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 <asm/uaccess.h>
36 struct block_device bdev
;
37 struct inode vfs_inode
;
40 static const struct address_space_operations def_blk_aops
;
42 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
44 return container_of(inode
, struct bdev_inode
, vfs_inode
);
47 struct block_device
*I_BDEV(struct inode
*inode
)
49 return &BDEV_I(inode
)->bdev
;
51 EXPORT_SYMBOL(I_BDEV
);
53 void __vfs_msg(struct super_block
*sb
, const char *prefix
, const char *fmt
, ...)
61 printk_ratelimited("%sVFS (%s): %pV\n", prefix
, sb
->s_id
, &vaf
);
65 static void bdev_write_inode(struct block_device
*bdev
)
67 struct inode
*inode
= bdev
->bd_inode
;
70 spin_lock(&inode
->i_lock
);
71 while (inode
->i_state
& I_DIRTY
) {
72 spin_unlock(&inode
->i_lock
);
73 ret
= write_inode_now(inode
, true);
75 char name
[BDEVNAME_SIZE
];
76 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
77 "for block device %s (err=%d).\n",
78 bdevname(bdev
, name
), ret
);
80 spin_lock(&inode
->i_lock
);
82 spin_unlock(&inode
->i_lock
);
85 /* Kill _all_ buffers and pagecache , dirty or not.. */
86 void kill_bdev(struct block_device
*bdev
)
88 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
90 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
94 truncate_inode_pages(mapping
, 0);
96 EXPORT_SYMBOL(kill_bdev
);
98 /* Invalidate clean unused buffers and pagecache. */
99 void invalidate_bdev(struct block_device
*bdev
)
101 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
103 if (mapping
->nrpages
== 0)
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping
, 0, -1);
109 /* 99% of the time, we don't need to flush the cleancache on the bdev.
110 * But, for the strange corners, lets be cautious
112 cleancache_invalidate_inode(mapping
);
114 EXPORT_SYMBOL(invalidate_bdev
);
116 int set_blocksize(struct block_device
*bdev
, int size
)
118 /* Size must be a power of two, and between 512 and PAGE_SIZE */
119 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
122 /* Size cannot be smaller than the size supported by the device */
123 if (size
< bdev_logical_block_size(bdev
))
126 /* Don't change the size if it is same as current */
127 if (bdev
->bd_block_size
!= size
) {
129 bdev
->bd_block_size
= size
;
130 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
136 EXPORT_SYMBOL(set_blocksize
);
138 int sb_set_blocksize(struct super_block
*sb
, int size
)
140 if (set_blocksize(sb
->s_bdev
, size
))
142 /* If we get here, we know size is power of two
143 * and it's value is between 512 and PAGE_SIZE */
144 sb
->s_blocksize
= size
;
145 sb
->s_blocksize_bits
= blksize_bits(size
);
146 return sb
->s_blocksize
;
149 EXPORT_SYMBOL(sb_set_blocksize
);
151 int sb_min_blocksize(struct super_block
*sb
, int size
)
153 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
156 return sb_set_blocksize(sb
, size
);
159 EXPORT_SYMBOL(sb_min_blocksize
);
162 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
163 struct buffer_head
*bh
, int create
)
165 bh
->b_bdev
= I_BDEV(inode
);
166 bh
->b_blocknr
= iblock
;
167 set_buffer_mapped(bh
);
171 static struct inode
*bdev_file_inode(struct file
*file
)
173 return file
->f_mapping
->host
;
177 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
179 struct file
*file
= iocb
->ki_filp
;
180 struct inode
*inode
= bdev_file_inode(file
);
183 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
184 NULL
, DIO_SKIP_DIO_COUNT
);
185 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
186 blkdev_get_block
, NULL
, NULL
,
190 int __sync_blockdev(struct block_device
*bdev
, int wait
)
195 return filemap_flush(bdev
->bd_inode
->i_mapping
);
196 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
200 * Write out and wait upon all the dirty data associated with a block
201 * device via its mapping. Does not take the superblock lock.
203 int sync_blockdev(struct block_device
*bdev
)
205 return __sync_blockdev(bdev
, 1);
207 EXPORT_SYMBOL(sync_blockdev
);
210 * Write out and wait upon all dirty data associated with this
211 * device. Filesystem data as well as the underlying block
212 * device. Takes the superblock lock.
214 int fsync_bdev(struct block_device
*bdev
)
216 struct super_block
*sb
= get_super(bdev
);
218 int res
= sync_filesystem(sb
);
222 return sync_blockdev(bdev
);
224 EXPORT_SYMBOL(fsync_bdev
);
227 * freeze_bdev -- lock a filesystem and force it into a consistent state
228 * @bdev: blockdevice to lock
230 * If a superblock is found on this device, we take the s_umount semaphore
231 * on it to make sure nobody unmounts until the snapshot creation is done.
232 * The reference counter (bd_fsfreeze_count) guarantees that only the last
233 * unfreeze process can unfreeze the frozen filesystem actually when multiple
234 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
235 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
238 struct super_block
*freeze_bdev(struct block_device
*bdev
)
240 struct super_block
*sb
;
243 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
244 if (++bdev
->bd_fsfreeze_count
> 1) {
246 * We don't even need to grab a reference - the first call
247 * to freeze_bdev grab an active reference and only the last
248 * thaw_bdev drops it.
250 sb
= get_super(bdev
);
252 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
256 sb
= get_active_super(bdev
);
259 if (sb
->s_op
->freeze_super
)
260 error
= sb
->s_op
->freeze_super(sb
);
262 error
= freeze_super(sb
);
264 deactivate_super(sb
);
265 bdev
->bd_fsfreeze_count
--;
266 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
267 return ERR_PTR(error
);
269 deactivate_super(sb
);
272 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
273 return sb
; /* thaw_bdev releases s->s_umount */
275 EXPORT_SYMBOL(freeze_bdev
);
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
288 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
289 if (!bdev
->bd_fsfreeze_count
)
293 if (--bdev
->bd_fsfreeze_count
> 0)
299 if (sb
->s_op
->thaw_super
)
300 error
= sb
->s_op
->thaw_super(sb
);
302 error
= thaw_super(sb
);
304 bdev
->bd_fsfreeze_count
++;
305 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
309 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
312 EXPORT_SYMBOL(thaw_bdev
);
314 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
316 return block_write_full_page(page
, blkdev_get_block
, wbc
);
319 static int blkdev_readpage(struct file
* file
, struct page
* page
)
321 return block_read_full_page(page
, blkdev_get_block
);
324 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
325 struct list_head
*pages
, unsigned nr_pages
)
327 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
330 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
331 loff_t pos
, unsigned len
, unsigned flags
,
332 struct page
**pagep
, void **fsdata
)
334 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
338 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
339 loff_t pos
, unsigned len
, unsigned copied
,
340 struct page
*page
, void *fsdata
)
343 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
353 * for a block special file file_inode(file)->i_size is zero
354 * so we compute the size by hand (just as in block_read/write above)
356 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
358 struct inode
*bd_inode
= bdev_file_inode(file
);
361 inode_lock(bd_inode
);
362 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
363 inode_unlock(bd_inode
);
367 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
369 struct inode
*bd_inode
= bdev_file_inode(filp
);
370 struct block_device
*bdev
= I_BDEV(bd_inode
);
373 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
378 * There is no need to serialise calls to blkdev_issue_flush with
379 * i_mutex and doing so causes performance issues with concurrent
380 * O_SYNC writers to a block device.
382 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
383 if (error
== -EOPNOTSUPP
)
388 EXPORT_SYMBOL(blkdev_fsync
);
391 * bdev_read_page() - Start reading a page from a block device
392 * @bdev: The device to read the page from
393 * @sector: The offset on the device to read the page to (need not be aligned)
394 * @page: The page to read
396 * On entry, the page should be locked. It will be unlocked when the page
397 * has been read. If the block driver implements rw_page synchronously,
398 * that will be true on exit from this function, but it need not be.
400 * Errors returned by this function are usually "soft", eg out of memory, or
401 * queue full; callers should try a different route to read this page rather
402 * than propagate an error back up the stack.
404 * Return: negative errno if an error occurs, 0 if submission was successful.
406 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
409 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
410 int result
= -EOPNOTSUPP
;
412 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
415 result
= blk_queue_enter(bdev
->bd_queue
, false);
418 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
419 blk_queue_exit(bdev
->bd_queue
);
422 EXPORT_SYMBOL_GPL(bdev_read_page
);
425 * bdev_write_page() - Start writing a page to a block device
426 * @bdev: The device to write the page to
427 * @sector: The offset on the device to write the page to (need not be aligned)
428 * @page: The page to write
429 * @wbc: The writeback_control for the write
431 * On entry, the page should be locked and not currently under writeback.
432 * On exit, if the write started successfully, the page will be unlocked and
433 * under writeback. If the write failed already (eg the driver failed to
434 * queue the page to the device), the page will still be locked. If the
435 * caller is a ->writepage implementation, it will need to unlock the page.
437 * Errors returned by this function are usually "soft", eg out of memory, or
438 * queue full; callers should try a different route to write this page rather
439 * than propagate an error back up the stack.
441 * Return: negative errno if an error occurs, 0 if submission was successful.
443 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
444 struct page
*page
, struct writeback_control
*wbc
)
447 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
448 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
450 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
452 result
= blk_queue_enter(bdev
->bd_queue
, false);
456 set_page_writeback(page
);
457 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
459 end_page_writeback(page
);
462 blk_queue_exit(bdev
->bd_queue
);
465 EXPORT_SYMBOL_GPL(bdev_write_page
);
468 * bdev_direct_access() - Get the address for directly-accessibly memory
469 * @bdev: The device containing the memory
470 * @dax: control and output parameters for ->direct_access
472 * If a block device is made up of directly addressable memory, this function
473 * will tell the caller the PFN and the address of the memory. The address
474 * may be directly dereferenced within the kernel without the need to call
475 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
478 * Return: negative errno if an error occurs, otherwise the number of bytes
479 * accessible at this address.
481 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
483 sector_t sector
= dax
->sector
;
484 long avail
, size
= dax
->size
;
485 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
488 * The device driver is allowed to sleep, in order to make the
489 * memory directly accessible.
495 if (!ops
->direct_access
)
497 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
498 part_nr_sects_read(bdev
->bd_part
))
500 sector
+= get_start_sect(bdev
);
501 if (sector
% (PAGE_SIZE
/ 512))
503 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
);
506 if (avail
> 0 && avail
& ~PAGE_MASK
)
508 return min(avail
, size
);
510 EXPORT_SYMBOL_GPL(bdev_direct_access
);
516 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
517 static struct kmem_cache
* bdev_cachep __read_mostly
;
519 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
521 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
524 return &ei
->vfs_inode
;
527 static void bdev_i_callback(struct rcu_head
*head
)
529 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
530 struct bdev_inode
*bdi
= BDEV_I(inode
);
532 kmem_cache_free(bdev_cachep
, bdi
);
535 static void bdev_destroy_inode(struct inode
*inode
)
537 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
540 static void init_once(void *foo
)
542 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
543 struct block_device
*bdev
= &ei
->bdev
;
545 memset(bdev
, 0, sizeof(*bdev
));
546 mutex_init(&bdev
->bd_mutex
);
547 INIT_LIST_HEAD(&bdev
->bd_inodes
);
548 INIT_LIST_HEAD(&bdev
->bd_list
);
550 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
552 inode_init_once(&ei
->vfs_inode
);
553 /* Initialize mutex for freeze. */
554 mutex_init(&bdev
->bd_fsfreeze_mutex
);
557 static inline void __bd_forget(struct inode
*inode
)
559 list_del_init(&inode
->i_devices
);
560 inode
->i_bdev
= NULL
;
561 inode
->i_mapping
= &inode
->i_data
;
564 static void bdev_evict_inode(struct inode
*inode
)
566 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
568 truncate_inode_pages_final(&inode
->i_data
);
569 invalidate_inode_buffers(inode
); /* is it needed here? */
571 spin_lock(&bdev_lock
);
572 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
573 __bd_forget(list_entry(p
, struct inode
, i_devices
));
575 list_del_init(&bdev
->bd_list
);
576 spin_unlock(&bdev_lock
);
579 static const struct super_operations bdev_sops
= {
580 .statfs
= simple_statfs
,
581 .alloc_inode
= bdev_alloc_inode
,
582 .destroy_inode
= bdev_destroy_inode
,
583 .drop_inode
= generic_delete_inode
,
584 .evict_inode
= bdev_evict_inode
,
587 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
588 int flags
, const char *dev_name
, void *data
)
591 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
593 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
597 static struct file_system_type bd_type
= {
600 .kill_sb
= kill_anon_super
,
603 struct super_block
*blockdev_superblock __read_mostly
;
604 EXPORT_SYMBOL_GPL(blockdev_superblock
);
606 void __init
bdev_cache_init(void)
609 static struct vfsmount
*bd_mnt
;
611 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
612 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
613 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
615 err
= register_filesystem(&bd_type
);
617 panic("Cannot register bdev pseudo-fs");
618 bd_mnt
= kern_mount(&bd_type
);
620 panic("Cannot create bdev pseudo-fs");
621 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
625 * Most likely _very_ bad one - but then it's hardly critical for small
626 * /dev and can be fixed when somebody will need really large one.
627 * Keep in mind that it will be fed through icache hash function too.
629 static inline unsigned long hash(dev_t dev
)
631 return MAJOR(dev
)+MINOR(dev
);
634 static int bdev_test(struct inode
*inode
, void *data
)
636 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
639 static int bdev_set(struct inode
*inode
, void *data
)
641 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
645 static LIST_HEAD(all_bdevs
);
647 struct block_device
*bdget(dev_t dev
)
649 struct block_device
*bdev
;
652 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
653 bdev_test
, bdev_set
, &dev
);
658 bdev
= &BDEV_I(inode
)->bdev
;
660 if (inode
->i_state
& I_NEW
) {
661 bdev
->bd_contains
= NULL
;
662 bdev
->bd_super
= NULL
;
663 bdev
->bd_inode
= inode
;
664 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
665 bdev
->bd_part_count
= 0;
666 bdev
->bd_invalidated
= 0;
667 inode
->i_mode
= S_IFBLK
;
669 inode
->i_bdev
= bdev
;
670 inode
->i_data
.a_ops
= &def_blk_aops
;
671 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
672 spin_lock(&bdev_lock
);
673 list_add(&bdev
->bd_list
, &all_bdevs
);
674 spin_unlock(&bdev_lock
);
675 unlock_new_inode(inode
);
680 EXPORT_SYMBOL(bdget
);
683 * bdgrab -- Grab a reference to an already referenced block device
684 * @bdev: Block device to grab a reference to.
686 struct block_device
*bdgrab(struct block_device
*bdev
)
688 ihold(bdev
->bd_inode
);
691 EXPORT_SYMBOL(bdgrab
);
693 long nr_blockdev_pages(void)
695 struct block_device
*bdev
;
697 spin_lock(&bdev_lock
);
698 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
699 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
701 spin_unlock(&bdev_lock
);
705 void bdput(struct block_device
*bdev
)
707 iput(bdev
->bd_inode
);
710 EXPORT_SYMBOL(bdput
);
712 static struct block_device
*bd_acquire(struct inode
*inode
)
714 struct block_device
*bdev
;
716 spin_lock(&bdev_lock
);
717 bdev
= inode
->i_bdev
;
720 spin_unlock(&bdev_lock
);
723 spin_unlock(&bdev_lock
);
725 bdev
= bdget(inode
->i_rdev
);
727 spin_lock(&bdev_lock
);
728 if (!inode
->i_bdev
) {
730 * We take an additional reference to bd_inode,
731 * and it's released in clear_inode() of inode.
732 * So, we can access it via ->i_mapping always
736 inode
->i_bdev
= bdev
;
737 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
738 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
740 spin_unlock(&bdev_lock
);
745 /* Call when you free inode */
747 void bd_forget(struct inode
*inode
)
749 struct block_device
*bdev
= NULL
;
751 spin_lock(&bdev_lock
);
752 if (!sb_is_blkdev_sb(inode
->i_sb
))
753 bdev
= inode
->i_bdev
;
755 spin_unlock(&bdev_lock
);
762 * bd_may_claim - test whether a block device can be claimed
763 * @bdev: block device of interest
764 * @whole: whole block device containing @bdev, may equal @bdev
765 * @holder: holder trying to claim @bdev
767 * Test whether @bdev can be claimed by @holder.
770 * spin_lock(&bdev_lock).
773 * %true if @bdev can be claimed, %false otherwise.
775 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
778 if (bdev
->bd_holder
== holder
)
779 return true; /* already a holder */
780 else if (bdev
->bd_holder
!= NULL
)
781 return false; /* held by someone else */
782 else if (bdev
->bd_contains
== bdev
)
783 return true; /* is a whole device which isn't held */
785 else if (whole
->bd_holder
== bd_may_claim
)
786 return true; /* is a partition of a device that is being partitioned */
787 else if (whole
->bd_holder
!= NULL
)
788 return false; /* is a partition of a held device */
790 return true; /* is a partition of an un-held device */
794 * bd_prepare_to_claim - prepare to claim a block device
795 * @bdev: block device of interest
796 * @whole: the whole device containing @bdev, may equal @bdev
797 * @holder: holder trying to claim @bdev
799 * Prepare to claim @bdev. This function fails if @bdev is already
800 * claimed by another holder and waits if another claiming is in
801 * progress. This function doesn't actually claim. On successful
802 * return, the caller has ownership of bd_claiming and bd_holder[s].
805 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
809 * 0 if @bdev can be claimed, -EBUSY otherwise.
811 static int bd_prepare_to_claim(struct block_device
*bdev
,
812 struct block_device
*whole
, void *holder
)
815 /* if someone else claimed, fail */
816 if (!bd_may_claim(bdev
, whole
, holder
))
819 /* if claiming is already in progress, wait for it to finish */
820 if (whole
->bd_claiming
) {
821 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
824 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
825 spin_unlock(&bdev_lock
);
827 finish_wait(wq
, &wait
);
828 spin_lock(&bdev_lock
);
837 * bd_start_claiming - start claiming a block device
838 * @bdev: block device of interest
839 * @holder: holder trying to claim @bdev
841 * @bdev is about to be opened exclusively. Check @bdev can be opened
842 * exclusively and mark that an exclusive open is in progress. Each
843 * successful call to this function must be matched with a call to
844 * either bd_finish_claiming() or bd_abort_claiming() (which do not
847 * This function is used to gain exclusive access to the block device
848 * without actually causing other exclusive open attempts to fail. It
849 * should be used when the open sequence itself requires exclusive
850 * access but may subsequently fail.
856 * Pointer to the block device containing @bdev on success, ERR_PTR()
859 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
862 struct gendisk
*disk
;
863 struct block_device
*whole
;
869 * @bdev might not have been initialized properly yet, look up
870 * and grab the outer block device the hard way.
872 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
874 return ERR_PTR(-ENXIO
);
877 * Normally, @bdev should equal what's returned from bdget_disk()
878 * if partno is 0; however, some drivers (floppy) use multiple
879 * bdev's for the same physical device and @bdev may be one of the
880 * aliases. Keep @bdev if partno is 0. This means claimer
881 * tracking is broken for those devices but it has always been that
885 whole
= bdget_disk(disk
, 0);
887 whole
= bdgrab(bdev
);
889 module_put(disk
->fops
->owner
);
892 return ERR_PTR(-ENOMEM
);
894 /* prepare to claim, if successful, mark claiming in progress */
895 spin_lock(&bdev_lock
);
897 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
899 whole
->bd_claiming
= holder
;
900 spin_unlock(&bdev_lock
);
903 spin_unlock(&bdev_lock
);
910 struct bd_holder_disk
{
911 struct list_head list
;
912 struct gendisk
*disk
;
916 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
917 struct gendisk
*disk
)
919 struct bd_holder_disk
*holder
;
921 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
922 if (holder
->disk
== disk
)
927 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
929 return sysfs_create_link(from
, to
, kobject_name(to
));
932 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
934 sysfs_remove_link(from
, kobject_name(to
));
938 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
939 * @bdev: the claimed slave bdev
940 * @disk: the holding disk
942 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
944 * This functions creates the following sysfs symlinks.
946 * - from "slaves" directory of the holder @disk to the claimed @bdev
947 * - from "holders" directory of the @bdev to the holder @disk
949 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
950 * passed to bd_link_disk_holder(), then:
952 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
953 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
955 * The caller must have claimed @bdev before calling this function and
956 * ensure that both @bdev and @disk are valid during the creation and
957 * lifetime of these symlinks.
963 * 0 on success, -errno on failure.
965 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
967 struct bd_holder_disk
*holder
;
970 mutex_lock(&bdev
->bd_mutex
);
972 WARN_ON_ONCE(!bdev
->bd_holder
);
974 /* FIXME: remove the following once add_disk() handles errors */
975 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
978 holder
= bd_find_holder_disk(bdev
, disk
);
984 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
990 INIT_LIST_HEAD(&holder
->list
);
994 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
998 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1002 * bdev could be deleted beneath us which would implicitly destroy
1003 * the holder directory. Hold on to it.
1005 kobject_get(bdev
->bd_part
->holder_dir
);
1007 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1011 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1015 mutex_unlock(&bdev
->bd_mutex
);
1018 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1021 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1022 * @bdev: the calimed slave bdev
1023 * @disk: the holding disk
1025 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1030 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1032 struct bd_holder_disk
*holder
;
1034 mutex_lock(&bdev
->bd_mutex
);
1036 holder
= bd_find_holder_disk(bdev
, disk
);
1038 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1039 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1040 del_symlink(bdev
->bd_part
->holder_dir
,
1041 &disk_to_dev(disk
)->kobj
);
1042 kobject_put(bdev
->bd_part
->holder_dir
);
1043 list_del_init(&holder
->list
);
1047 mutex_unlock(&bdev
->bd_mutex
);
1049 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1053 * flush_disk - invalidates all buffer-cache entries on a disk
1055 * @bdev: struct block device to be flushed
1056 * @kill_dirty: flag to guide handling of dirty inodes
1058 * Invalidates all buffer-cache entries on a disk. It should be called
1059 * when a disk has been changed -- either by a media change or online
1062 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1064 if (__invalidate_device(bdev
, kill_dirty
)) {
1065 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1066 "resized disk %s\n",
1067 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1072 if (disk_part_scan_enabled(bdev
->bd_disk
))
1073 bdev
->bd_invalidated
= 1;
1077 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1078 * @disk: struct gendisk to check
1079 * @bdev: struct bdev to adjust.
1081 * This routine checks to see if the bdev size does not match the disk size
1082 * and adjusts it if it differs.
1084 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1086 loff_t disk_size
, bdev_size
;
1088 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1089 bdev_size
= i_size_read(bdev
->bd_inode
);
1090 if (disk_size
!= bdev_size
) {
1092 "%s: detected capacity change from %lld to %lld\n",
1093 disk
->disk_name
, bdev_size
, disk_size
);
1094 i_size_write(bdev
->bd_inode
, disk_size
);
1095 flush_disk(bdev
, false);
1098 EXPORT_SYMBOL(check_disk_size_change
);
1101 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1102 * @disk: struct gendisk to be revalidated
1104 * This routine is a wrapper for lower-level driver's revalidate_disk
1105 * call-backs. It is used to do common pre and post operations needed
1106 * for all revalidate_disk operations.
1108 int revalidate_disk(struct gendisk
*disk
)
1110 struct block_device
*bdev
;
1113 if (disk
->fops
->revalidate_disk
)
1114 ret
= disk
->fops
->revalidate_disk(disk
);
1115 blk_integrity_revalidate(disk
);
1116 bdev
= bdget_disk(disk
, 0);
1120 mutex_lock(&bdev
->bd_mutex
);
1121 check_disk_size_change(disk
, bdev
);
1122 bdev
->bd_invalidated
= 0;
1123 mutex_unlock(&bdev
->bd_mutex
);
1127 EXPORT_SYMBOL(revalidate_disk
);
1130 * This routine checks whether a removable media has been changed,
1131 * and invalidates all buffer-cache-entries in that case. This
1132 * is a relatively slow routine, so we have to try to minimize using
1133 * it. Thus it is called only upon a 'mount' or 'open'. This
1134 * is the best way of combining speed and utility, I think.
1135 * People changing diskettes in the middle of an operation deserve
1138 int check_disk_change(struct block_device
*bdev
)
1140 struct gendisk
*disk
= bdev
->bd_disk
;
1141 const struct block_device_operations
*bdops
= disk
->fops
;
1142 unsigned int events
;
1144 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1145 DISK_EVENT_EJECT_REQUEST
);
1146 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1149 flush_disk(bdev
, true);
1150 if (bdops
->revalidate_disk
)
1151 bdops
->revalidate_disk(bdev
->bd_disk
);
1155 EXPORT_SYMBOL(check_disk_change
);
1157 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1159 unsigned bsize
= bdev_logical_block_size(bdev
);
1161 inode_lock(bdev
->bd_inode
);
1162 i_size_write(bdev
->bd_inode
, size
);
1163 inode_unlock(bdev
->bd_inode
);
1164 while (bsize
< PAGE_SIZE
) {
1169 bdev
->bd_block_size
= bsize
;
1170 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1172 EXPORT_SYMBOL(bd_set_size
);
1174 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1179 * mutex_lock(part->bd_mutex)
1180 * mutex_lock_nested(whole->bd_mutex, 1)
1183 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1185 struct gendisk
*disk
;
1186 struct module
*owner
;
1191 if (mode
& FMODE_READ
)
1193 if (mode
& FMODE_WRITE
)
1196 * hooks: /n/, see "layering violations".
1199 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1209 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1212 owner
= disk
->fops
->owner
;
1214 disk_block_events(disk
);
1215 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1216 if (!bdev
->bd_openers
) {
1217 bdev
->bd_disk
= disk
;
1218 bdev
->bd_queue
= disk
->queue
;
1219 bdev
->bd_contains
= bdev
;
1220 if (IS_ENABLED(CONFIG_BLK_DEV_DAX
) && disk
->fops
->direct_access
)
1221 bdev
->bd_inode
->i_flags
= S_DAX
;
1223 bdev
->bd_inode
->i_flags
= 0;
1227 bdev
->bd_part
= disk_get_part(disk
, partno
);
1232 if (disk
->fops
->open
) {
1233 ret
= disk
->fops
->open(bdev
, mode
);
1234 if (ret
== -ERESTARTSYS
) {
1235 /* Lost a race with 'disk' being
1236 * deleted, try again.
1239 disk_put_part(bdev
->bd_part
);
1240 bdev
->bd_part
= NULL
;
1241 bdev
->bd_disk
= NULL
;
1242 bdev
->bd_queue
= NULL
;
1243 mutex_unlock(&bdev
->bd_mutex
);
1244 disk_unblock_events(disk
);
1252 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1253 if (!blkdev_dax_capable(bdev
))
1254 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1258 * If the device is invalidated, rescan partition
1259 * if open succeeded or failed with -ENOMEDIUM.
1260 * The latter is necessary to prevent ghost
1261 * partitions on a removed medium.
1263 if (bdev
->bd_invalidated
) {
1265 rescan_partitions(disk
, bdev
);
1266 else if (ret
== -ENOMEDIUM
)
1267 invalidate_partitions(disk
, bdev
);
1273 struct block_device
*whole
;
1274 whole
= bdget_disk(disk
, 0);
1279 ret
= __blkdev_get(whole
, mode
, 1);
1282 bdev
->bd_contains
= whole
;
1283 bdev
->bd_part
= disk_get_part(disk
, partno
);
1284 if (!(disk
->flags
& GENHD_FL_UP
) ||
1285 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1289 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1290 if (!blkdev_dax_capable(bdev
))
1291 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1294 if (bdev
->bd_contains
== bdev
) {
1296 if (bdev
->bd_disk
->fops
->open
)
1297 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1298 /* the same as first opener case, read comment there */
1299 if (bdev
->bd_invalidated
) {
1301 rescan_partitions(bdev
->bd_disk
, bdev
);
1302 else if (ret
== -ENOMEDIUM
)
1303 invalidate_partitions(bdev
->bd_disk
, bdev
);
1306 goto out_unlock_bdev
;
1308 /* only one opener holds refs to the module and disk */
1314 bdev
->bd_part_count
++;
1315 mutex_unlock(&bdev
->bd_mutex
);
1316 disk_unblock_events(disk
);
1320 disk_put_part(bdev
->bd_part
);
1321 bdev
->bd_disk
= NULL
;
1322 bdev
->bd_part
= NULL
;
1323 bdev
->bd_queue
= NULL
;
1324 if (bdev
!= bdev
->bd_contains
)
1325 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1326 bdev
->bd_contains
= NULL
;
1328 mutex_unlock(&bdev
->bd_mutex
);
1329 disk_unblock_events(disk
);
1339 * blkdev_get - open a block device
1340 * @bdev: block_device to open
1341 * @mode: FMODE_* mask
1342 * @holder: exclusive holder identifier
1344 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1345 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1346 * @holder is invalid. Exclusive opens may nest for the same @holder.
1348 * On success, the reference count of @bdev is unchanged. On failure,
1355 * 0 on success, -errno on failure.
1357 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1359 struct block_device
*whole
= NULL
;
1362 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1364 if ((mode
& FMODE_EXCL
) && holder
) {
1365 whole
= bd_start_claiming(bdev
, holder
);
1366 if (IS_ERR(whole
)) {
1368 return PTR_ERR(whole
);
1372 res
= __blkdev_get(bdev
, mode
, 0);
1375 struct gendisk
*disk
= whole
->bd_disk
;
1377 /* finish claiming */
1378 mutex_lock(&bdev
->bd_mutex
);
1379 spin_lock(&bdev_lock
);
1382 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1384 * Note that for a whole device bd_holders
1385 * will be incremented twice, and bd_holder
1386 * will be set to bd_may_claim before being
1389 whole
->bd_holders
++;
1390 whole
->bd_holder
= bd_may_claim
;
1392 bdev
->bd_holder
= holder
;
1395 /* tell others that we're done */
1396 BUG_ON(whole
->bd_claiming
!= holder
);
1397 whole
->bd_claiming
= NULL
;
1398 wake_up_bit(&whole
->bd_claiming
, 0);
1400 spin_unlock(&bdev_lock
);
1403 * Block event polling for write claims if requested. Any
1404 * write holder makes the write_holder state stick until
1405 * all are released. This is good enough and tracking
1406 * individual writeable reference is too fragile given the
1407 * way @mode is used in blkdev_get/put().
1409 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1410 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1411 bdev
->bd_write_holder
= true;
1412 disk_block_events(disk
);
1415 mutex_unlock(&bdev
->bd_mutex
);
1421 EXPORT_SYMBOL(blkdev_get
);
1424 * blkdev_get_by_path - open a block device by name
1425 * @path: path to the block device to open
1426 * @mode: FMODE_* mask
1427 * @holder: exclusive holder identifier
1429 * Open the blockdevice described by the device file at @path. @mode
1430 * and @holder are identical to blkdev_get().
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_path(const char *path
, fmode_t mode
,
1443 struct block_device
*bdev
;
1446 bdev
= lookup_bdev(path
);
1450 err
= blkdev_get(bdev
, mode
, holder
);
1452 return ERR_PTR(err
);
1454 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1455 blkdev_put(bdev
, mode
);
1456 return ERR_PTR(-EACCES
);
1461 EXPORT_SYMBOL(blkdev_get_by_path
);
1464 * blkdev_get_by_dev - open a block device by device number
1465 * @dev: device number of block device to open
1466 * @mode: FMODE_* mask
1467 * @holder: exclusive holder identifier
1469 * Open the blockdevice described by device number @dev. @mode and
1470 * @holder are identical to blkdev_get().
1472 * Use it ONLY if you really do not have anything better - i.e. when
1473 * you are behind a truly sucky interface and all you are given is a
1474 * device number. _Never_ to be used for internal purposes. If you
1475 * ever need it - reconsider your API.
1477 * On success, the returned block_device has reference count of one.
1483 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1485 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1487 struct block_device
*bdev
;
1492 return ERR_PTR(-ENOMEM
);
1494 err
= blkdev_get(bdev
, mode
, holder
);
1496 return ERR_PTR(err
);
1500 EXPORT_SYMBOL(blkdev_get_by_dev
);
1502 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1504 struct block_device
*bdev
;
1507 * Preserve backwards compatibility and allow large file access
1508 * even if userspace doesn't ask for it explicitly. Some mkfs
1509 * binary needs it. We might want to drop this workaround
1510 * during an unstable branch.
1512 filp
->f_flags
|= O_LARGEFILE
;
1514 if (filp
->f_flags
& O_NDELAY
)
1515 filp
->f_mode
|= FMODE_NDELAY
;
1516 if (filp
->f_flags
& O_EXCL
)
1517 filp
->f_mode
|= FMODE_EXCL
;
1518 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1519 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1521 bdev
= bd_acquire(inode
);
1525 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1527 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1530 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1532 struct gendisk
*disk
= bdev
->bd_disk
;
1533 struct block_device
*victim
= NULL
;
1535 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1537 bdev
->bd_part_count
--;
1539 if (!--bdev
->bd_openers
) {
1540 WARN_ON_ONCE(bdev
->bd_holders
);
1541 sync_blockdev(bdev
);
1544 bdev_write_inode(bdev
);
1546 * Detaching bdev inode from its wb in __destroy_inode()
1547 * is too late: the queue which embeds its bdi (along with
1548 * root wb) can be gone as soon as we put_disk() below.
1550 inode_detach_wb(bdev
->bd_inode
);
1552 if (bdev
->bd_contains
== bdev
) {
1553 if (disk
->fops
->release
)
1554 disk
->fops
->release(disk
, mode
);
1556 if (!bdev
->bd_openers
) {
1557 struct module
*owner
= disk
->fops
->owner
;
1559 disk_put_part(bdev
->bd_part
);
1560 bdev
->bd_part
= NULL
;
1561 bdev
->bd_disk
= NULL
;
1562 if (bdev
!= bdev
->bd_contains
)
1563 victim
= bdev
->bd_contains
;
1564 bdev
->bd_contains
= NULL
;
1569 mutex_unlock(&bdev
->bd_mutex
);
1572 __blkdev_put(victim
, mode
, 1);
1575 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1577 mutex_lock(&bdev
->bd_mutex
);
1579 if (mode
& FMODE_EXCL
) {
1583 * Release a claim on the device. The holder fields
1584 * are protected with bdev_lock. bd_mutex is to
1585 * synchronize disk_holder unlinking.
1587 spin_lock(&bdev_lock
);
1589 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1590 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1592 /* bd_contains might point to self, check in a separate step */
1593 if ((bdev_free
= !bdev
->bd_holders
))
1594 bdev
->bd_holder
= NULL
;
1595 if (!bdev
->bd_contains
->bd_holders
)
1596 bdev
->bd_contains
->bd_holder
= NULL
;
1598 spin_unlock(&bdev_lock
);
1601 * If this was the last claim, remove holder link and
1602 * unblock evpoll if it was a write holder.
1604 if (bdev_free
&& bdev
->bd_write_holder
) {
1605 disk_unblock_events(bdev
->bd_disk
);
1606 bdev
->bd_write_holder
= false;
1611 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1612 * event. This is to ensure detection of media removal commanded
1613 * from userland - e.g. eject(1).
1615 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1617 mutex_unlock(&bdev
->bd_mutex
);
1619 __blkdev_put(bdev
, mode
, 0);
1621 EXPORT_SYMBOL(blkdev_put
);
1623 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1625 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1626 blkdev_put(bdev
, filp
->f_mode
);
1630 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1632 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1633 fmode_t mode
= file
->f_mode
;
1636 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1637 * to updated it before every ioctl.
1639 if (file
->f_flags
& O_NDELAY
)
1640 mode
|= FMODE_NDELAY
;
1642 mode
&= ~FMODE_NDELAY
;
1644 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1648 * Write data to the block device. Only intended for the block device itself
1649 * and the raw driver which basically is a fake block device.
1651 * Does not take i_mutex for the write and thus is not for general purpose
1654 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1656 struct file
*file
= iocb
->ki_filp
;
1657 struct inode
*bd_inode
= bdev_file_inode(file
);
1658 loff_t size
= i_size_read(bd_inode
);
1659 struct blk_plug plug
;
1662 if (bdev_read_only(I_BDEV(bd_inode
)))
1665 if (!iov_iter_count(from
))
1668 if (iocb
->ki_pos
>= size
)
1671 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1673 blk_start_plug(&plug
);
1674 ret
= __generic_file_write_iter(iocb
, from
);
1677 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1681 blk_finish_plug(&plug
);
1684 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1686 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1688 struct file
*file
= iocb
->ki_filp
;
1689 struct inode
*bd_inode
= bdev_file_inode(file
);
1690 loff_t size
= i_size_read(bd_inode
);
1691 loff_t pos
= iocb
->ki_pos
;
1697 iov_iter_truncate(to
, size
);
1698 return generic_file_read_iter(iocb
, to
);
1700 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1703 * Try to release a page associated with block device when the system
1704 * is under memory pressure.
1706 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1708 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1710 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1711 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1713 return try_to_free_buffers(page
);
1716 static int blkdev_writepages(struct address_space
*mapping
,
1717 struct writeback_control
*wbc
)
1719 if (dax_mapping(mapping
)) {
1720 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1722 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1724 return generic_writepages(mapping
, wbc
);
1727 static const struct address_space_operations def_blk_aops
= {
1728 .readpage
= blkdev_readpage
,
1729 .readpages
= blkdev_readpages
,
1730 .writepage
= blkdev_writepage
,
1731 .write_begin
= blkdev_write_begin
,
1732 .write_end
= blkdev_write_end
,
1733 .writepages
= blkdev_writepages
,
1734 .releasepage
= blkdev_releasepage
,
1735 .direct_IO
= blkdev_direct_IO
,
1736 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1739 #ifdef CONFIG_FS_DAX
1741 * In the raw block case we do not need to contend with truncation nor
1742 * unwritten file extents. Without those concerns there is no need for
1743 * additional locking beyond the mmap_sem context that these routines
1744 * are already executing under.
1746 * Note, there is no protection if the block device is dynamically
1747 * resized (partition grow/shrink) during a fault. A stable block device
1748 * size is already not enforced in the blkdev_direct_IO path.
1750 * For DAX, it is the responsibility of the block device driver to
1751 * ensure the whole-disk device size is stable while requests are in
1754 * Finally, unlike the filemap_page_mkwrite() case there is no
1755 * filesystem superblock to sync against freezing. We still include a
1756 * pfn_mkwrite callback for dax drivers to receive write fault
1759 static int blkdev_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1761 return __dax_fault(vma
, vmf
, blkdev_get_block
);
1764 static int blkdev_dax_pfn_mkwrite(struct vm_area_struct
*vma
,
1765 struct vm_fault
*vmf
)
1767 return dax_pfn_mkwrite(vma
, vmf
);
1770 static int blkdev_dax_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
1771 pmd_t
*pmd
, unsigned int flags
)
1773 return __dax_pmd_fault(vma
, addr
, pmd
, flags
, blkdev_get_block
);
1776 static const struct vm_operations_struct blkdev_dax_vm_ops
= {
1777 .fault
= blkdev_dax_fault
,
1778 .pmd_fault
= blkdev_dax_pmd_fault
,
1779 .pfn_mkwrite
= blkdev_dax_pfn_mkwrite
,
1782 static const struct vm_operations_struct blkdev_default_vm_ops
= {
1783 .fault
= filemap_fault
,
1784 .map_pages
= filemap_map_pages
,
1787 static int blkdev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1789 struct inode
*bd_inode
= bdev_file_inode(file
);
1791 file_accessed(file
);
1792 if (IS_DAX(bd_inode
)) {
1793 vma
->vm_ops
= &blkdev_dax_vm_ops
;
1794 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1796 vma
->vm_ops
= &blkdev_default_vm_ops
;
1802 #define blkdev_mmap generic_file_mmap
1805 const struct file_operations def_blk_fops
= {
1806 .open
= blkdev_open
,
1807 .release
= blkdev_close
,
1808 .llseek
= block_llseek
,
1809 .read_iter
= blkdev_read_iter
,
1810 .write_iter
= blkdev_write_iter
,
1811 .mmap
= blkdev_mmap
,
1812 .fsync
= blkdev_fsync
,
1813 .unlocked_ioctl
= block_ioctl
,
1814 #ifdef CONFIG_COMPAT
1815 .compat_ioctl
= compat_blkdev_ioctl
,
1817 .splice_read
= generic_file_splice_read
,
1818 .splice_write
= iter_file_splice_write
,
1821 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1824 mm_segment_t old_fs
= get_fs();
1826 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1831 EXPORT_SYMBOL(ioctl_by_bdev
);
1834 * lookup_bdev - lookup a struct block_device by name
1835 * @pathname: special file representing the block device
1837 * Get a reference to the blockdevice at @pathname in the current
1838 * namespace if possible and return it. Return ERR_PTR(error)
1841 struct block_device
*lookup_bdev(const char *pathname
)
1843 struct block_device
*bdev
;
1844 struct inode
*inode
;
1848 if (!pathname
|| !*pathname
)
1849 return ERR_PTR(-EINVAL
);
1851 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1853 return ERR_PTR(error
);
1855 inode
= d_backing_inode(path
.dentry
);
1857 if (!S_ISBLK(inode
->i_mode
))
1860 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1863 bdev
= bd_acquire(inode
);
1870 bdev
= ERR_PTR(error
);
1873 EXPORT_SYMBOL(lookup_bdev
);
1875 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1877 struct super_block
*sb
= get_super(bdev
);
1882 * no need to lock the super, get_super holds the
1883 * read mutex so the filesystem cannot go away
1884 * under us (->put_super runs with the write lock
1887 shrink_dcache_sb(sb
);
1888 res
= invalidate_inodes(sb
, kill_dirty
);
1891 invalidate_bdev(bdev
);
1894 EXPORT_SYMBOL(__invalidate_device
);
1896 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1898 struct inode
*inode
, *old_inode
= NULL
;
1900 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1901 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1902 struct address_space
*mapping
= inode
->i_mapping
;
1904 spin_lock(&inode
->i_lock
);
1905 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1906 mapping
->nrpages
== 0) {
1907 spin_unlock(&inode
->i_lock
);
1911 spin_unlock(&inode
->i_lock
);
1912 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1914 * We hold a reference to 'inode' so it couldn't have been
1915 * removed from s_inodes list while we dropped the
1916 * s_inode_list_lock We cannot iput the inode now as we can
1917 * be holding the last reference and we cannot iput it under
1918 * s_inode_list_lock. So we keep the reference and iput it
1924 func(I_BDEV(inode
), arg
);
1926 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1928 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);