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 <asm/uaccess.h>
37 struct block_device bdev
;
38 struct inode vfs_inode
;
41 static const struct address_space_operations def_blk_aops
;
43 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
45 return container_of(inode
, struct bdev_inode
, vfs_inode
);
48 struct block_device
*I_BDEV(struct inode
*inode
)
50 return &BDEV_I(inode
)->bdev
;
52 EXPORT_SYMBOL(I_BDEV
);
54 void __vfs_msg(struct super_block
*sb
, const char *prefix
, const char *fmt
, ...)
62 printk_ratelimited("%sVFS (%s): %pV\n", prefix
, sb
->s_id
, &vaf
);
66 static void bdev_write_inode(struct block_device
*bdev
)
68 struct inode
*inode
= bdev
->bd_inode
;
71 spin_lock(&inode
->i_lock
);
72 while (inode
->i_state
& I_DIRTY
) {
73 spin_unlock(&inode
->i_lock
);
74 ret
= write_inode_now(inode
, true);
76 char name
[BDEVNAME_SIZE
];
77 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
78 "for block device %s (err=%d).\n",
79 bdevname(bdev
, name
), ret
);
81 spin_lock(&inode
->i_lock
);
83 spin_unlock(&inode
->i_lock
);
86 /* Kill _all_ buffers and pagecache , dirty or not.. */
87 void kill_bdev(struct block_device
*bdev
)
89 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
91 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
95 truncate_inode_pages(mapping
, 0);
97 EXPORT_SYMBOL(kill_bdev
);
99 /* Invalidate clean unused buffers and pagecache. */
100 void invalidate_bdev(struct block_device
*bdev
)
102 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
104 if (mapping
->nrpages
== 0)
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping
, 0, -1);
110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
113 cleancache_invalidate_inode(mapping
);
115 EXPORT_SYMBOL(invalidate_bdev
);
117 int set_blocksize(struct block_device
*bdev
, int size
)
119 /* Size must be a power of two, and between 512 and PAGE_SIZE */
120 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
123 /* Size cannot be smaller than the size supported by the device */
124 if (size
< bdev_logical_block_size(bdev
))
127 /* Don't change the size if it is same as current */
128 if (bdev
->bd_block_size
!= size
) {
130 bdev
->bd_block_size
= size
;
131 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
137 EXPORT_SYMBOL(set_blocksize
);
139 int sb_set_blocksize(struct super_block
*sb
, int size
)
141 if (set_blocksize(sb
->s_bdev
, size
))
143 /* If we get here, we know size is power of two
144 * and it's value is between 512 and PAGE_SIZE */
145 sb
->s_blocksize
= size
;
146 sb
->s_blocksize_bits
= blksize_bits(size
);
147 return sb
->s_blocksize
;
150 EXPORT_SYMBOL(sb_set_blocksize
);
152 int sb_min_blocksize(struct super_block
*sb
, int size
)
154 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
157 return sb_set_blocksize(sb
, size
);
160 EXPORT_SYMBOL(sb_min_blocksize
);
163 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
164 struct buffer_head
*bh
, int create
)
166 bh
->b_bdev
= I_BDEV(inode
);
167 bh
->b_blocknr
= iblock
;
168 set_buffer_mapped(bh
);
172 static struct inode
*bdev_file_inode(struct file
*file
)
174 return file
->f_mapping
->host
;
178 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
180 struct file
*file
= iocb
->ki_filp
;
181 struct inode
*inode
= bdev_file_inode(file
);
184 return dax_do_io(iocb
, inode
, iter
, blkdev_get_block
,
185 NULL
, DIO_SKIP_DIO_COUNT
);
186 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
,
187 blkdev_get_block
, NULL
, NULL
,
191 int __sync_blockdev(struct block_device
*bdev
, int wait
)
196 return filemap_flush(bdev
->bd_inode
->i_mapping
);
197 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
201 * Write out and wait upon all the dirty data associated with a block
202 * device via its mapping. Does not take the superblock lock.
204 int sync_blockdev(struct block_device
*bdev
)
206 return __sync_blockdev(bdev
, 1);
208 EXPORT_SYMBOL(sync_blockdev
);
211 * Write out and wait upon all dirty data associated with this
212 * device. Filesystem data as well as the underlying block
213 * device. Takes the superblock lock.
215 int fsync_bdev(struct block_device
*bdev
)
217 struct super_block
*sb
= get_super(bdev
);
219 int res
= sync_filesystem(sb
);
223 return sync_blockdev(bdev
);
225 EXPORT_SYMBOL(fsync_bdev
);
228 * freeze_bdev -- lock a filesystem and force it into a consistent state
229 * @bdev: blockdevice to lock
231 * If a superblock is found on this device, we take the s_umount semaphore
232 * on it to make sure nobody unmounts until the snapshot creation is done.
233 * The reference counter (bd_fsfreeze_count) guarantees that only the last
234 * unfreeze process can unfreeze the frozen filesystem actually when multiple
235 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
236 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
239 struct super_block
*freeze_bdev(struct block_device
*bdev
)
241 struct super_block
*sb
;
244 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
245 if (++bdev
->bd_fsfreeze_count
> 1) {
247 * We don't even need to grab a reference - the first call
248 * to freeze_bdev grab an active reference and only the last
249 * thaw_bdev drops it.
251 sb
= get_super(bdev
);
253 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
257 sb
= get_active_super(bdev
);
260 if (sb
->s_op
->freeze_super
)
261 error
= sb
->s_op
->freeze_super(sb
);
263 error
= freeze_super(sb
);
265 deactivate_super(sb
);
266 bdev
->bd_fsfreeze_count
--;
267 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
268 return ERR_PTR(error
);
270 deactivate_super(sb
);
273 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
274 return sb
; /* thaw_bdev releases s->s_umount */
276 EXPORT_SYMBOL(freeze_bdev
);
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
285 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
289 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
290 if (!bdev
->bd_fsfreeze_count
)
294 if (--bdev
->bd_fsfreeze_count
> 0)
300 if (sb
->s_op
->thaw_super
)
301 error
= sb
->s_op
->thaw_super(sb
);
303 error
= thaw_super(sb
);
305 bdev
->bd_fsfreeze_count
++;
306 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
310 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
313 EXPORT_SYMBOL(thaw_bdev
);
315 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
317 return block_write_full_page(page
, blkdev_get_block
, wbc
);
320 static int blkdev_readpage(struct file
* file
, struct page
* page
)
322 return block_read_full_page(page
, blkdev_get_block
);
325 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
326 struct list_head
*pages
, unsigned nr_pages
)
328 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
331 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
332 loff_t pos
, unsigned len
, unsigned flags
,
333 struct page
**pagep
, void **fsdata
)
335 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
339 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
340 loff_t pos
, unsigned len
, unsigned copied
,
341 struct page
*page
, void *fsdata
)
344 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
354 * for a block special file file_inode(file)->i_size is zero
355 * so we compute the size by hand (just as in block_read/write above)
357 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
359 struct inode
*bd_inode
= bdev_file_inode(file
);
362 inode_lock(bd_inode
);
363 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
364 inode_unlock(bd_inode
);
368 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
370 struct inode
*bd_inode
= bdev_file_inode(filp
);
371 struct block_device
*bdev
= I_BDEV(bd_inode
);
374 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
379 * There is no need to serialise calls to blkdev_issue_flush with
380 * i_mutex and doing so causes performance issues with concurrent
381 * O_SYNC writers to a block device.
383 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
384 if (error
== -EOPNOTSUPP
)
389 EXPORT_SYMBOL(blkdev_fsync
);
392 * bdev_read_page() - Start reading a page from a block device
393 * @bdev: The device to read the page from
394 * @sector: The offset on the device to read the page to (need not be aligned)
395 * @page: The page to read
397 * On entry, the page should be locked. It will be unlocked when the page
398 * has been read. If the block driver implements rw_page synchronously,
399 * that will be true on exit from this function, but it need not be.
401 * Errors returned by this function are usually "soft", eg out of memory, or
402 * queue full; callers should try a different route to read this page rather
403 * than propagate an error back up the stack.
405 * Return: negative errno if an error occurs, 0 if submission was successful.
407 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
410 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
411 int result
= -EOPNOTSUPP
;
413 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
416 result
= blk_queue_enter(bdev
->bd_queue
, false);
419 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
421 blk_queue_exit(bdev
->bd_queue
);
424 EXPORT_SYMBOL_GPL(bdev_read_page
);
427 * bdev_write_page() - Start writing a page to a block device
428 * @bdev: The device to write the page to
429 * @sector: The offset on the device to write the page to (need not be aligned)
430 * @page: The page to write
431 * @wbc: The writeback_control for the write
433 * On entry, the page should be locked and not currently under writeback.
434 * On exit, if the write started successfully, the page will be unlocked and
435 * under writeback. If the write failed already (eg the driver failed to
436 * queue the page to the device), the page will still be locked. If the
437 * caller is a ->writepage implementation, it will need to unlock the page.
439 * Errors returned by this function are usually "soft", eg out of memory, or
440 * queue full; callers should try a different route to write this page rather
441 * than propagate an error back up the stack.
443 * Return: negative errno if an error occurs, 0 if submission was successful.
445 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
446 struct page
*page
, struct writeback_control
*wbc
)
449 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
451 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
453 result
= blk_queue_enter(bdev
->bd_queue
, false);
457 set_page_writeback(page
);
458 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
,
461 end_page_writeback(page
);
464 blk_queue_exit(bdev
->bd_queue
);
467 EXPORT_SYMBOL_GPL(bdev_write_page
);
470 * bdev_direct_access() - Get the address for directly-accessibly memory
471 * @bdev: The device containing the memory
472 * @dax: control and output parameters for ->direct_access
474 * If a block device is made up of directly addressable memory, this function
475 * will tell the caller the PFN and the address of the memory. The address
476 * may be directly dereferenced within the kernel without the need to call
477 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
480 * Return: negative errno if an error occurs, otherwise the number of bytes
481 * accessible at this address.
483 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
485 sector_t sector
= dax
->sector
;
486 long avail
, size
= dax
->size
;
487 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
490 * The device driver is allowed to sleep, in order to make the
491 * memory directly accessible.
497 if (!blk_queue_dax(bdev_get_queue(bdev
)) || !ops
->direct_access
)
499 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
500 part_nr_sects_read(bdev
->bd_part
))
502 sector
+= get_start_sect(bdev
);
503 if (sector
% (PAGE_SIZE
/ 512))
505 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
, size
);
508 if (avail
> 0 && avail
& ~PAGE_MASK
)
510 return min(avail
, size
);
512 EXPORT_SYMBOL_GPL(bdev_direct_access
);
515 * bdev_dax_supported() - Check if the device supports dax for filesystem
516 * @sb: The superblock of the device
517 * @blocksize: The block size of the device
519 * This is a library function for filesystems to check if the block device
520 * can be mounted with dax option.
522 * Return: negative errno if unsupported, 0 if supported.
524 int bdev_dax_supported(struct super_block
*sb
, int blocksize
)
526 struct blk_dax_ctl dax
= {
532 if (blocksize
!= PAGE_SIZE
) {
533 vfs_msg(sb
, KERN_ERR
, "error: unsupported blocksize for dax");
537 err
= bdev_direct_access(sb
->s_bdev
, &dax
);
541 vfs_msg(sb
, KERN_ERR
,
542 "error: device does not support dax");
545 vfs_msg(sb
, KERN_ERR
,
546 "error: unaligned partition for dax");
549 vfs_msg(sb
, KERN_ERR
,
550 "error: dax access failed (%d)", err
);
557 EXPORT_SYMBOL_GPL(bdev_dax_supported
);
560 * bdev_dax_capable() - Return if the raw device is capable for dax
561 * @bdev: The device for raw block device access
563 bool bdev_dax_capable(struct block_device
*bdev
)
565 struct blk_dax_ctl dax
= {
569 if (!IS_ENABLED(CONFIG_FS_DAX
))
573 if (bdev_direct_access(bdev
, &dax
) < 0)
576 dax
.sector
= bdev
->bd_part
->nr_sects
- (PAGE_SIZE
/ 512);
577 if (bdev_direct_access(bdev
, &dax
) < 0)
587 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
588 static struct kmem_cache
* bdev_cachep __read_mostly
;
590 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
592 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
595 return &ei
->vfs_inode
;
598 static void bdev_i_callback(struct rcu_head
*head
)
600 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
601 struct bdev_inode
*bdi
= BDEV_I(inode
);
603 kmem_cache_free(bdev_cachep
, bdi
);
606 static void bdev_destroy_inode(struct inode
*inode
)
608 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
611 static void init_once(void *foo
)
613 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
614 struct block_device
*bdev
= &ei
->bdev
;
616 memset(bdev
, 0, sizeof(*bdev
));
617 mutex_init(&bdev
->bd_mutex
);
618 INIT_LIST_HEAD(&bdev
->bd_list
);
620 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
622 inode_init_once(&ei
->vfs_inode
);
623 /* Initialize mutex for freeze. */
624 mutex_init(&bdev
->bd_fsfreeze_mutex
);
627 static void bdev_evict_inode(struct inode
*inode
)
629 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
630 truncate_inode_pages_final(&inode
->i_data
);
631 invalidate_inode_buffers(inode
); /* is it needed here? */
633 spin_lock(&bdev_lock
);
634 list_del_init(&bdev
->bd_list
);
635 spin_unlock(&bdev_lock
);
638 static const struct super_operations bdev_sops
= {
639 .statfs
= simple_statfs
,
640 .alloc_inode
= bdev_alloc_inode
,
641 .destroy_inode
= bdev_destroy_inode
,
642 .drop_inode
= generic_delete_inode
,
643 .evict_inode
= bdev_evict_inode
,
646 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
647 int flags
, const char *dev_name
, void *data
)
650 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
652 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
656 static struct file_system_type bd_type
= {
659 .kill_sb
= kill_anon_super
,
662 struct super_block
*blockdev_superblock __read_mostly
;
663 EXPORT_SYMBOL_GPL(blockdev_superblock
);
665 void __init
bdev_cache_init(void)
668 static struct vfsmount
*bd_mnt
;
670 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
671 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
672 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
674 err
= register_filesystem(&bd_type
);
676 panic("Cannot register bdev pseudo-fs");
677 bd_mnt
= kern_mount(&bd_type
);
679 panic("Cannot create bdev pseudo-fs");
680 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
684 * Most likely _very_ bad one - but then it's hardly critical for small
685 * /dev and can be fixed when somebody will need really large one.
686 * Keep in mind that it will be fed through icache hash function too.
688 static inline unsigned long hash(dev_t dev
)
690 return MAJOR(dev
)+MINOR(dev
);
693 static int bdev_test(struct inode
*inode
, void *data
)
695 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
698 static int bdev_set(struct inode
*inode
, void *data
)
700 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
704 static LIST_HEAD(all_bdevs
);
706 struct block_device
*bdget(dev_t dev
)
708 struct block_device
*bdev
;
711 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
712 bdev_test
, bdev_set
, &dev
);
717 bdev
= &BDEV_I(inode
)->bdev
;
719 if (inode
->i_state
& I_NEW
) {
720 bdev
->bd_contains
= NULL
;
721 bdev
->bd_super
= NULL
;
722 bdev
->bd_inode
= inode
;
723 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
724 bdev
->bd_part_count
= 0;
725 bdev
->bd_invalidated
= 0;
726 inode
->i_mode
= S_IFBLK
;
728 inode
->i_bdev
= bdev
;
729 inode
->i_data
.a_ops
= &def_blk_aops
;
730 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
731 spin_lock(&bdev_lock
);
732 list_add(&bdev
->bd_list
, &all_bdevs
);
733 spin_unlock(&bdev_lock
);
734 unlock_new_inode(inode
);
739 EXPORT_SYMBOL(bdget
);
742 * bdgrab -- Grab a reference to an already referenced block device
743 * @bdev: Block device to grab a reference to.
745 struct block_device
*bdgrab(struct block_device
*bdev
)
747 ihold(bdev
->bd_inode
);
750 EXPORT_SYMBOL(bdgrab
);
752 long nr_blockdev_pages(void)
754 struct block_device
*bdev
;
756 spin_lock(&bdev_lock
);
757 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
758 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
760 spin_unlock(&bdev_lock
);
764 void bdput(struct block_device
*bdev
)
766 iput(bdev
->bd_inode
);
769 EXPORT_SYMBOL(bdput
);
771 static struct block_device
*bd_acquire(struct inode
*inode
)
773 struct block_device
*bdev
;
775 spin_lock(&bdev_lock
);
776 bdev
= inode
->i_bdev
;
779 spin_unlock(&bdev_lock
);
782 spin_unlock(&bdev_lock
);
784 bdev
= bdget(inode
->i_rdev
);
786 spin_lock(&bdev_lock
);
787 if (!inode
->i_bdev
) {
789 * We take an additional reference to bd_inode,
790 * and it's released in clear_inode() of inode.
791 * So, we can access it via ->i_mapping always
795 inode
->i_bdev
= bdev
;
796 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
798 spin_unlock(&bdev_lock
);
803 /* Call when you free inode */
805 void bd_forget(struct inode
*inode
)
807 struct block_device
*bdev
= NULL
;
809 spin_lock(&bdev_lock
);
810 if (!sb_is_blkdev_sb(inode
->i_sb
))
811 bdev
= inode
->i_bdev
;
812 inode
->i_bdev
= NULL
;
813 inode
->i_mapping
= &inode
->i_data
;
814 spin_unlock(&bdev_lock
);
821 * bd_may_claim - test whether a block device can be claimed
822 * @bdev: block device of interest
823 * @whole: whole block device containing @bdev, may equal @bdev
824 * @holder: holder trying to claim @bdev
826 * Test whether @bdev can be claimed by @holder.
829 * spin_lock(&bdev_lock).
832 * %true if @bdev can be claimed, %false otherwise.
834 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
837 if (bdev
->bd_holder
== holder
)
838 return true; /* already a holder */
839 else if (bdev
->bd_holder
!= NULL
)
840 return false; /* held by someone else */
841 else if (bdev
->bd_contains
== bdev
)
842 return true; /* is a whole device which isn't held */
844 else if (whole
->bd_holder
== bd_may_claim
)
845 return true; /* is a partition of a device that is being partitioned */
846 else if (whole
->bd_holder
!= NULL
)
847 return false; /* is a partition of a held device */
849 return true; /* is a partition of an un-held device */
853 * bd_prepare_to_claim - prepare to claim a block device
854 * @bdev: block device of interest
855 * @whole: the whole device containing @bdev, may equal @bdev
856 * @holder: holder trying to claim @bdev
858 * Prepare to claim @bdev. This function fails if @bdev is already
859 * claimed by another holder and waits if another claiming is in
860 * progress. This function doesn't actually claim. On successful
861 * return, the caller has ownership of bd_claiming and bd_holder[s].
864 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
868 * 0 if @bdev can be claimed, -EBUSY otherwise.
870 static int bd_prepare_to_claim(struct block_device
*bdev
,
871 struct block_device
*whole
, void *holder
)
874 /* if someone else claimed, fail */
875 if (!bd_may_claim(bdev
, whole
, holder
))
878 /* if claiming is already in progress, wait for it to finish */
879 if (whole
->bd_claiming
) {
880 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
883 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
884 spin_unlock(&bdev_lock
);
886 finish_wait(wq
, &wait
);
887 spin_lock(&bdev_lock
);
896 * bd_start_claiming - start claiming a block device
897 * @bdev: block device of interest
898 * @holder: holder trying to claim @bdev
900 * @bdev is about to be opened exclusively. Check @bdev can be opened
901 * exclusively and mark that an exclusive open is in progress. Each
902 * successful call to this function must be matched with a call to
903 * either bd_finish_claiming() or bd_abort_claiming() (which do not
906 * This function is used to gain exclusive access to the block device
907 * without actually causing other exclusive open attempts to fail. It
908 * should be used when the open sequence itself requires exclusive
909 * access but may subsequently fail.
915 * Pointer to the block device containing @bdev on success, ERR_PTR()
918 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
921 struct gendisk
*disk
;
922 struct block_device
*whole
;
928 * @bdev might not have been initialized properly yet, look up
929 * and grab the outer block device the hard way.
931 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
933 return ERR_PTR(-ENXIO
);
936 * Normally, @bdev should equal what's returned from bdget_disk()
937 * if partno is 0; however, some drivers (floppy) use multiple
938 * bdev's for the same physical device and @bdev may be one of the
939 * aliases. Keep @bdev if partno is 0. This means claimer
940 * tracking is broken for those devices but it has always been that
944 whole
= bdget_disk(disk
, 0);
946 whole
= bdgrab(bdev
);
948 module_put(disk
->fops
->owner
);
951 return ERR_PTR(-ENOMEM
);
953 /* prepare to claim, if successful, mark claiming in progress */
954 spin_lock(&bdev_lock
);
956 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
958 whole
->bd_claiming
= holder
;
959 spin_unlock(&bdev_lock
);
962 spin_unlock(&bdev_lock
);
969 struct bd_holder_disk
{
970 struct list_head list
;
971 struct gendisk
*disk
;
975 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
976 struct gendisk
*disk
)
978 struct bd_holder_disk
*holder
;
980 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
981 if (holder
->disk
== disk
)
986 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
988 return sysfs_create_link(from
, to
, kobject_name(to
));
991 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
993 sysfs_remove_link(from
, kobject_name(to
));
997 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
998 * @bdev: the claimed slave bdev
999 * @disk: the holding disk
1001 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1003 * This functions creates the following sysfs symlinks.
1005 * - from "slaves" directory of the holder @disk to the claimed @bdev
1006 * - from "holders" directory of the @bdev to the holder @disk
1008 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1009 * passed to bd_link_disk_holder(), then:
1011 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1012 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1014 * The caller must have claimed @bdev before calling this function and
1015 * ensure that both @bdev and @disk are valid during the creation and
1016 * lifetime of these symlinks.
1022 * 0 on success, -errno on failure.
1024 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1026 struct bd_holder_disk
*holder
;
1029 mutex_lock(&bdev
->bd_mutex
);
1031 WARN_ON_ONCE(!bdev
->bd_holder
);
1033 /* FIXME: remove the following once add_disk() handles errors */
1034 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1037 holder
= bd_find_holder_disk(bdev
, disk
);
1043 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1049 INIT_LIST_HEAD(&holder
->list
);
1050 holder
->disk
= disk
;
1053 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1057 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1061 * bdev could be deleted beneath us which would implicitly destroy
1062 * the holder directory. Hold on to it.
1064 kobject_get(bdev
->bd_part
->holder_dir
);
1066 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1070 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1074 mutex_unlock(&bdev
->bd_mutex
);
1077 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1080 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1081 * @bdev: the calimed slave bdev
1082 * @disk: the holding disk
1084 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1089 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1091 struct bd_holder_disk
*holder
;
1093 mutex_lock(&bdev
->bd_mutex
);
1095 holder
= bd_find_holder_disk(bdev
, disk
);
1097 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1098 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1099 del_symlink(bdev
->bd_part
->holder_dir
,
1100 &disk_to_dev(disk
)->kobj
);
1101 kobject_put(bdev
->bd_part
->holder_dir
);
1102 list_del_init(&holder
->list
);
1106 mutex_unlock(&bdev
->bd_mutex
);
1108 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1112 * flush_disk - invalidates all buffer-cache entries on a disk
1114 * @bdev: struct block device to be flushed
1115 * @kill_dirty: flag to guide handling of dirty inodes
1117 * Invalidates all buffer-cache entries on a disk. It should be called
1118 * when a disk has been changed -- either by a media change or online
1121 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1123 if (__invalidate_device(bdev
, kill_dirty
)) {
1124 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1125 "resized disk %s\n",
1126 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1131 if (disk_part_scan_enabled(bdev
->bd_disk
))
1132 bdev
->bd_invalidated
= 1;
1136 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1137 * @disk: struct gendisk to check
1138 * @bdev: struct bdev to adjust.
1140 * This routine checks to see if the bdev size does not match the disk size
1141 * and adjusts it if it differs.
1143 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1145 loff_t disk_size
, bdev_size
;
1147 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1148 bdev_size
= i_size_read(bdev
->bd_inode
);
1149 if (disk_size
!= bdev_size
) {
1151 "%s: detected capacity change from %lld to %lld\n",
1152 disk
->disk_name
, bdev_size
, disk_size
);
1153 i_size_write(bdev
->bd_inode
, disk_size
);
1154 flush_disk(bdev
, false);
1157 EXPORT_SYMBOL(check_disk_size_change
);
1160 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1161 * @disk: struct gendisk to be revalidated
1163 * This routine is a wrapper for lower-level driver's revalidate_disk
1164 * call-backs. It is used to do common pre and post operations needed
1165 * for all revalidate_disk operations.
1167 int revalidate_disk(struct gendisk
*disk
)
1169 struct block_device
*bdev
;
1172 if (disk
->fops
->revalidate_disk
)
1173 ret
= disk
->fops
->revalidate_disk(disk
);
1174 blk_integrity_revalidate(disk
);
1175 bdev
= bdget_disk(disk
, 0);
1179 mutex_lock(&bdev
->bd_mutex
);
1180 check_disk_size_change(disk
, bdev
);
1181 bdev
->bd_invalidated
= 0;
1182 mutex_unlock(&bdev
->bd_mutex
);
1186 EXPORT_SYMBOL(revalidate_disk
);
1189 * This routine checks whether a removable media has been changed,
1190 * and invalidates all buffer-cache-entries in that case. This
1191 * is a relatively slow routine, so we have to try to minimize using
1192 * it. Thus it is called only upon a 'mount' or 'open'. This
1193 * is the best way of combining speed and utility, I think.
1194 * People changing diskettes in the middle of an operation deserve
1197 int check_disk_change(struct block_device
*bdev
)
1199 struct gendisk
*disk
= bdev
->bd_disk
;
1200 const struct block_device_operations
*bdops
= disk
->fops
;
1201 unsigned int events
;
1203 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1204 DISK_EVENT_EJECT_REQUEST
);
1205 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1208 flush_disk(bdev
, true);
1209 if (bdops
->revalidate_disk
)
1210 bdops
->revalidate_disk(bdev
->bd_disk
);
1214 EXPORT_SYMBOL(check_disk_change
);
1216 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1218 unsigned bsize
= bdev_logical_block_size(bdev
);
1220 inode_lock(bdev
->bd_inode
);
1221 i_size_write(bdev
->bd_inode
, size
);
1222 inode_unlock(bdev
->bd_inode
);
1223 while (bsize
< PAGE_SIZE
) {
1228 bdev
->bd_block_size
= bsize
;
1229 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1231 EXPORT_SYMBOL(bd_set_size
);
1233 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1238 * mutex_lock(part->bd_mutex)
1239 * mutex_lock_nested(whole->bd_mutex, 1)
1242 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1244 struct gendisk
*disk
;
1245 struct module
*owner
;
1250 if (mode
& FMODE_READ
)
1252 if (mode
& FMODE_WRITE
)
1255 * hooks: /n/, see "layering violations".
1258 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1268 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1271 owner
= disk
->fops
->owner
;
1273 disk_block_events(disk
);
1274 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1275 if (!bdev
->bd_openers
) {
1276 bdev
->bd_disk
= disk
;
1277 bdev
->bd_queue
= disk
->queue
;
1278 bdev
->bd_contains
= bdev
;
1279 bdev
->bd_inode
->i_flags
= 0;
1283 bdev
->bd_part
= disk_get_part(disk
, partno
);
1288 if (disk
->fops
->open
) {
1289 ret
= disk
->fops
->open(bdev
, mode
);
1290 if (ret
== -ERESTARTSYS
) {
1291 /* Lost a race with 'disk' being
1292 * deleted, try again.
1295 disk_put_part(bdev
->bd_part
);
1296 bdev
->bd_part
= NULL
;
1297 bdev
->bd_disk
= NULL
;
1298 bdev
->bd_queue
= NULL
;
1299 mutex_unlock(&bdev
->bd_mutex
);
1300 disk_unblock_events(disk
);
1308 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1309 if (!bdev_dax_capable(bdev
))
1310 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1314 * If the device is invalidated, rescan partition
1315 * if open succeeded or failed with -ENOMEDIUM.
1316 * The latter is necessary to prevent ghost
1317 * partitions on a removed medium.
1319 if (bdev
->bd_invalidated
) {
1321 rescan_partitions(disk
, bdev
);
1322 else if (ret
== -ENOMEDIUM
)
1323 invalidate_partitions(disk
, bdev
);
1329 struct block_device
*whole
;
1330 whole
= bdget_disk(disk
, 0);
1335 ret
= __blkdev_get(whole
, mode
, 1);
1338 bdev
->bd_contains
= whole
;
1339 bdev
->bd_part
= disk_get_part(disk
, partno
);
1340 if (!(disk
->flags
& GENHD_FL_UP
) ||
1341 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1345 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1346 if (!bdev_dax_capable(bdev
))
1347 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1350 if (bdev
->bd_contains
== bdev
) {
1352 if (bdev
->bd_disk
->fops
->open
)
1353 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1354 /* the same as first opener case, read comment there */
1355 if (bdev
->bd_invalidated
) {
1357 rescan_partitions(bdev
->bd_disk
, bdev
);
1358 else if (ret
== -ENOMEDIUM
)
1359 invalidate_partitions(bdev
->bd_disk
, bdev
);
1362 goto out_unlock_bdev
;
1364 /* only one opener holds refs to the module and disk */
1370 bdev
->bd_part_count
++;
1371 mutex_unlock(&bdev
->bd_mutex
);
1372 disk_unblock_events(disk
);
1376 disk_put_part(bdev
->bd_part
);
1377 bdev
->bd_disk
= NULL
;
1378 bdev
->bd_part
= NULL
;
1379 bdev
->bd_queue
= NULL
;
1380 if (bdev
!= bdev
->bd_contains
)
1381 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1382 bdev
->bd_contains
= NULL
;
1384 mutex_unlock(&bdev
->bd_mutex
);
1385 disk_unblock_events(disk
);
1395 * blkdev_get - open a block device
1396 * @bdev: block_device to open
1397 * @mode: FMODE_* mask
1398 * @holder: exclusive holder identifier
1400 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1401 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1402 * @holder is invalid. Exclusive opens may nest for the same @holder.
1404 * On success, the reference count of @bdev is unchanged. On failure,
1411 * 0 on success, -errno on failure.
1413 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1415 struct block_device
*whole
= NULL
;
1418 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1420 if ((mode
& FMODE_EXCL
) && holder
) {
1421 whole
= bd_start_claiming(bdev
, holder
);
1422 if (IS_ERR(whole
)) {
1424 return PTR_ERR(whole
);
1428 res
= __blkdev_get(bdev
, mode
, 0);
1431 struct gendisk
*disk
= whole
->bd_disk
;
1433 /* finish claiming */
1434 mutex_lock(&bdev
->bd_mutex
);
1435 spin_lock(&bdev_lock
);
1438 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1440 * Note that for a whole device bd_holders
1441 * will be incremented twice, and bd_holder
1442 * will be set to bd_may_claim before being
1445 whole
->bd_holders
++;
1446 whole
->bd_holder
= bd_may_claim
;
1448 bdev
->bd_holder
= holder
;
1451 /* tell others that we're done */
1452 BUG_ON(whole
->bd_claiming
!= holder
);
1453 whole
->bd_claiming
= NULL
;
1454 wake_up_bit(&whole
->bd_claiming
, 0);
1456 spin_unlock(&bdev_lock
);
1459 * Block event polling for write claims if requested. Any
1460 * write holder makes the write_holder state stick until
1461 * all are released. This is good enough and tracking
1462 * individual writeable reference is too fragile given the
1463 * way @mode is used in blkdev_get/put().
1465 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1466 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1467 bdev
->bd_write_holder
= true;
1468 disk_block_events(disk
);
1471 mutex_unlock(&bdev
->bd_mutex
);
1477 EXPORT_SYMBOL(blkdev_get
);
1480 * blkdev_get_by_path - open a block device by name
1481 * @path: path to the block device to open
1482 * @mode: FMODE_* mask
1483 * @holder: exclusive holder identifier
1485 * Open the blockdevice described by the device file at @path. @mode
1486 * and @holder are identical to blkdev_get().
1488 * On success, the returned block_device has reference count of one.
1494 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1496 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1499 struct block_device
*bdev
;
1502 bdev
= lookup_bdev(path
);
1506 err
= blkdev_get(bdev
, mode
, holder
);
1508 return ERR_PTR(err
);
1510 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1511 blkdev_put(bdev
, mode
);
1512 return ERR_PTR(-EACCES
);
1517 EXPORT_SYMBOL(blkdev_get_by_path
);
1520 * blkdev_get_by_dev - open a block device by device number
1521 * @dev: device number of block device to open
1522 * @mode: FMODE_* mask
1523 * @holder: exclusive holder identifier
1525 * Open the blockdevice described by device number @dev. @mode and
1526 * @holder are identical to blkdev_get().
1528 * Use it ONLY if you really do not have anything better - i.e. when
1529 * you are behind a truly sucky interface and all you are given is a
1530 * device number. _Never_ to be used for internal purposes. If you
1531 * ever need it - reconsider your API.
1533 * On success, the returned block_device has reference count of one.
1539 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1541 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1543 struct block_device
*bdev
;
1548 return ERR_PTR(-ENOMEM
);
1550 err
= blkdev_get(bdev
, mode
, holder
);
1552 return ERR_PTR(err
);
1556 EXPORT_SYMBOL(blkdev_get_by_dev
);
1558 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1560 struct block_device
*bdev
;
1563 * Preserve backwards compatibility and allow large file access
1564 * even if userspace doesn't ask for it explicitly. Some mkfs
1565 * binary needs it. We might want to drop this workaround
1566 * during an unstable branch.
1568 filp
->f_flags
|= O_LARGEFILE
;
1570 if (filp
->f_flags
& O_NDELAY
)
1571 filp
->f_mode
|= FMODE_NDELAY
;
1572 if (filp
->f_flags
& O_EXCL
)
1573 filp
->f_mode
|= FMODE_EXCL
;
1574 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1575 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1577 bdev
= bd_acquire(inode
);
1581 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1583 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1586 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1588 struct gendisk
*disk
= bdev
->bd_disk
;
1589 struct block_device
*victim
= NULL
;
1591 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1593 bdev
->bd_part_count
--;
1595 if (!--bdev
->bd_openers
) {
1596 WARN_ON_ONCE(bdev
->bd_holders
);
1597 sync_blockdev(bdev
);
1600 bdev_write_inode(bdev
);
1602 * Detaching bdev inode from its wb in __destroy_inode()
1603 * is too late: the queue which embeds its bdi (along with
1604 * root wb) can be gone as soon as we put_disk() below.
1606 inode_detach_wb(bdev
->bd_inode
);
1608 if (bdev
->bd_contains
== bdev
) {
1609 if (disk
->fops
->release
)
1610 disk
->fops
->release(disk
, mode
);
1612 if (!bdev
->bd_openers
) {
1613 struct module
*owner
= disk
->fops
->owner
;
1615 disk_put_part(bdev
->bd_part
);
1616 bdev
->bd_part
= NULL
;
1617 bdev
->bd_disk
= NULL
;
1618 if (bdev
!= bdev
->bd_contains
)
1619 victim
= bdev
->bd_contains
;
1620 bdev
->bd_contains
= NULL
;
1625 mutex_unlock(&bdev
->bd_mutex
);
1628 __blkdev_put(victim
, mode
, 1);
1631 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1633 mutex_lock(&bdev
->bd_mutex
);
1635 if (mode
& FMODE_EXCL
) {
1639 * Release a claim on the device. The holder fields
1640 * are protected with bdev_lock. bd_mutex is to
1641 * synchronize disk_holder unlinking.
1643 spin_lock(&bdev_lock
);
1645 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1646 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1648 /* bd_contains might point to self, check in a separate step */
1649 if ((bdev_free
= !bdev
->bd_holders
))
1650 bdev
->bd_holder
= NULL
;
1651 if (!bdev
->bd_contains
->bd_holders
)
1652 bdev
->bd_contains
->bd_holder
= NULL
;
1654 spin_unlock(&bdev_lock
);
1657 * If this was the last claim, remove holder link and
1658 * unblock evpoll if it was a write holder.
1660 if (bdev_free
&& bdev
->bd_write_holder
) {
1661 disk_unblock_events(bdev
->bd_disk
);
1662 bdev
->bd_write_holder
= false;
1667 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1668 * event. This is to ensure detection of media removal commanded
1669 * from userland - e.g. eject(1).
1671 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1673 mutex_unlock(&bdev
->bd_mutex
);
1675 __blkdev_put(bdev
, mode
, 0);
1677 EXPORT_SYMBOL(blkdev_put
);
1679 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1681 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1682 blkdev_put(bdev
, filp
->f_mode
);
1686 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1688 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1689 fmode_t mode
= file
->f_mode
;
1692 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1693 * to updated it before every ioctl.
1695 if (file
->f_flags
& O_NDELAY
)
1696 mode
|= FMODE_NDELAY
;
1698 mode
&= ~FMODE_NDELAY
;
1700 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1704 * Write data to the block device. Only intended for the block device itself
1705 * and the raw driver which basically is a fake block device.
1707 * Does not take i_mutex for the write and thus is not for general purpose
1710 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1712 struct file
*file
= iocb
->ki_filp
;
1713 struct inode
*bd_inode
= bdev_file_inode(file
);
1714 loff_t size
= i_size_read(bd_inode
);
1715 struct blk_plug plug
;
1718 if (bdev_read_only(I_BDEV(bd_inode
)))
1721 if (!iov_iter_count(from
))
1724 if (iocb
->ki_pos
>= size
)
1727 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1729 blk_start_plug(&plug
);
1730 ret
= __generic_file_write_iter(iocb
, from
);
1732 ret
= generic_write_sync(iocb
, ret
);
1733 blk_finish_plug(&plug
);
1736 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1738 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1740 struct file
*file
= iocb
->ki_filp
;
1741 struct inode
*bd_inode
= bdev_file_inode(file
);
1742 loff_t size
= i_size_read(bd_inode
);
1743 loff_t pos
= iocb
->ki_pos
;
1749 iov_iter_truncate(to
, size
);
1750 return generic_file_read_iter(iocb
, to
);
1752 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1755 * Try to release a page associated with block device when the system
1756 * is under memory pressure.
1758 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1760 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1762 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1763 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1765 return try_to_free_buffers(page
);
1768 static int blkdev_writepages(struct address_space
*mapping
,
1769 struct writeback_control
*wbc
)
1771 if (dax_mapping(mapping
)) {
1772 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1774 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1776 return generic_writepages(mapping
, wbc
);
1779 static const struct address_space_operations def_blk_aops
= {
1780 .readpage
= blkdev_readpage
,
1781 .readpages
= blkdev_readpages
,
1782 .writepage
= blkdev_writepage
,
1783 .write_begin
= blkdev_write_begin
,
1784 .write_end
= blkdev_write_end
,
1785 .writepages
= blkdev_writepages
,
1786 .releasepage
= blkdev_releasepage
,
1787 .direct_IO
= blkdev_direct_IO
,
1788 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1791 const struct file_operations def_blk_fops
= {
1792 .open
= blkdev_open
,
1793 .release
= blkdev_close
,
1794 .llseek
= block_llseek
,
1795 .read_iter
= blkdev_read_iter
,
1796 .write_iter
= blkdev_write_iter
,
1797 .mmap
= generic_file_mmap
,
1798 .fsync
= blkdev_fsync
,
1799 .unlocked_ioctl
= block_ioctl
,
1800 #ifdef CONFIG_COMPAT
1801 .compat_ioctl
= compat_blkdev_ioctl
,
1803 .splice_read
= generic_file_splice_read
,
1804 .splice_write
= iter_file_splice_write
,
1807 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1810 mm_segment_t old_fs
= get_fs();
1812 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1817 EXPORT_SYMBOL(ioctl_by_bdev
);
1820 * lookup_bdev - lookup a struct block_device by name
1821 * @pathname: special file representing the block device
1823 * Get a reference to the blockdevice at @pathname in the current
1824 * namespace if possible and return it. Return ERR_PTR(error)
1827 struct block_device
*lookup_bdev(const char *pathname
)
1829 struct block_device
*bdev
;
1830 struct inode
*inode
;
1834 if (!pathname
|| !*pathname
)
1835 return ERR_PTR(-EINVAL
);
1837 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1839 return ERR_PTR(error
);
1841 inode
= d_backing_inode(path
.dentry
);
1843 if (!S_ISBLK(inode
->i_mode
))
1846 if (!may_open_dev(&path
))
1849 bdev
= bd_acquire(inode
);
1856 bdev
= ERR_PTR(error
);
1859 EXPORT_SYMBOL(lookup_bdev
);
1861 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1863 struct super_block
*sb
= get_super(bdev
);
1868 * no need to lock the super, get_super holds the
1869 * read mutex so the filesystem cannot go away
1870 * under us (->put_super runs with the write lock
1873 shrink_dcache_sb(sb
);
1874 res
= invalidate_inodes(sb
, kill_dirty
);
1877 invalidate_bdev(bdev
);
1880 EXPORT_SYMBOL(__invalidate_device
);
1882 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1884 struct inode
*inode
, *old_inode
= NULL
;
1886 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1887 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1888 struct address_space
*mapping
= inode
->i_mapping
;
1890 spin_lock(&inode
->i_lock
);
1891 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1892 mapping
->nrpages
== 0) {
1893 spin_unlock(&inode
->i_lock
);
1897 spin_unlock(&inode
->i_lock
);
1898 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1900 * We hold a reference to 'inode' so it couldn't have been
1901 * removed from s_inodes list while we dropped the
1902 * s_inode_list_lock We cannot iput the inode now as we can
1903 * be holding the last reference and we cannot iput it under
1904 * s_inode_list_lock. So we keep the reference and iput it
1910 func(I_BDEV(inode
), arg
);
1912 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1914 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);