
;
1659 if (filp
->f_flags
& O_NDELAY
)
1660 filp
->f_mode
|= FMODE_NDELAY
;
1661 if (filp
->f_flags
& O_EXCL
)
1662 filp
->f_mode
|= FMODE_EXCL
;
1663 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1664 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1666 bdev
= bd_acquire(inode
);
1670 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1672 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1675 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1677 struct gendisk
*disk
= bdev
->bd_disk
;
1678 struct block_device
*victim
= NULL
;
1680 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1682 bdev
->bd_part_count
--;
1684 if (!--bdev
->bd_openers
) {
1685 WARN_ON_ONCE(bdev
->bd_holders
);
1686 sync_blockdev(bdev
);
1689 bdev_write_inode(bdev
);
1691 * Detaching bdev inode from its wb in __destroy_inode()
1692 * is too late: the queue which embeds its bdi (along with
1693 * root wb) can be gone as soon as we put_disk() below.
1695 inode_detach_wb(bdev
->bd_inode
);
1697 if (bdev
->bd_contains
== bdev
) {
1698 if (disk
->fops
->release
)
1699 disk
->fops
->release(disk
, mode
);
1701 if (!bdev
->bd_openers
) {
1702 struct module
*owner
= disk
->fops
->owner
;
1704 disk_put_part(bdev
->bd_part
);
1705 bdev
->bd_part
= NULL
;
1706 bdev
->bd_disk
= NULL
;
1707 if (bdev
!= bdev
->bd_contains
)
1708 victim
= bdev
->bd_contains
;
1709 bdev
->bd_contains
= NULL
;
1714 mutex_unlock(&bdev
->bd_mutex
);
1717 __blkdev_put(victim
, mode
, 1);
1720 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1722 mutex_lock(&bdev
->bd_mutex
);
1724 if (mode
& FMODE_EXCL
) {
1728 * Release a claim on the device. The holder fields
1729 * are protected with bdev_lock. bd_mutex is to
1730 * synchronize disk_holder unlinking.
1732 spin_lock(&bdev_lock
);
1734 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1735 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1737 /* bd_contains might point to self, check in a separate step */
1738 if ((bdev_free
= !bdev
->bd_holders
))
1739 bdev
->bd_holder
= NULL
;
1740 if (!bdev
->bd_contains
->bd_holders
)
1741 bdev
->bd_contains
->bd_holder
= NULL
;
1743 spin_unlock(&bdev_lock
);
1746 * If this was the last claim, remove holder link and
1747 * unblock evpoll if it was a write holder.
1749 if (bdev_free
&& bdev
->bd_write_holder
) {
1750 disk_unblock_events(bdev
->bd_disk
);
1751 bdev
->bd_write_holder
= false;
1756 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1757 * event. This is to ensure detection of media removal commanded
1758 * from userland - e.g. eject(1).
1760 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1762 mutex_unlock(&bdev
->bd_mutex
);
1764 __blkdev_put(bdev
, mode
, 0);
1766 EXPORT_SYMBOL(blkdev_put
);
1768 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1770 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1771 blkdev_put(bdev
, filp
->f_mode
);
1775 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1777 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1778 fmode_t mode
= file
->f_mode
;
1781 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1782 * to updated it before every ioctl.
1784 if (file
->f_flags
& O_NDELAY
)
1785 mode
|= FMODE_NDELAY
;
1787 mode
&= ~FMODE_NDELAY
;
1789 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1793 * Write data to the block device. Only intended for the block device itself
1794 * and the raw driver which basically is a fake block device.
1796 * Does not take i_mutex for the write and thus is not for general purpose
1799 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1801 struct file
*file
= iocb
->ki_filp
;
1802 struct inode
*bd_inode
= bdev_file_inode(file
);
1803 loff_t size
= i_size_read(bd_inode
);
1804 struct blk_plug plug
;
1807 if (bdev_read_only(I_BDEV(bd_inode
)))
1810 if (!iov_iter_count(from
))
1813 if (iocb
->ki_pos
>= size
)
1816 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1818 blk_start_plug(&plug
);
1819 ret
= __generic_file_write_iter(iocb
, from
);
1821 ret
= generic_write_sync(iocb
, ret
);
1822 blk_finish_plug(&plug
);
1825 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1827 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1829 struct file
*file
= iocb
->ki_filp
;
1830 struct inode
*bd_inode
= bdev_file_inode(file
);
1831 loff_t size
= i_size_read(bd_inode
);
1832 loff_t pos
= iocb
->ki_pos
;
1838 iov_iter_truncate(to
, size
);
1839 return generic_file_read_iter(iocb
, to
);
1841 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1844 * Try to release a page associated with block device when the system
1845 * is under memory pressure.
1847 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1849 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1851 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1852 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1854 return try_to_free_buffers(page
);
1857 static int blkdev_writepages(struct address_space
*mapping
,
1858 struct writeback_control
*wbc
)
1860 if (dax_mapping(mapping
)) {
1861 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1863 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1865 return generic_writepages(mapping
, wbc
);
1868 static const struct address_space_operations def_blk_aops
= {
1869 .readpage
= blkdev_readpage
,
1870 .readpages
= blkdev_readpages
,
1871 .writepage
= blkdev_writepage
,
1872 .write_begin
= blkdev_write_begin
,
1873 .write_end
= blkdev_write_end
,
1874 .writepages
= blkdev_writepages
,
1875 .releasepage
= blkdev_releasepage
,
1876 .direct_IO
= blkdev_direct_IO
,
1877 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1880 #define BLKDEV_FALLOC_FL_SUPPORTED \
1881 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1882 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1884 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
1887 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1888 struct request_queue
*q
= bdev_get_queue(bdev
);
1889 struct address_space
*mapping
;
1890 loff_t end
= start
+ len
- 1;
1894 /* Fail if we don't recognize the flags. */
1895 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
1898 /* Don't go off the end of the device. */
1899 isize
= i_size_read(bdev
->bd_inode
);
1903 if (mode
& FALLOC_FL_KEEP_SIZE
) {
1904 len
= isize
- start
;
1905 end
= start
+ len
- 1;
1911 * Don't allow IO that isn't aligned to logical block size.
1913 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
1916 /* Invalidate the page cache, including dirty pages. */
1917 mapping
= bdev
->bd_inode
->i_mapping
;
1918 truncate_inode_pages_range(mapping
, start
, end
);
1921 case FALLOC_FL_ZERO_RANGE
:
1922 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
1923 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
1926 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
1927 /* Only punch if the device can do zeroing discard. */
1928 if (!blk_queue_discard(q
) || !q
->limits
.discard_zeroes_data
)
1930 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
1933 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
1934 if (!blk_queue_discard(q
))
1936 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
1946 * Invalidate again; if someone wandered in and dirtied a page,
1947 * the caller will be given -EBUSY. The third argument is
1948 * inclusive, so the rounding here is safe.
1950 return invalidate_inode_pages2_range(mapping
,
1951 start
>> PAGE_SHIFT
,
1955 const struct file_operations def_blk_fops
= {
1956 .open
= blkdev_open
,
1957 .release
= blkdev_close
,
1958 .llseek
= block_llseek
,
1959 .read_iter
= blkdev_read_iter
,
1960 .write_iter
= blkdev_write_iter
,
1961 .mmap
= generic_file_mmap
,
1962 .fsync
= blkdev_fsync
,
1963 .unlocked_ioctl
= block_ioctl
,
1964 #ifdef CONFIG_COMPAT
1965 .compat_ioctl
= compat_blkdev_ioctl
,
1967 .splice_read
= generic_file_splice_read
,
1968 .splice_write
= iter_file_splice_write
,
1969 .fallocate
= blkdev_fallocate
,
1972 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1975 mm_segment_t old_fs
= get_fs();
1977 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1982 EXPORT_SYMBOL(ioctl_by_bdev
);
1985 * lookup_bdev - lookup a struct block_device by name
1986 * @pathname: special file representing the block device
1988 * Get a reference to the blockdevice at @pathname in the current
1989 * namespace if possible and return it. Return ERR_PTR(error)
1992 struct block_device
*lookup_bdev(const char *pathname
)
1994 struct block_device
*bdev
;
1995 struct inode
*inode
;
1999 if (!pathname
|| !*pathname
)
2000 return ERR_PTR(-EINVAL
);
2002 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2004 return ERR_PTR(error
);
2006 inode
= d_backing_inode(path
.dentry
);
2008 if (!S_ISBLK(inode
->i_mode
))
2011 if (!may_open_dev(&path
))
2014 bdev
= bd_acquire(inode
);
2021 bdev
= ERR_PTR(error
);
2024 EXPORT_SYMBOL(lookup_bdev
);
2026 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2028 struct super_block
*sb
= get_super(bdev
);
2033 * no need to lock the super, get_super holds the
2034 * read mutex so the filesystem cannot go away
2035 * under us (->put_super runs with the write lock
2038 shrink_dcache_sb(sb
);
2039 res
= invalidate_inodes(sb
, kill_dirty
);
2042 invalidate_bdev(bdev
);
2045 EXPORT_SYMBOL(__invalidate_device
);
2047 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2049 struct inode
*inode
, *old_inode
= NULL
;
2051 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2052 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2053 struct address_space
*mapping
= inode
->i_mapping
;
2055 spin_lock(&inode
->i_lock
);
2056 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2057 mapping
->nrpages
== 0) {
2058 spin_unlock(&inode
->i_lock
);
2062 spin_unlock(&inode
->i_lock
);
2063 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2065 * We hold a reference to 'inode' so it couldn't have been
2066 * removed from s_inodes list while we dropped the
2067 * s_inode_list_lock We cannot iput the inode now as we can
2068 * be holding the last reference and we cannot iput it under
2069 * s_inode_list_lock. So we keep the reference and iput it
2075 func(I_BDEV(inode
), arg
);
2077 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2079 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);