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/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <asm/uaccess.h>
34 struct block_device bdev
;
35 struct inode vfs_inode
;
38 static const struct address_space_operations def_blk_aops
;
40 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
42 return container_of(inode
, struct bdev_inode
, vfs_inode
);
45 inline struct block_device
*I_BDEV(struct inode
*inode
)
47 return &BDEV_I(inode
)->bdev
;
49 EXPORT_SYMBOL(I_BDEV
);
51 static void bdev_write_inode(struct inode
*inode
)
53 spin_lock(&inode
->i_lock
);
54 while (inode
->i_state
& I_DIRTY
) {
55 spin_unlock(&inode
->i_lock
);
56 WARN_ON_ONCE(write_inode_now(inode
, true));
57 spin_lock(&inode
->i_lock
);
59 spin_unlock(&inode
->i_lock
);
62 /* Kill _all_ buffers and pagecache , dirty or not.. */
63 void kill_bdev(struct block_device
*bdev
)
65 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
67 if (mapping
->nrpages
== 0 && mapping
->nrshadows
== 0)
71 truncate_inode_pages(mapping
, 0);
73 EXPORT_SYMBOL(kill_bdev
);
75 /* Invalidate clean unused buffers and pagecache. */
76 void invalidate_bdev(struct block_device
*bdev
)
78 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
80 if (mapping
->nrpages
== 0)
84 lru_add_drain_all(); /* make sure all lru add caches are flushed */
85 invalidate_mapping_pages(mapping
, 0, -1);
86 /* 99% of the time, we don't need to flush the cleancache on the bdev.
87 * But, for the strange corners, lets be cautious
89 cleancache_invalidate_inode(mapping
);
91 EXPORT_SYMBOL(invalidate_bdev
);
93 int set_blocksize(struct block_device
*bdev
, int size
)
95 /* Size must be a power of two, and between 512 and PAGE_SIZE */
96 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
99 /* Size cannot be smaller than the size supported by the device */
100 if (size
< bdev_logical_block_size(bdev
))
103 /* Don't change the size if it is same as current */
104 if (bdev
->bd_block_size
!= size
) {
106 bdev
->bd_block_size
= size
;
107 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
113 EXPORT_SYMBOL(set_blocksize
);
115 int sb_set_blocksize(struct super_block
*sb
, int size
)
117 if (set_blocksize(sb
->s_bdev
, size
))
119 /* If we get here, we know size is power of two
120 * and it's value is between 512 and PAGE_SIZE */
121 sb
->s_blocksize
= size
;
122 sb
->s_blocksize_bits
= blksize_bits(size
);
123 return sb
->s_blocksize
;
126 EXPORT_SYMBOL(sb_set_blocksize
);
128 int sb_min_blocksize(struct super_block
*sb
, int size
)
130 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
133 return sb_set_blocksize(sb
, size
);
136 EXPORT_SYMBOL(sb_min_blocksize
);
139 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
140 struct buffer_head
*bh
, int create
)
142 bh
->b_bdev
= I_BDEV(inode
);
143 bh
->b_blocknr
= iblock
;
144 set_buffer_mapped(bh
);
149 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
151 struct file
*file
= iocb
->ki_filp
;
152 struct inode
*inode
= file
->f_mapping
->host
;
155 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
156 NULL
, DIO_SKIP_DIO_COUNT
);
157 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
158 blkdev_get_block
, NULL
, NULL
,
162 int __sync_blockdev(struct block_device
*bdev
, int wait
)
167 return filemap_flush(bdev
->bd_inode
->i_mapping
);
168 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
172 * Write out and wait upon all the dirty data associated with a block
173 * device via its mapping. Does not take the superblock lock.
175 int sync_blockdev(struct block_device
*bdev
)
177 return __sync_blockdev(bdev
, 1);
179 EXPORT_SYMBOL(sync_blockdev
);
182 * Write out and wait upon all dirty data associated with this
183 * device. Filesystem data as well as the underlying block
184 * device. Takes the superblock lock.
186 int fsync_bdev(struct block_device
*bdev
)
188 struct super_block
*sb
= get_super(bdev
);
190 int res
= sync_filesystem(sb
);
194 return sync_blockdev(bdev
);
196 EXPORT_SYMBOL(fsync_bdev
);
199 * freeze_bdev -- lock a filesystem and force it into a consistent state
200 * @bdev: blockdevice to lock
202 * If a superblock is found on this device, we take the s_umount semaphore
203 * on it to make sure nobody unmounts until the snapshot creation is done.
204 * The reference counter (bd_fsfreeze_count) guarantees that only the last
205 * unfreeze process can unfreeze the frozen filesystem actually when multiple
206 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
207 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
210 struct super_block
*freeze_bdev(struct block_device
*bdev
)
212 struct super_block
*sb
;
215 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
216 if (++bdev
->bd_fsfreeze_count
> 1) {
218 * We don't even need to grab a reference - the first call
219 * to freeze_bdev grab an active reference and only the last
220 * thaw_bdev drops it.
222 sb
= get_super(bdev
);
224 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
228 sb
= get_active_super(bdev
);
231 if (sb
->s_op
->freeze_super
)
232 error
= sb
->s_op
->freeze_super(sb
);
234 error
= freeze_super(sb
);
236 deactivate_super(sb
);
237 bdev
->bd_fsfreeze_count
--;
238 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
239 return ERR_PTR(error
);
241 deactivate_super(sb
);
244 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
245 return sb
; /* thaw_bdev releases s->s_umount */
247 EXPORT_SYMBOL(freeze_bdev
);
250 * thaw_bdev -- unlock filesystem
251 * @bdev: blockdevice to unlock
252 * @sb: associated superblock
254 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
256 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
260 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
261 if (!bdev
->bd_fsfreeze_count
)
265 if (--bdev
->bd_fsfreeze_count
> 0)
271 if (sb
->s_op
->thaw_super
)
272 error
= sb
->s_op
->thaw_super(sb
);
274 error
= thaw_super(sb
);
276 bdev
->bd_fsfreeze_count
++;
277 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
281 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
284 EXPORT_SYMBOL(thaw_bdev
);
286 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
288 return block_write_full_page(page
, blkdev_get_block
, wbc
);
291 static int blkdev_readpage(struct file
* file
, struct page
* page
)
293 return block_read_full_page(page
, blkdev_get_block
);
296 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
297 struct list_head
*pages
, unsigned nr_pages
)
299 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
302 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
303 loff_t pos
, unsigned len
, unsigned flags
,
304 struct page
**pagep
, void **fsdata
)
306 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
310 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
311 loff_t pos
, unsigned len
, unsigned copied
,
312 struct page
*page
, void *fsdata
)
315 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
318 page_cache_release(page
);
325 * for a block special file file_inode(file)->i_size is zero
326 * so we compute the size by hand (just as in block_read/write above)
328 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
330 struct inode
*bd_inode
= file
->f_mapping
->host
;
333 mutex_lock(&bd_inode
->i_mutex
);
334 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
335 mutex_unlock(&bd_inode
->i_mutex
);
339 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
341 struct inode
*bd_inode
= filp
->f_mapping
->host
;
342 struct block_device
*bdev
= I_BDEV(bd_inode
);
345 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
350 * There is no need to serialise calls to blkdev_issue_flush with
351 * i_mutex and doing so causes performance issues with concurrent
352 * O_SYNC writers to a block device.
354 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
355 if (error
== -EOPNOTSUPP
)
360 EXPORT_SYMBOL(blkdev_fsync
);
363 * bdev_read_page() - Start reading a page from a block device
364 * @bdev: The device to read the page from
365 * @sector: The offset on the device to read the page to (need not be aligned)
366 * @page: The page to read
368 * On entry, the page should be locked. It will be unlocked when the page
369 * has been read. If the block driver implements rw_page synchronously,
370 * that will be true on exit from this function, but it need not be.
372 * Errors returned by this function are usually "soft", eg out of memory, or
373 * queue full; callers should try a different route to read this page rather
374 * than propagate an error back up the stack.
376 * Return: negative errno if an error occurs, 0 if submission was successful.
378 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
381 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
384 return ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
386 EXPORT_SYMBOL_GPL(bdev_read_page
);
389 * bdev_write_page() - Start writing a page to a block device
390 * @bdev: The device to write the page to
391 * @sector: The offset on the device to write the page to (need not be aligned)
392 * @page: The page to write
393 * @wbc: The writeback_control for the write
395 * On entry, the page should be locked and not currently under writeback.
396 * On exit, if the write started successfully, the page will be unlocked and
397 * under writeback. If the write failed already (eg the driver failed to
398 * queue the page to the device), the page will still be locked. If the
399 * caller is a ->writepage implementation, it will need to unlock the page.
401 * Errors returned by this function are usually "soft", eg out of memory, or
402 * queue full; callers should try a different route to write 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_write_page(struct block_device
*bdev
, sector_t sector
,
408 struct page
*page
, struct writeback_control
*wbc
)
411 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
412 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
415 set_page_writeback(page
);
416 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
418 end_page_writeback(page
);
423 EXPORT_SYMBOL_GPL(bdev_write_page
);
426 * bdev_direct_access() - Get the address for directly-accessibly memory
427 * @bdev: The device containing the memory
428 * @sector: The offset within the device
429 * @addr: Where to put the address of the memory
430 * @pfn: The Page Frame Number for the memory
431 * @size: The number of bytes requested
433 * If a block device is made up of directly addressable memory, this function
434 * will tell the caller the PFN and the address of the memory. The address
435 * may be directly dereferenced within the kernel without the need to call
436 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
439 * Return: negative errno if an error occurs, otherwise the number of bytes
440 * accessible at this address.
442 long bdev_direct_access(struct block_device
*bdev
, sector_t sector
,
443 void **addr
, unsigned long *pfn
, long size
)
446 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
449 * The device driver is allowed to sleep, in order to make the
450 * memory directly accessible.
456 if (!ops
->direct_access
)
458 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
459 part_nr_sects_read(bdev
->bd_part
))
461 sector
+= get_start_sect(bdev
);
462 if (sector
% (PAGE_SIZE
/ 512))
464 avail
= ops
->direct_access(bdev
, sector
, addr
, pfn
, size
);
467 return min(avail
, size
);
469 EXPORT_SYMBOL_GPL(bdev_direct_access
);
475 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
476 static struct kmem_cache
* bdev_cachep __read_mostly
;
478 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
480 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
483 return &ei
->vfs_inode
;
486 static void bdev_i_callback(struct rcu_head
*head
)
488 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
489 struct bdev_inode
*bdi
= BDEV_I(inode
);
491 kmem_cache_free(bdev_cachep
, bdi
);
494 static void bdev_destroy_inode(struct inode
*inode
)
496 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
499 static void init_once(void *foo
)
501 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
502 struct block_device
*bdev
= &ei
->bdev
;
504 memset(bdev
, 0, sizeof(*bdev
));
505 mutex_init(&bdev
->bd_mutex
);
506 INIT_LIST_HEAD(&bdev
->bd_inodes
);
507 INIT_LIST_HEAD(&bdev
->bd_list
);
509 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
511 inode_init_once(&ei
->vfs_inode
);
512 /* Initialize mutex for freeze. */
513 mutex_init(&bdev
->bd_fsfreeze_mutex
);
516 static inline void __bd_forget(struct inode
*inode
)
518 list_del_init(&inode
->i_devices
);
519 inode
->i_bdev
= NULL
;
520 inode
->i_mapping
= &inode
->i_data
;
523 static void bdev_evict_inode(struct inode
*inode
)
525 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
527 truncate_inode_pages_final(&inode
->i_data
);
528 invalidate_inode_buffers(inode
); /* is it needed here? */
530 spin_lock(&bdev_lock
);
531 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
532 __bd_forget(list_entry(p
, struct inode
, i_devices
));
534 list_del_init(&bdev
->bd_list
);
535 spin_unlock(&bdev_lock
);
538 static const struct super_operations bdev_sops
= {
539 .statfs
= simple_statfs
,
540 .alloc_inode
= bdev_alloc_inode
,
541 .destroy_inode
= bdev_destroy_inode
,
542 .drop_inode
= generic_delete_inode
,
543 .evict_inode
= bdev_evict_inode
,
546 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
547 int flags
, const char *dev_name
, void *data
)
549 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
552 static struct file_system_type bd_type
= {
555 .kill_sb
= kill_anon_super
,
558 static struct super_block
*blockdev_superblock __read_mostly
;
560 void __init
bdev_cache_init(void)
563 static struct vfsmount
*bd_mnt
;
565 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
566 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
567 SLAB_MEM_SPREAD
|SLAB_PANIC
),
569 err
= register_filesystem(&bd_type
);
571 panic("Cannot register bdev pseudo-fs");
572 bd_mnt
= kern_mount(&bd_type
);
574 panic("Cannot create bdev pseudo-fs");
575 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
579 * Most likely _very_ bad one - but then it's hardly critical for small
580 * /dev and can be fixed when somebody will need really large one.
581 * Keep in mind that it will be fed through icache hash function too.
583 static inline unsigned long hash(dev_t dev
)
585 return MAJOR(dev
)+MINOR(dev
);
588 static int bdev_test(struct inode
*inode
, void *data
)
590 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
593 static int bdev_set(struct inode
*inode
, void *data
)
595 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
599 static LIST_HEAD(all_bdevs
);
601 struct block_device
*bdget(dev_t dev
)
603 struct block_device
*bdev
;
606 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
607 bdev_test
, bdev_set
, &dev
);
612 bdev
= &BDEV_I(inode
)->bdev
;
614 if (inode
->i_state
& I_NEW
) {
615 bdev
->bd_contains
= NULL
;
616 bdev
->bd_super
= NULL
;
617 bdev
->bd_inode
= inode
;
618 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
619 bdev
->bd_part_count
= 0;
620 bdev
->bd_invalidated
= 0;
621 inode
->i_mode
= S_IFBLK
;
623 inode
->i_bdev
= bdev
;
624 inode
->i_data
.a_ops
= &def_blk_aops
;
625 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
626 spin_lock(&bdev_lock
);
627 list_add(&bdev
->bd_list
, &all_bdevs
);
628 spin_unlock(&bdev_lock
);
629 unlock_new_inode(inode
);
634 EXPORT_SYMBOL(bdget
);
637 * bdgrab -- Grab a reference to an already referenced block device
638 * @bdev: Block device to grab a reference to.
640 struct block_device
*bdgrab(struct block_device
*bdev
)
642 ihold(bdev
->bd_inode
);
645 EXPORT_SYMBOL(bdgrab
);
647 long nr_blockdev_pages(void)
649 struct block_device
*bdev
;
651 spin_lock(&bdev_lock
);
652 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
653 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
655 spin_unlock(&bdev_lock
);
659 void bdput(struct block_device
*bdev
)
661 iput(bdev
->bd_inode
);
664 EXPORT_SYMBOL(bdput
);
666 static struct block_device
*bd_acquire(struct inode
*inode
)
668 struct block_device
*bdev
;
670 spin_lock(&bdev_lock
);
671 bdev
= inode
->i_bdev
;
673 ihold(bdev
->bd_inode
);
674 spin_unlock(&bdev_lock
);
677 spin_unlock(&bdev_lock
);
679 bdev
= bdget(inode
->i_rdev
);
681 spin_lock(&bdev_lock
);
682 if (!inode
->i_bdev
) {
684 * We take an additional reference to bd_inode,
685 * and it's released in clear_inode() of inode.
686 * So, we can access it via ->i_mapping always
689 ihold(bdev
->bd_inode
);
690 inode
->i_bdev
= bdev
;
691 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
692 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
694 spin_unlock(&bdev_lock
);
699 int sb_is_blkdev_sb(struct super_block
*sb
)
701 return sb
== blockdev_superblock
;
704 /* Call when you free inode */
706 void bd_forget(struct inode
*inode
)
708 struct block_device
*bdev
= NULL
;
710 spin_lock(&bdev_lock
);
711 if (!sb_is_blkdev_sb(inode
->i_sb
))
712 bdev
= inode
->i_bdev
;
714 spin_unlock(&bdev_lock
);
717 iput(bdev
->bd_inode
);
721 * bd_may_claim - test whether a block device can be claimed
722 * @bdev: block device of interest
723 * @whole: whole block device containing @bdev, may equal @bdev
724 * @holder: holder trying to claim @bdev
726 * Test whether @bdev can be claimed by @holder.
729 * spin_lock(&bdev_lock).
732 * %true if @bdev can be claimed, %false otherwise.
734 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
737 if (bdev
->bd_holder
== holder
)
738 return true; /* already a holder */
739 else if (bdev
->bd_holder
!= NULL
)
740 return false; /* held by someone else */
741 else if (bdev
->bd_contains
== bdev
)
742 return true; /* is a whole device which isn't held */
744 else if (whole
->bd_holder
== bd_may_claim
)
745 return true; /* is a partition of a device that is being partitioned */
746 else if (whole
->bd_holder
!= NULL
)
747 return false; /* is a partition of a held device */
749 return true; /* is a partition of an un-held device */
753 * bd_prepare_to_claim - prepare to claim a block device
754 * @bdev: block device of interest
755 * @whole: the whole device containing @bdev, may equal @bdev
756 * @holder: holder trying to claim @bdev
758 * Prepare to claim @bdev. This function fails if @bdev is already
759 * claimed by another holder and waits if another claiming is in
760 * progress. This function doesn't actually claim. On successful
761 * return, the caller has ownership of bd_claiming and bd_holder[s].
764 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
768 * 0 if @bdev can be claimed, -EBUSY otherwise.
770 static int bd_prepare_to_claim(struct block_device
*bdev
,
771 struct block_device
*whole
, void *holder
)
774 /* if someone else claimed, fail */
775 if (!bd_may_claim(bdev
, whole
, holder
))
778 /* if claiming is already in progress, wait for it to finish */
779 if (whole
->bd_claiming
) {
780 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
783 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
784 spin_unlock(&bdev_lock
);
786 finish_wait(wq
, &wait
);
787 spin_lock(&bdev_lock
);
796 * bd_start_claiming - start claiming a block device
797 * @bdev: block device of interest
798 * @holder: holder trying to claim @bdev
800 * @bdev is about to be opened exclusively. Check @bdev can be opened
801 * exclusively and mark that an exclusive open is in progress. Each
802 * successful call to this function must be matched with a call to
803 * either bd_finish_claiming() or bd_abort_claiming() (which do not
806 * This function is used to gain exclusive access to the block device
807 * without actually causing other exclusive open attempts to fail. It
808 * should be used when the open sequence itself requires exclusive
809 * access but may subsequently fail.
815 * Pointer to the block device containing @bdev on success, ERR_PTR()
818 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
821 struct gendisk
*disk
;
822 struct block_device
*whole
;
828 * @bdev might not have been initialized properly yet, look up
829 * and grab the outer block device the hard way.
831 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
833 return ERR_PTR(-ENXIO
);
836 * Normally, @bdev should equal what's returned from bdget_disk()
837 * if partno is 0; however, some drivers (floppy) use multiple
838 * bdev's for the same physical device and @bdev may be one of the
839 * aliases. Keep @bdev if partno is 0. This means claimer
840 * tracking is broken for those devices but it has always been that
844 whole
= bdget_disk(disk
, 0);
846 whole
= bdgrab(bdev
);
848 module_put(disk
->fops
->owner
);
851 return ERR_PTR(-ENOMEM
);
853 /* prepare to claim, if successful, mark claiming in progress */
854 spin_lock(&bdev_lock
);
856 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
858 whole
->bd_claiming
= holder
;
859 spin_unlock(&bdev_lock
);
862 spin_unlock(&bdev_lock
);
869 struct bd_holder_disk
{
870 struct list_head list
;
871 struct gendisk
*disk
;
875 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
876 struct gendisk
*disk
)
878 struct bd_holder_disk
*holder
;
880 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
881 if (holder
->disk
== disk
)
886 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
888 return sysfs_create_link(from
, to
, kobject_name(to
));
891 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
893 sysfs_remove_link(from
, kobject_name(to
));
897 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
898 * @bdev: the claimed slave bdev
899 * @disk: the holding disk
901 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
903 * This functions creates the following sysfs symlinks.
905 * - from "slaves" directory of the holder @disk to the claimed @bdev
906 * - from "holders" directory of the @bdev to the holder @disk
908 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
909 * passed to bd_link_disk_holder(), then:
911 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
912 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
914 * The caller must have claimed @bdev before calling this function and
915 * ensure that both @bdev and @disk are valid during the creation and
916 * lifetime of these symlinks.
922 * 0 on success, -errno on failure.
924 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
926 struct bd_holder_disk
*holder
;
929 mutex_lock(&bdev
->bd_mutex
);
931 WARN_ON_ONCE(!bdev
->bd_holder
);
933 /* FIXME: remove the following once add_disk() handles errors */
934 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
937 holder
= bd_find_holder_disk(bdev
, disk
);
943 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
949 INIT_LIST_HEAD(&holder
->list
);
953 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
957 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
961 * bdev could be deleted beneath us which would implicitly destroy
962 * the holder directory. Hold on to it.
964 kobject_get(bdev
->bd_part
->holder_dir
);
966 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
970 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
974 mutex_unlock(&bdev
->bd_mutex
);
977 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
980 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
981 * @bdev: the calimed slave bdev
982 * @disk: the holding disk
984 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
989 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
991 struct bd_holder_disk
*holder
;
993 mutex_lock(&bdev
->bd_mutex
);
995 holder
= bd_find_holder_disk(bdev
, disk
);
997 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
998 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
999 del_symlink(bdev
->bd_part
->holder_dir
,
1000 &disk_to_dev(disk
)->kobj
);
1001 kobject_put(bdev
->bd_part
->holder_dir
);
1002 list_del_init(&holder
->list
);
1006 mutex_unlock(&bdev
->bd_mutex
);
1008 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1012 * flush_disk - invalidates all buffer-cache entries on a disk
1014 * @bdev: struct block device to be flushed
1015 * @kill_dirty: flag to guide handling of dirty inodes
1017 * Invalidates all buffer-cache entries on a disk. It should be called
1018 * when a disk has been changed -- either by a media change or online
1021 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1023 if (__invalidate_device(bdev
, kill_dirty
)) {
1024 char name
[BDEVNAME_SIZE
] = "";
1027 disk_name(bdev
->bd_disk
, 0, name
);
1028 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1029 "resized disk %s\n", name
);
1034 if (disk_part_scan_enabled(bdev
->bd_disk
))
1035 bdev
->bd_invalidated
= 1;
1039 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1040 * @disk: struct gendisk to check
1041 * @bdev: struct bdev to adjust.
1043 * This routine checks to see if the bdev size does not match the disk size
1044 * and adjusts it if it differs.
1046 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1048 loff_t disk_size
, bdev_size
;
1050 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1051 bdev_size
= i_size_read(bdev
->bd_inode
);
1052 if (disk_size
!= bdev_size
) {
1053 char name
[BDEVNAME_SIZE
];
1055 disk_name(disk
, 0, name
);
1057 "%s: detected capacity change from %lld to %lld\n",
1058 name
, bdev_size
, disk_size
);
1059 i_size_write(bdev
->bd_inode
, disk_size
);
1060 flush_disk(bdev
, false);
1063 EXPORT_SYMBOL(check_disk_size_change
);
1066 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1067 * @disk: struct gendisk to be revalidated
1069 * This routine is a wrapper for lower-level driver's revalidate_disk
1070 * call-backs. It is used to do common pre and post operations needed
1071 * for all revalidate_disk operations.
1073 int revalidate_disk(struct gendisk
*disk
)
1075 struct block_device
*bdev
;
1078 if (disk
->fops
->revalidate_disk
)
1079 ret
= disk
->fops
->revalidate_disk(disk
);
1081 bdev
= bdget_disk(disk
, 0);
1085 mutex_lock(&bdev
->bd_mutex
);
1086 check_disk_size_change(disk
, bdev
);
1087 bdev
->bd_invalidated
= 0;
1088 mutex_unlock(&bdev
->bd_mutex
);
1092 EXPORT_SYMBOL(revalidate_disk
);
1095 * This routine checks whether a removable media has been changed,
1096 * and invalidates all buffer-cache-entries in that case. This
1097 * is a relatively slow routine, so we have to try to minimize using
1098 * it. Thus it is called only upon a 'mount' or 'open'. This
1099 * is the best way of combining speed and utility, I think.
1100 * People changing diskettes in the middle of an operation deserve
1103 int check_disk_change(struct block_device
*bdev
)
1105 struct gendisk
*disk
= bdev
->bd_disk
;
1106 const struct block_device_operations
*bdops
= disk
->fops
;
1107 unsigned int events
;
1109 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1110 DISK_EVENT_EJECT_REQUEST
);
1111 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1114 flush_disk(bdev
, true);
1115 if (bdops
->revalidate_disk
)
1116 bdops
->revalidate_disk(bdev
->bd_disk
);
1120 EXPORT_SYMBOL(check_disk_change
);
1122 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1124 unsigned bsize
= bdev_logical_block_size(bdev
);
1126 mutex_lock(&bdev
->bd_inode
->i_mutex
);
1127 i_size_write(bdev
->bd_inode
, size
);
1128 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
1129 while (bsize
< PAGE_CACHE_SIZE
) {
1134 bdev
->bd_block_size
= bsize
;
1135 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1137 EXPORT_SYMBOL(bd_set_size
);
1139 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1144 * mutex_lock(part->bd_mutex)
1145 * mutex_lock_nested(whole->bd_mutex, 1)
1148 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1150 struct gendisk
*disk
;
1151 struct module
*owner
;
1156 if (mode
& FMODE_READ
)
1158 if (mode
& FMODE_WRITE
)
1161 * hooks: /n/, see "layering violations".
1164 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1174 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1177 owner
= disk
->fops
->owner
;
1179 disk_block_events(disk
);
1180 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1181 if (!bdev
->bd_openers
) {
1182 bdev
->bd_disk
= disk
;
1183 bdev
->bd_queue
= disk
->queue
;
1184 bdev
->bd_contains
= bdev
;
1185 bdev
->bd_inode
->i_flags
= disk
->fops
->direct_access
? S_DAX
: 0;
1188 bdev
->bd_part
= disk_get_part(disk
, partno
);
1193 if (disk
->fops
->open
) {
1194 ret
= disk
->fops
->open(bdev
, mode
);
1195 if (ret
== -ERESTARTSYS
) {
1196 /* Lost a race with 'disk' being
1197 * deleted, try again.
1200 disk_put_part(bdev
->bd_part
);
1201 bdev
->bd_part
= NULL
;
1202 bdev
->bd_disk
= NULL
;
1203 bdev
->bd_queue
= NULL
;
1204 mutex_unlock(&bdev
->bd_mutex
);
1205 disk_unblock_events(disk
);
1213 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1216 * If the device is invalidated, rescan partition
1217 * if open succeeded or failed with -ENOMEDIUM.
1218 * The latter is necessary to prevent ghost
1219 * partitions on a removed medium.
1221 if (bdev
->bd_invalidated
) {
1223 rescan_partitions(disk
, bdev
);
1224 else if (ret
== -ENOMEDIUM
)
1225 invalidate_partitions(disk
, bdev
);
1230 struct block_device
*whole
;
1231 whole
= bdget_disk(disk
, 0);
1236 ret
= __blkdev_get(whole
, mode
, 1);
1239 bdev
->bd_contains
= whole
;
1240 bdev
->bd_part
= disk_get_part(disk
, partno
);
1241 if (!(disk
->flags
& GENHD_FL_UP
) ||
1242 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1246 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1249 if (bdev
->bd_contains
== bdev
) {
1251 if (bdev
->bd_disk
->fops
->open
)
1252 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1253 /* the same as first opener case, read comment there */
1254 if (bdev
->bd_invalidated
) {
1256 rescan_partitions(bdev
->bd_disk
, bdev
);
1257 else if (ret
== -ENOMEDIUM
)
1258 invalidate_partitions(bdev
->bd_disk
, bdev
);
1261 goto out_unlock_bdev
;
1263 /* only one opener holds refs to the module and disk */
1269 bdev
->bd_part_count
++;
1270 mutex_unlock(&bdev
->bd_mutex
);
1271 disk_unblock_events(disk
);
1275 disk_put_part(bdev
->bd_part
);
1276 bdev
->bd_disk
= NULL
;
1277 bdev
->bd_part
= NULL
;
1278 bdev
->bd_queue
= NULL
;
1279 if (bdev
!= bdev
->bd_contains
)
1280 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1281 bdev
->bd_contains
= NULL
;
1283 mutex_unlock(&bdev
->bd_mutex
);
1284 disk_unblock_events(disk
);
1294 * blkdev_get - open a block device
1295 * @bdev: block_device to open
1296 * @mode: FMODE_* mask
1297 * @holder: exclusive holder identifier
1299 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1300 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1301 * @holder is invalid. Exclusive opens may nest for the same @holder.
1303 * On success, the reference count of @bdev is unchanged. On failure,
1310 * 0 on success, -errno on failure.
1312 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1314 struct block_device
*whole
= NULL
;
1317 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1319 if ((mode
& FMODE_EXCL
) && holder
) {
1320 whole
= bd_start_claiming(bdev
, holder
);
1321 if (IS_ERR(whole
)) {
1323 return PTR_ERR(whole
);
1327 res
= __blkdev_get(bdev
, mode
, 0);
1330 struct gendisk
*disk
= whole
->bd_disk
;
1332 /* finish claiming */
1333 mutex_lock(&bdev
->bd_mutex
);
1334 spin_lock(&bdev_lock
);
1337 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1339 * Note that for a whole device bd_holders
1340 * will be incremented twice, and bd_holder
1341 * will be set to bd_may_claim before being
1344 whole
->bd_holders
++;
1345 whole
->bd_holder
= bd_may_claim
;
1347 bdev
->bd_holder
= holder
;
1350 /* tell others that we're done */
1351 BUG_ON(whole
->bd_claiming
!= holder
);
1352 whole
->bd_claiming
= NULL
;
1353 wake_up_bit(&whole
->bd_claiming
, 0);
1355 spin_unlock(&bdev_lock
);
1358 * Block event polling for write claims if requested. Any
1359 * write holder makes the write_holder state stick until
1360 * all are released. This is good enough and tracking
1361 * individual writeable reference is too fragile given the
1362 * way @mode is used in blkdev_get/put().
1364 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1365 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1366 bdev
->bd_write_holder
= true;
1367 disk_block_events(disk
);
1370 mutex_unlock(&bdev
->bd_mutex
);
1376 EXPORT_SYMBOL(blkdev_get
);
1379 * blkdev_get_by_path - open a block device by name
1380 * @path: path to the block device to open
1381 * @mode: FMODE_* mask
1382 * @holder: exclusive holder identifier
1384 * Open the blockdevice described by the device file at @path. @mode
1385 * and @holder are identical to blkdev_get().
1387 * On success, the returned block_device has reference count of one.
1393 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1395 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1398 struct block_device
*bdev
;
1401 bdev
= lookup_bdev(path
);
1405 err
= blkdev_get(bdev
, mode
, holder
);
1407 return ERR_PTR(err
);
1409 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1410 blkdev_put(bdev
, mode
);
1411 return ERR_PTR(-EACCES
);
1416 EXPORT_SYMBOL(blkdev_get_by_path
);
1419 * blkdev_get_by_dev - open a block device by device number
1420 * @dev: device number of block device to open
1421 * @mode: FMODE_* mask
1422 * @holder: exclusive holder identifier
1424 * Open the blockdevice described by device number @dev. @mode and
1425 * @holder are identical to blkdev_get().
1427 * Use it ONLY if you really do not have anything better - i.e. when
1428 * you are behind a truly sucky interface and all you are given is a
1429 * device number. _Never_ to be used for internal purposes. If you
1430 * ever need it - reconsider your API.
1432 * On success, the returned block_device has reference count of one.
1438 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1440 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1442 struct block_device
*bdev
;
1447 return ERR_PTR(-ENOMEM
);
1449 err
= blkdev_get(bdev
, mode
, holder
);
1451 return ERR_PTR(err
);
1455 EXPORT_SYMBOL(blkdev_get_by_dev
);
1457 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1459 struct block_device
*bdev
;
1462 * Preserve backwards compatibility and allow large file access
1463 * even if userspace doesn't ask for it explicitly. Some mkfs
1464 * binary needs it. We might want to drop this workaround
1465 * during an unstable branch.
1467 filp
->f_flags
|= O_LARGEFILE
;
1469 if (filp
->f_flags
& O_NDELAY
)
1470 filp
->f_mode
|= FMODE_NDELAY
;
1471 if (filp
->f_flags
& O_EXCL
)
1472 filp
->f_mode
|= FMODE_EXCL
;
1473 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1474 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1476 bdev
= bd_acquire(inode
);
1480 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1482 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1485 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1487 struct gendisk
*disk
= bdev
->bd_disk
;
1488 struct block_device
*victim
= NULL
;
1490 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1492 bdev
->bd_part_count
--;
1494 if (!--bdev
->bd_openers
) {
1495 WARN_ON_ONCE(bdev
->bd_holders
);
1496 sync_blockdev(bdev
);
1499 * ->release can cause the queue to disappear, so flush all
1500 * dirty data before.
1502 bdev_write_inode(bdev
->bd_inode
);
1504 if (bdev
->bd_contains
== bdev
) {
1505 if (disk
->fops
->release
)
1506 disk
->fops
->release(disk
, mode
);
1508 if (!bdev
->bd_openers
) {
1509 struct module
*owner
= disk
->fops
->owner
;
1511 disk_put_part(bdev
->bd_part
);
1512 bdev
->bd_part
= NULL
;
1513 bdev
->bd_disk
= NULL
;
1514 if (bdev
!= bdev
->bd_contains
)
1515 victim
= bdev
->bd_contains
;
1516 bdev
->bd_contains
= NULL
;
1521 mutex_unlock(&bdev
->bd_mutex
);
1524 __blkdev_put(victim
, mode
, 1);
1527 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1529 mutex_lock(&bdev
->bd_mutex
);
1531 if (mode
& FMODE_EXCL
) {
1535 * Release a claim on the device. The holder fields
1536 * are protected with bdev_lock. bd_mutex is to
1537 * synchronize disk_holder unlinking.
1539 spin_lock(&bdev_lock
);
1541 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1542 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1544 /* bd_contains might point to self, check in a separate step */
1545 if ((bdev_free
= !bdev
->bd_holders
))
1546 bdev
->bd_holder
= NULL
;
1547 if (!bdev
->bd_contains
->bd_holders
)
1548 bdev
->bd_contains
->bd_holder
= NULL
;
1550 spin_unlock(&bdev_lock
);
1553 * If this was the last claim, remove holder link and
1554 * unblock evpoll if it was a write holder.
1556 if (bdev_free
&& bdev
->bd_write_holder
) {
1557 disk_unblock_events(bdev
->bd_disk
);
1558 bdev
->bd_write_holder
= false;
1563 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1564 * event. This is to ensure detection of media removal commanded
1565 * from userland - e.g. eject(1).
1567 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1569 mutex_unlock(&bdev
->bd_mutex
);
1571 __blkdev_put(bdev
, mode
, 0);
1573 EXPORT_SYMBOL(blkdev_put
);
1575 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1577 struct block_device
*bdev
= I_BDEV(filp
->f_mapping
->host
);
1578 blkdev_put(bdev
, filp
->f_mode
);
1582 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1584 struct block_device
*bdev
= I_BDEV(file
->f_mapping
->host
);
1585 fmode_t mode
= file
->f_mode
;
1588 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1589 * to updated it before every ioctl.
1591 if (file
->f_flags
& O_NDELAY
)
1592 mode
|= FMODE_NDELAY
;
1594 mode
&= ~FMODE_NDELAY
;
1596 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1600 * Write data to the block device. Only intended for the block device itself
1601 * and the raw driver which basically is a fake block device.
1603 * Does not take i_mutex for the write and thus is not for general purpose
1606 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1608 struct file
*file
= iocb
->ki_filp
;
1609 struct inode
*bd_inode
= file
->f_mapping
->host
;
1610 loff_t size
= i_size_read(bd_inode
);
1611 struct blk_plug plug
;
1614 if (bdev_read_only(I_BDEV(bd_inode
)))
1617 if (!iov_iter_count(from
))
1620 if (iocb
->ki_pos
>= size
)
1623 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1625 blk_start_plug(&plug
);
1626 ret
= __generic_file_write_iter(iocb
, from
);
1629 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1633 blk_finish_plug(&plug
);
1636 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1638 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1640 struct file
*file
= iocb
->ki_filp
;
1641 struct inode
*bd_inode
= file
->f_mapping
->host
;
1642 loff_t size
= i_size_read(bd_inode
);
1643 loff_t pos
= iocb
->ki_pos
;
1649 iov_iter_truncate(to
, size
);
1650 return generic_file_read_iter(iocb
, to
);
1652 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1655 * Try to release a page associated with block device when the system
1656 * is under memory pressure.
1658 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1660 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1662 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1663 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1665 return try_to_free_buffers(page
);
1668 static const struct address_space_operations def_blk_aops
= {
1669 .readpage
= blkdev_readpage
,
1670 .readpages
= blkdev_readpages
,
1671 .writepage
= blkdev_writepage
,
1672 .write_begin
= blkdev_write_begin
,
1673 .write_end
= blkdev_write_end
,
1674 .writepages
= generic_writepages
,
1675 .releasepage
= blkdev_releasepage
,
1676 .direct_IO
= blkdev_direct_IO
,
1677 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1680 const struct file_operations def_blk_fops
= {
1681 .open
= blkdev_open
,
1682 .release
= blkdev_close
,
1683 .llseek
= block_llseek
,
1684 .read_iter
= blkdev_read_iter
,
1685 .write_iter
= blkdev_write_iter
,
1686 .mmap
= generic_file_mmap
,
1687 .fsync
= blkdev_fsync
,
1688 .unlocked_ioctl
= block_ioctl
,
1689 #ifdef CONFIG_COMPAT
1690 .compat_ioctl
= compat_blkdev_ioctl
,
1692 .splice_read
= generic_file_splice_read
,
1693 .splice_write
= iter_file_splice_write
,
1696 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1699 mm_segment_t old_fs
= get_fs();
1701 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1706 EXPORT_SYMBOL(ioctl_by_bdev
);
1709 * lookup_bdev - lookup a struct block_device by name
1710 * @pathname: special file representing the block device
1712 * Get a reference to the blockdevice at @pathname in the current
1713 * namespace if possible and return it. Return ERR_PTR(error)
1716 struct block_device
*lookup_bdev(const char *pathname
)
1718 struct block_device
*bdev
;
1719 struct inode
*inode
;
1723 if (!pathname
|| !*pathname
)
1724 return ERR_PTR(-EINVAL
);
1726 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1728 return ERR_PTR(error
);
1730 inode
= d_backing_inode(path
.dentry
);
1732 if (!S_ISBLK(inode
->i_mode
))
1735 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1738 bdev
= bd_acquire(inode
);
1745 bdev
= ERR_PTR(error
);
1748 EXPORT_SYMBOL(lookup_bdev
);
1750 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1752 struct super_block
*sb
= get_super(bdev
);
1757 * no need to lock the super, get_super holds the
1758 * read mutex so the filesystem cannot go away
1759 * under us (->put_super runs with the write lock
1762 shrink_dcache_sb(sb
);
1763 res
= invalidate_inodes(sb
, kill_dirty
);
1766 invalidate_bdev(bdev
);
1769 EXPORT_SYMBOL(__invalidate_device
);
1771 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1773 struct inode
*inode
, *old_inode
= NULL
;
1775 spin_lock(&inode_sb_list_lock
);
1776 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1777 struct address_space
*mapping
= inode
->i_mapping
;
1779 spin_lock(&inode
->i_lock
);
1780 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1781 mapping
->nrpages
== 0) {
1782 spin_unlock(&inode
->i_lock
);
1786 spin_unlock(&inode
->i_lock
);
1787 spin_unlock(&inode_sb_list_lock
);
1789 * We hold a reference to 'inode' so it couldn't have been
1790 * removed from s_inodes list while we dropped the
1791 * inode_sb_list_lock. We cannot iput the inode now as we can
1792 * be holding the last reference and we cannot iput it under
1793 * inode_sb_list_lock. So we keep the reference and iput it
1799 func(I_BDEV(inode
), arg
);
1801 spin_lock(&inode_sb_list_lock
);
1803 spin_unlock(&inode_sb_list_lock
);