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 static void bdev_write_inode(struct block_device
*bdev
)
55 struct inode
*inode
= bdev
->bd_inode
;
58 spin_lock(&inode
->i_lock
);
59 while (inode
->i_state
& I_DIRTY
) {
60 spin_unlock(&inode
->i_lock
);
61 ret
= write_inode_now(inode
, true);
63 char name
[BDEVNAME_SIZE
];
64 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
65 "for block device %s (err=%d).\n",
66 bdevname(bdev
, name
), ret
);
68 spin_lock(&inode
->i_lock
);
70 spin_unlock(&inode
->i_lock
);
73 /* Kill _all_ buffers and pagecache , dirty or not.. */
74 void kill_bdev(struct block_device
*bdev
)
76 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
78 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
82 truncate_inode_pages(mapping
, 0);
84 EXPORT_SYMBOL(kill_bdev
);
86 /* Invalidate clean unused buffers and pagecache. */
87 void invalidate_bdev(struct block_device
*bdev
)
89 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
91 if (mapping
->nrpages
== 0)
95 lru_add_drain_all(); /* make sure all lru add caches are flushed */
96 invalidate_mapping_pages(mapping
, 0, -1);
97 /* 99% of the time, we don't need to flush the cleancache on the bdev.
98 * But, for the strange corners, lets be cautious
100 cleancache_invalidate_inode(mapping
);
102 EXPORT_SYMBOL(invalidate_bdev
);
104 int set_blocksize(struct block_device
*bdev
, int size
)
106 /* Size must be a power of two, and between 512 and PAGE_SIZE */
107 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
110 /* Size cannot be smaller than the size supported by the device */
111 if (size
< bdev_logical_block_size(bdev
))
114 /* Don't change the size if it is same as current */
115 if (bdev
->bd_block_size
!= size
) {
117 bdev
->bd_block_size
= size
;
118 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
124 EXPORT_SYMBOL(set_blocksize
);
126 int sb_set_blocksize(struct super_block
*sb
, int size
)
128 if (set_blocksize(sb
->s_bdev
, size
))
130 /* If we get here, we know size is power of two
131 * and it's value is between 512 and PAGE_SIZE */
132 sb
->s_blocksize
= size
;
133 sb
->s_blocksize_bits
= blksize_bits(size
);
134 return sb
->s_blocksize
;
137 EXPORT_SYMBOL(sb_set_blocksize
);
139 int sb_min_blocksize(struct super_block
*sb
, int size
)
141 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
144 return sb_set_blocksize(sb
, size
);
147 EXPORT_SYMBOL(sb_min_blocksize
);
150 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
151 struct buffer_head
*bh
, int create
)
153 bh
->b_bdev
= I_BDEV(inode
);
154 bh
->b_blocknr
= iblock
;
155 set_buffer_mapped(bh
);
159 static struct inode
*bdev_file_inode(struct file
*file
)
161 return file
->f_mapping
->host
;
165 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, loff_t offset
)
167 struct file
*file
= iocb
->ki_filp
;
168 struct inode
*inode
= bdev_file_inode(file
);
171 return dax_do_io(iocb
, inode
, iter
, offset
, blkdev_get_block
,
172 NULL
, DIO_SKIP_DIO_COUNT
);
173 return __blockdev_direct_IO(iocb
, inode
, I_BDEV(inode
), iter
, offset
,
174 blkdev_get_block
, NULL
, NULL
,
178 int __sync_blockdev(struct block_device
*bdev
, int wait
)
183 return filemap_flush(bdev
->bd_inode
->i_mapping
);
184 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
188 * Write out and wait upon all the dirty data associated with a block
189 * device via its mapping. Does not take the superblock lock.
191 int sync_blockdev(struct block_device
*bdev
)
193 return __sync_blockdev(bdev
, 1);
195 EXPORT_SYMBOL(sync_blockdev
);
198 * Write out and wait upon all dirty data associated with this
199 * device. Filesystem data as well as the underlying block
200 * device. Takes the superblock lock.
202 int fsync_bdev(struct block_device
*bdev
)
204 struct super_block
*sb
= get_super(bdev
);
206 int res
= sync_filesystem(sb
);
210 return sync_blockdev(bdev
);
212 EXPORT_SYMBOL(fsync_bdev
);
215 * freeze_bdev -- lock a filesystem and force it into a consistent state
216 * @bdev: blockdevice to lock
218 * If a superblock is found on this device, we take the s_umount semaphore
219 * on it to make sure nobody unmounts until the snapshot creation is done.
220 * The reference counter (bd_fsfreeze_count) guarantees that only the last
221 * unfreeze process can unfreeze the frozen filesystem actually when multiple
222 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
226 struct super_block
*freeze_bdev(struct block_device
*bdev
)
228 struct super_block
*sb
;
231 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
232 if (++bdev
->bd_fsfreeze_count
> 1) {
234 * We don't even need to grab a reference - the first call
235 * to freeze_bdev grab an active reference and only the last
236 * thaw_bdev drops it.
238 sb
= get_super(bdev
);
240 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
244 sb
= get_active_super(bdev
);
247 if (sb
->s_op
->freeze_super
)
248 error
= sb
->s_op
->freeze_super(sb
);
250 error
= freeze_super(sb
);
252 deactivate_super(sb
);
253 bdev
->bd_fsfreeze_count
--;
254 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
255 return ERR_PTR(error
);
257 deactivate_super(sb
);
260 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
261 return sb
; /* thaw_bdev releases s->s_umount */
263 EXPORT_SYMBOL(freeze_bdev
);
266 * thaw_bdev -- unlock filesystem
267 * @bdev: blockdevice to unlock
268 * @sb: associated superblock
270 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
272 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
276 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
277 if (!bdev
->bd_fsfreeze_count
)
281 if (--bdev
->bd_fsfreeze_count
> 0)
287 if (sb
->s_op
->thaw_super
)
288 error
= sb
->s_op
->thaw_super(sb
);
290 error
= thaw_super(sb
);
292 bdev
->bd_fsfreeze_count
++;
293 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
297 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
300 EXPORT_SYMBOL(thaw_bdev
);
302 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
304 return block_write_full_page(page
, blkdev_get_block
, wbc
);
307 static int blkdev_readpage(struct file
* file
, struct page
* page
)
309 return block_read_full_page(page
, blkdev_get_block
);
312 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
313 struct list_head
*pages
, unsigned nr_pages
)
315 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
318 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
319 loff_t pos
, unsigned len
, unsigned flags
,
320 struct page
**pagep
, void **fsdata
)
322 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
326 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
327 loff_t pos
, unsigned len
, unsigned copied
,
328 struct page
*page
, void *fsdata
)
331 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
334 page_cache_release(page
);
341 * for a block special file file_inode(file)->i_size is zero
342 * so we compute the size by hand (just as in block_read/write above)
344 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
346 struct inode
*bd_inode
= bdev_file_inode(file
);
349 inode_lock(bd_inode
);
350 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
351 inode_unlock(bd_inode
);
355 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
357 struct inode
*bd_inode
= bdev_file_inode(filp
);
358 struct block_device
*bdev
= I_BDEV(bd_inode
);
361 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
366 * There is no need to serialise calls to blkdev_issue_flush with
367 * i_mutex and doing so causes performance issues with concurrent
368 * O_SYNC writers to a block device.
370 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
371 if (error
== -EOPNOTSUPP
)
376 EXPORT_SYMBOL(blkdev_fsync
);
379 * bdev_read_page() - Start reading a page from a block device
380 * @bdev: The device to read the page from
381 * @sector: The offset on the device to read the page to (need not be aligned)
382 * @page: The page to read
384 * On entry, the page should be locked. It will be unlocked when the page
385 * has been read. If the block driver implements rw_page synchronously,
386 * that will be true on exit from this function, but it need not be.
388 * Errors returned by this function are usually "soft", eg out of memory, or
389 * queue full; callers should try a different route to read this page rather
390 * than propagate an error back up the stack.
392 * Return: negative errno if an error occurs, 0 if submission was successful.
394 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
397 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
398 int result
= -EOPNOTSUPP
;
400 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
403 result
= blk_queue_enter(bdev
->bd_queue
, false);
406 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, READ
);
407 blk_queue_exit(bdev
->bd_queue
);
410 EXPORT_SYMBOL_GPL(bdev_read_page
);
413 * bdev_write_page() - Start writing a page to a block device
414 * @bdev: The device to write the page to
415 * @sector: The offset on the device to write the page to (need not be aligned)
416 * @page: The page to write
417 * @wbc: The writeback_control for the write
419 * On entry, the page should be locked and not currently under writeback.
420 * On exit, if the write started successfully, the page will be unlocked and
421 * under writeback. If the write failed already (eg the driver failed to
422 * queue the page to the device), the page will still be locked. If the
423 * caller is a ->writepage implementation, it will need to unlock the page.
425 * Errors returned by this function are usually "soft", eg out of memory, or
426 * queue full; callers should try a different route to write this page rather
427 * than propagate an error back up the stack.
429 * Return: negative errno if an error occurs, 0 if submission was successful.
431 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
432 struct page
*page
, struct writeback_control
*wbc
)
435 int rw
= (wbc
->sync_mode
== WB_SYNC_ALL
) ? WRITE_SYNC
: WRITE
;
436 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
438 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
440 result
= blk_queue_enter(bdev
->bd_queue
, false);
444 set_page_writeback(page
);
445 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, rw
);
447 end_page_writeback(page
);
450 blk_queue_exit(bdev
->bd_queue
);
453 EXPORT_SYMBOL_GPL(bdev_write_page
);
456 * bdev_direct_access() - Get the address for directly-accessibly memory
457 * @bdev: The device containing the memory
458 * @dax: control and output parameters for ->direct_access
460 * If a block device is made up of directly addressable memory, this function
461 * will tell the caller the PFN and the address of the memory. The address
462 * may be directly dereferenced within the kernel without the need to call
463 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
466 * Return: negative errno if an error occurs, otherwise the number of bytes
467 * accessible at this address.
469 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
471 sector_t sector
= dax
->sector
;
472 long avail
, size
= dax
->size
;
473 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
476 * The device driver is allowed to sleep, in order to make the
477 * memory directly accessible.
483 if (!ops
->direct_access
)
485 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
486 part_nr_sects_read(bdev
->bd_part
))
488 sector
+= get_start_sect(bdev
);
489 if (sector
% (PAGE_SIZE
/ 512))
491 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
);
494 if (avail
> 0 && avail
& ~PAGE_MASK
)
496 return min(avail
, size
);
498 EXPORT_SYMBOL_GPL(bdev_direct_access
);
504 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
505 static struct kmem_cache
* bdev_cachep __read_mostly
;
507 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
509 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
512 return &ei
->vfs_inode
;
515 static void bdev_i_callback(struct rcu_head
*head
)
517 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
518 struct bdev_inode
*bdi
= BDEV_I(inode
);
520 kmem_cache_free(bdev_cachep
, bdi
);
523 static void bdev_destroy_inode(struct inode
*inode
)
525 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
528 static void init_once(void *foo
)
530 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
531 struct block_device
*bdev
= &ei
->bdev
;
533 memset(bdev
, 0, sizeof(*bdev
));
534 mutex_init(&bdev
->bd_mutex
);
535 INIT_LIST_HEAD(&bdev
->bd_inodes
);
536 INIT_LIST_HEAD(&bdev
->bd_list
);
538 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
540 inode_init_once(&ei
->vfs_inode
);
541 /* Initialize mutex for freeze. */
542 mutex_init(&bdev
->bd_fsfreeze_mutex
);
545 static inline void __bd_forget(struct inode
*inode
)
547 list_del_init(&inode
->i_devices
);
548 inode
->i_bdev
= NULL
;
549 inode
->i_mapping
= &inode
->i_data
;
552 static void bdev_evict_inode(struct inode
*inode
)
554 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
556 truncate_inode_pages_final(&inode
->i_data
);
557 invalidate_inode_buffers(inode
); /* is it needed here? */
559 spin_lock(&bdev_lock
);
560 while ( (p
= bdev
->bd_inodes
.next
) != &bdev
->bd_inodes
) {
561 __bd_forget(list_entry(p
, struct inode
, i_devices
));
563 list_del_init(&bdev
->bd_list
);
564 spin_unlock(&bdev_lock
);
567 static const struct super_operations bdev_sops
= {
568 .statfs
= simple_statfs
,
569 .alloc_inode
= bdev_alloc_inode
,
570 .destroy_inode
= bdev_destroy_inode
,
571 .drop_inode
= generic_delete_inode
,
572 .evict_inode
= bdev_evict_inode
,
575 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
576 int flags
, const char *dev_name
, void *data
)
578 return mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
581 static struct file_system_type bd_type
= {
584 .kill_sb
= kill_anon_super
,
587 struct super_block
*blockdev_superblock __read_mostly
;
588 EXPORT_SYMBOL_GPL(blockdev_superblock
);
590 void __init
bdev_cache_init(void)
593 static struct vfsmount
*bd_mnt
;
595 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
596 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
597 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
599 err
= register_filesystem(&bd_type
);
601 panic("Cannot register bdev pseudo-fs");
602 bd_mnt
= kern_mount(&bd_type
);
604 panic("Cannot create bdev pseudo-fs");
605 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
609 * Most likely _very_ bad one - but then it's hardly critical for small
610 * /dev and can be fixed when somebody will need really large one.
611 * Keep in mind that it will be fed through icache hash function too.
613 static inline unsigned long hash(dev_t dev
)
615 return MAJOR(dev
)+MINOR(dev
);
618 static int bdev_test(struct inode
*inode
, void *data
)
620 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
623 static int bdev_set(struct inode
*inode
, void *data
)
625 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
629 static LIST_HEAD(all_bdevs
);
631 struct block_device
*bdget(dev_t dev
)
633 struct block_device
*bdev
;
636 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
637 bdev_test
, bdev_set
, &dev
);
642 bdev
= &BDEV_I(inode
)->bdev
;
644 if (inode
->i_state
& I_NEW
) {
645 bdev
->bd_contains
= NULL
;
646 bdev
->bd_super
= NULL
;
647 bdev
->bd_inode
= inode
;
648 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
649 bdev
->bd_part_count
= 0;
650 bdev
->bd_invalidated
= 0;
651 inode
->i_mode
= S_IFBLK
;
653 inode
->i_bdev
= bdev
;
654 inode
->i_data
.a_ops
= &def_blk_aops
;
655 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
656 spin_lock(&bdev_lock
);
657 list_add(&bdev
->bd_list
, &all_bdevs
);
658 spin_unlock(&bdev_lock
);
659 unlock_new_inode(inode
);
664 EXPORT_SYMBOL(bdget
);
667 * bdgrab -- Grab a reference to an already referenced block device
668 * @bdev: Block device to grab a reference to.
670 struct block_device
*bdgrab(struct block_device
*bdev
)
672 ihold(bdev
->bd_inode
);
675 EXPORT_SYMBOL(bdgrab
);
677 long nr_blockdev_pages(void)
679 struct block_device
*bdev
;
681 spin_lock(&bdev_lock
);
682 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
683 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
685 spin_unlock(&bdev_lock
);
689 void bdput(struct block_device
*bdev
)
691 iput(bdev
->bd_inode
);
694 EXPORT_SYMBOL(bdput
);
696 static struct block_device
*bd_acquire(struct inode
*inode
)
698 struct block_device
*bdev
;
700 spin_lock(&bdev_lock
);
701 bdev
= inode
->i_bdev
;
704 spin_unlock(&bdev_lock
);
707 spin_unlock(&bdev_lock
);
709 bdev
= bdget(inode
->i_rdev
);
711 spin_lock(&bdev_lock
);
712 if (!inode
->i_bdev
) {
714 * We take an additional reference to bd_inode,
715 * and it's released in clear_inode() of inode.
716 * So, we can access it via ->i_mapping always
720 inode
->i_bdev
= bdev
;
721 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
722 list_add(&inode
->i_devices
, &bdev
->bd_inodes
);
724 spin_unlock(&bdev_lock
);
729 /* Call when you free inode */
731 void bd_forget(struct inode
*inode
)
733 struct block_device
*bdev
= NULL
;
735 spin_lock(&bdev_lock
);
736 if (!sb_is_blkdev_sb(inode
->i_sb
))
737 bdev
= inode
->i_bdev
;
739 spin_unlock(&bdev_lock
);
746 * bd_may_claim - test whether a block device can be claimed
747 * @bdev: block device of interest
748 * @whole: whole block device containing @bdev, may equal @bdev
749 * @holder: holder trying to claim @bdev
751 * Test whether @bdev can be claimed by @holder.
754 * spin_lock(&bdev_lock).
757 * %true if @bdev can be claimed, %false otherwise.
759 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
762 if (bdev
->bd_holder
== holder
)
763 return true; /* already a holder */
764 else if (bdev
->bd_holder
!= NULL
)
765 return false; /* held by someone else */
766 else if (bdev
->bd_contains
== bdev
)
767 return true; /* is a whole device which isn't held */
769 else if (whole
->bd_holder
== bd_may_claim
)
770 return true; /* is a partition of a device that is being partitioned */
771 else if (whole
->bd_holder
!= NULL
)
772 return false; /* is a partition of a held device */
774 return true; /* is a partition of an un-held device */
778 * bd_prepare_to_claim - prepare to claim a block device
779 * @bdev: block device of interest
780 * @whole: the whole device containing @bdev, may equal @bdev
781 * @holder: holder trying to claim @bdev
783 * Prepare to claim @bdev. This function fails if @bdev is already
784 * claimed by another holder and waits if another claiming is in
785 * progress. This function doesn't actually claim. On successful
786 * return, the caller has ownership of bd_claiming and bd_holder[s].
789 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
793 * 0 if @bdev can be claimed, -EBUSY otherwise.
795 static int bd_prepare_to_claim(struct block_device
*bdev
,
796 struct block_device
*whole
, void *holder
)
799 /* if someone else claimed, fail */
800 if (!bd_may_claim(bdev
, whole
, holder
))
803 /* if claiming is already in progress, wait for it to finish */
804 if (whole
->bd_claiming
) {
805 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
808 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
809 spin_unlock(&bdev_lock
);
811 finish_wait(wq
, &wait
);
812 spin_lock(&bdev_lock
);
821 * bd_start_claiming - start claiming a block device
822 * @bdev: block device of interest
823 * @holder: holder trying to claim @bdev
825 * @bdev is about to be opened exclusively. Check @bdev can be opened
826 * exclusively and mark that an exclusive open is in progress. Each
827 * successful call to this function must be matched with a call to
828 * either bd_finish_claiming() or bd_abort_claiming() (which do not
831 * This function is used to gain exclusive access to the block device
832 * without actually causing other exclusive open attempts to fail. It
833 * should be used when the open sequence itself requires exclusive
834 * access but may subsequently fail.
840 * Pointer to the block device containing @bdev on success, ERR_PTR()
843 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
846 struct gendisk
*disk
;
847 struct block_device
*whole
;
853 * @bdev might not have been initialized properly yet, look up
854 * and grab the outer block device the hard way.
856 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
858 return ERR_PTR(-ENXIO
);
861 * Normally, @bdev should equal what's returned from bdget_disk()
862 * if partno is 0; however, some drivers (floppy) use multiple
863 * bdev's for the same physical device and @bdev may be one of the
864 * aliases. Keep @bdev if partno is 0. This means claimer
865 * tracking is broken for those devices but it has always been that
869 whole
= bdget_disk(disk
, 0);
871 whole
= bdgrab(bdev
);
873 module_put(disk
->fops
->owner
);
876 return ERR_PTR(-ENOMEM
);
878 /* prepare to claim, if successful, mark claiming in progress */
879 spin_lock(&bdev_lock
);
881 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
883 whole
->bd_claiming
= holder
;
884 spin_unlock(&bdev_lock
);
887 spin_unlock(&bdev_lock
);
894 struct bd_holder_disk
{
895 struct list_head list
;
896 struct gendisk
*disk
;
900 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
901 struct gendisk
*disk
)
903 struct bd_holder_disk
*holder
;
905 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
906 if (holder
->disk
== disk
)
911 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
913 return sysfs_create_link(from
, to
, kobject_name(to
));
916 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
918 sysfs_remove_link(from
, kobject_name(to
));
922 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
923 * @bdev: the claimed slave bdev
924 * @disk: the holding disk
926 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
928 * This functions creates the following sysfs symlinks.
930 * - from "slaves" directory of the holder @disk to the claimed @bdev
931 * - from "holders" directory of the @bdev to the holder @disk
933 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
934 * passed to bd_link_disk_holder(), then:
936 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
937 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
939 * The caller must have claimed @bdev before calling this function and
940 * ensure that both @bdev and @disk are valid during the creation and
941 * lifetime of these symlinks.
947 * 0 on success, -errno on failure.
949 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
951 struct bd_holder_disk
*holder
;
954 mutex_lock(&bdev
->bd_mutex
);
956 WARN_ON_ONCE(!bdev
->bd_holder
);
958 /* FIXME: remove the following once add_disk() handles errors */
959 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
962 holder
= bd_find_holder_disk(bdev
, disk
);
968 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
974 INIT_LIST_HEAD(&holder
->list
);
978 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
982 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
986 * bdev could be deleted beneath us which would implicitly destroy
987 * the holder directory. Hold on to it.
989 kobject_get(bdev
->bd_part
->holder_dir
);
991 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
995 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
999 mutex_unlock(&bdev
->bd_mutex
);
1002 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1005 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1006 * @bdev: the calimed slave bdev
1007 * @disk: the holding disk
1009 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1014 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1016 struct bd_holder_disk
*holder
;
1018 mutex_lock(&bdev
->bd_mutex
);
1020 holder
= bd_find_holder_disk(bdev
, disk
);
1022 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1023 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1024 del_symlink(bdev
->bd_part
->holder_dir
,
1025 &disk_to_dev(disk
)->kobj
);
1026 kobject_put(bdev
->bd_part
->holder_dir
);
1027 list_del_init(&holder
->list
);
1031 mutex_unlock(&bdev
->bd_mutex
);
1033 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1037 * flush_disk - invalidates all buffer-cache entries on a disk
1039 * @bdev: struct block device to be flushed
1040 * @kill_dirty: flag to guide handling of dirty inodes
1042 * Invalidates all buffer-cache entries on a disk. It should be called
1043 * when a disk has been changed -- either by a media change or online
1046 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1048 if (__invalidate_device(bdev
, kill_dirty
)) {
1049 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1050 "resized disk %s\n",
1051 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1056 if (disk_part_scan_enabled(bdev
->bd_disk
))
1057 bdev
->bd_invalidated
= 1;
1061 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1062 * @disk: struct gendisk to check
1063 * @bdev: struct bdev to adjust.
1065 * This routine checks to see if the bdev size does not match the disk size
1066 * and adjusts it if it differs.
1068 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1070 loff_t disk_size
, bdev_size
;
1072 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1073 bdev_size
= i_size_read(bdev
->bd_inode
);
1074 if (disk_size
!= bdev_size
) {
1076 "%s: detected capacity change from %lld to %lld\n",
1077 disk
->disk_name
, bdev_size
, disk_size
);
1078 i_size_write(bdev
->bd_inode
, disk_size
);
1079 flush_disk(bdev
, false);
1082 EXPORT_SYMBOL(check_disk_size_change
);
1085 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1086 * @disk: struct gendisk to be revalidated
1088 * This routine is a wrapper for lower-level driver's revalidate_disk
1089 * call-backs. It is used to do common pre and post operations needed
1090 * for all revalidate_disk operations.
1092 int revalidate_disk(struct gendisk
*disk
)
1094 struct block_device
*bdev
;
1097 if (disk
->fops
->revalidate_disk
)
1098 ret
= disk
->fops
->revalidate_disk(disk
);
1099 blk_integrity_revalidate(disk
);
1100 bdev
= bdget_disk(disk
, 0);
1104 mutex_lock(&bdev
->bd_mutex
);
1105 check_disk_size_change(disk
, bdev
);
1106 bdev
->bd_invalidated
= 0;
1107 mutex_unlock(&bdev
->bd_mutex
);
1111 EXPORT_SYMBOL(revalidate_disk
);
1114 * This routine checks whether a removable media has been changed,
1115 * and invalidates all buffer-cache-entries in that case. This
1116 * is a relatively slow routine, so we have to try to minimize using
1117 * it. Thus it is called only upon a 'mount' or 'open'. This
1118 * is the best way of combining speed and utility, I think.
1119 * People changing diskettes in the middle of an operation deserve
1122 int check_disk_change(struct block_device
*bdev
)
1124 struct gendisk
*disk
= bdev
->bd_disk
;
1125 const struct block_device_operations
*bdops
= disk
->fops
;
1126 unsigned int events
;
1128 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1129 DISK_EVENT_EJECT_REQUEST
);
1130 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1133 flush_disk(bdev
, true);
1134 if (bdops
->revalidate_disk
)
1135 bdops
->revalidate_disk(bdev
->bd_disk
);
1139 EXPORT_SYMBOL(check_disk_change
);
1141 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1143 unsigned bsize
= bdev_logical_block_size(bdev
);
1145 inode_lock(bdev
->bd_inode
);
1146 i_size_write(bdev
->bd_inode
, size
);
1147 inode_unlock(bdev
->bd_inode
);
1148 while (bsize
< PAGE_CACHE_SIZE
) {
1153 bdev
->bd_block_size
= bsize
;
1154 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1156 EXPORT_SYMBOL(bd_set_size
);
1158 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1163 * mutex_lock(part->bd_mutex)
1164 * mutex_lock_nested(whole->bd_mutex, 1)
1167 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1169 struct gendisk
*disk
;
1170 struct module
*owner
;
1175 if (mode
& FMODE_READ
)
1177 if (mode
& FMODE_WRITE
)
1180 * hooks: /n/, see "layering violations".
1183 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1193 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1196 owner
= disk
->fops
->owner
;
1198 disk_block_events(disk
);
1199 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1200 if (!bdev
->bd_openers
) {
1201 bdev
->bd_disk
= disk
;
1202 bdev
->bd_queue
= disk
->queue
;
1203 bdev
->bd_contains
= bdev
;
1204 if (IS_ENABLED(CONFIG_BLK_DEV_DAX
) && disk
->fops
->direct_access
)
1205 bdev
->bd_inode
->i_flags
= S_DAX
;
1207 bdev
->bd_inode
->i_flags
= 0;
1211 bdev
->bd_part
= disk_get_part(disk
, partno
);
1216 if (disk
->fops
->open
) {
1217 ret
= disk
->fops
->open(bdev
, mode
);
1218 if (ret
== -ERESTARTSYS
) {
1219 /* Lost a race with 'disk' being
1220 * deleted, try again.
1223 disk_put_part(bdev
->bd_part
);
1224 bdev
->bd_part
= NULL
;
1225 bdev
->bd_disk
= NULL
;
1226 bdev
->bd_queue
= NULL
;
1227 mutex_unlock(&bdev
->bd_mutex
);
1228 disk_unblock_events(disk
);
1236 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1237 if (!blkdev_dax_capable(bdev
))
1238 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1242 * If the device is invalidated, rescan partition
1243 * if open succeeded or failed with -ENOMEDIUM.
1244 * The latter is necessary to prevent ghost
1245 * partitions on a removed medium.
1247 if (bdev
->bd_invalidated
) {
1249 rescan_partitions(disk
, bdev
);
1250 else if (ret
== -ENOMEDIUM
)
1251 invalidate_partitions(disk
, bdev
);
1257 struct block_device
*whole
;
1258 whole
= bdget_disk(disk
, 0);
1263 ret
= __blkdev_get(whole
, mode
, 1);
1266 bdev
->bd_contains
= whole
;
1267 bdev
->bd_part
= disk_get_part(disk
, partno
);
1268 if (!(disk
->flags
& GENHD_FL_UP
) ||
1269 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1273 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1274 if (!blkdev_dax_capable(bdev
))
1275 bdev
->bd_inode
->i_flags
&= ~S_DAX
;
1278 if (bdev
->bd_contains
== bdev
) {
1280 if (bdev
->bd_disk
->fops
->open
)
1281 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1282 /* the same as first opener case, read comment there */
1283 if (bdev
->bd_invalidated
) {
1285 rescan_partitions(bdev
->bd_disk
, bdev
);
1286 else if (ret
== -ENOMEDIUM
)
1287 invalidate_partitions(bdev
->bd_disk
, bdev
);
1290 goto out_unlock_bdev
;
1292 /* only one opener holds refs to the module and disk */
1298 bdev
->bd_part_count
++;
1299 mutex_unlock(&bdev
->bd_mutex
);
1300 disk_unblock_events(disk
);
1304 disk_put_part(bdev
->bd_part
);
1305 bdev
->bd_disk
= NULL
;
1306 bdev
->bd_part
= NULL
;
1307 bdev
->bd_queue
= NULL
;
1308 if (bdev
!= bdev
->bd_contains
)
1309 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1310 bdev
->bd_contains
= NULL
;
1312 mutex_unlock(&bdev
->bd_mutex
);
1313 disk_unblock_events(disk
);
1323 * blkdev_get - open a block device
1324 * @bdev: block_device to open
1325 * @mode: FMODE_* mask
1326 * @holder: exclusive holder identifier
1328 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1329 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1330 * @holder is invalid. Exclusive opens may nest for the same @holder.
1332 * On success, the reference count of @bdev is unchanged. On failure,
1339 * 0 on success, -errno on failure.
1341 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1343 struct block_device
*whole
= NULL
;
1346 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1348 if ((mode
& FMODE_EXCL
) && holder
) {
1349 whole
= bd_start_claiming(bdev
, holder
);
1350 if (IS_ERR(whole
)) {
1352 return PTR_ERR(whole
);
1356 res
= __blkdev_get(bdev
, mode
, 0);
1359 struct gendisk
*disk
= whole
->bd_disk
;
1361 /* finish claiming */
1362 mutex_lock(&bdev
->bd_mutex
);
1363 spin_lock(&bdev_lock
);
1366 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1368 * Note that for a whole device bd_holders
1369 * will be incremented twice, and bd_holder
1370 * will be set to bd_may_claim before being
1373 whole
->bd_holders
++;
1374 whole
->bd_holder
= bd_may_claim
;
1376 bdev
->bd_holder
= holder
;
1379 /* tell others that we're done */
1380 BUG_ON(whole
->bd_claiming
!= holder
);
1381 whole
->bd_claiming
= NULL
;
1382 wake_up_bit(&whole
->bd_claiming
, 0);
1384 spin_unlock(&bdev_lock
);
1387 * Block event polling for write claims if requested. Any
1388 * write holder makes the write_holder state stick until
1389 * all are released. This is good enough and tracking
1390 * individual writeable reference is too fragile given the
1391 * way @mode is used in blkdev_get/put().
1393 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1394 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1395 bdev
->bd_write_holder
= true;
1396 disk_block_events(disk
);
1399 mutex_unlock(&bdev
->bd_mutex
);
1405 EXPORT_SYMBOL(blkdev_get
);
1408 * blkdev_get_by_path - open a block device by name
1409 * @path: path to the block device to open
1410 * @mode: FMODE_* mask
1411 * @holder: exclusive holder identifier
1413 * Open the blockdevice described by the device file at @path. @mode
1414 * and @holder are identical to blkdev_get().
1416 * On success, the returned block_device has reference count of one.
1422 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1424 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1427 struct block_device
*bdev
;
1430 bdev
= lookup_bdev(path
);
1434 err
= blkdev_get(bdev
, mode
, holder
);
1436 return ERR_PTR(err
);
1438 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1439 blkdev_put(bdev
, mode
);
1440 return ERR_PTR(-EACCES
);
1445 EXPORT_SYMBOL(blkdev_get_by_path
);
1448 * blkdev_get_by_dev - open a block device by device number
1449 * @dev: device number of block device to open
1450 * @mode: FMODE_* mask
1451 * @holder: exclusive holder identifier
1453 * Open the blockdevice described by device number @dev. @mode and
1454 * @holder are identical to blkdev_get().
1456 * Use it ONLY if you really do not have anything better - i.e. when
1457 * you are behind a truly sucky interface and all you are given is a
1458 * device number. _Never_ to be used for internal purposes. If you
1459 * ever need it - reconsider your API.
1461 * On success, the returned block_device has reference count of one.
1467 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1469 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1471 struct block_device
*bdev
;
1476 return ERR_PTR(-ENOMEM
);
1478 err
= blkdev_get(bdev
, mode
, holder
);
1480 return ERR_PTR(err
);
1484 EXPORT_SYMBOL(blkdev_get_by_dev
);
1486 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1488 struct block_device
*bdev
;
1491 * Preserve backwards compatibility and allow large file access
1492 * even if userspace doesn't ask for it explicitly. Some mkfs
1493 * binary needs it. We might want to drop this workaround
1494 * during an unstable branch.
1496 filp
->f_flags
|= O_LARGEFILE
;
1498 if (filp
->f_flags
& O_NDELAY
)
1499 filp
->f_mode
|= FMODE_NDELAY
;
1500 if (filp
->f_flags
& O_EXCL
)
1501 filp
->f_mode
|= FMODE_EXCL
;
1502 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1503 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1505 bdev
= bd_acquire(inode
);
1509 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1511 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1514 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1516 struct gendisk
*disk
= bdev
->bd_disk
;
1517 struct block_device
*victim
= NULL
;
1519 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1521 bdev
->bd_part_count
--;
1523 if (!--bdev
->bd_openers
) {
1524 WARN_ON_ONCE(bdev
->bd_holders
);
1525 sync_blockdev(bdev
);
1528 bdev_write_inode(bdev
);
1530 * Detaching bdev inode from its wb in __destroy_inode()
1531 * is too late: the queue which embeds its bdi (along with
1532 * root wb) can be gone as soon as we put_disk() below.
1534 inode_detach_wb(bdev
->bd_inode
);
1536 if (bdev
->bd_contains
== bdev
) {
1537 if (disk
->fops
->release
)
1538 disk
->fops
->release(disk
, mode
);
1540 if (!bdev
->bd_openers
) {
1541 struct module
*owner
= disk
->fops
->owner
;
1543 disk_put_part(bdev
->bd_part
);
1544 bdev
->bd_part
= NULL
;
1545 bdev
->bd_disk
= NULL
;
1546 if (bdev
!= bdev
->bd_contains
)
1547 victim
= bdev
->bd_contains
;
1548 bdev
->bd_contains
= NULL
;
1553 mutex_unlock(&bdev
->bd_mutex
);
1556 __blkdev_put(victim
, mode
, 1);
1559 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1561 mutex_lock(&bdev
->bd_mutex
);
1563 if (mode
& FMODE_EXCL
) {
1567 * Release a claim on the device. The holder fields
1568 * are protected with bdev_lock. bd_mutex is to
1569 * synchronize disk_holder unlinking.
1571 spin_lock(&bdev_lock
);
1573 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1574 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1576 /* bd_contains might point to self, check in a separate step */
1577 if ((bdev_free
= !bdev
->bd_holders
))
1578 bdev
->bd_holder
= NULL
;
1579 if (!bdev
->bd_contains
->bd_holders
)
1580 bdev
->bd_contains
->bd_holder
= NULL
;
1582 spin_unlock(&bdev_lock
);
1585 * If this was the last claim, remove holder link and
1586 * unblock evpoll if it was a write holder.
1588 if (bdev_free
&& bdev
->bd_write_holder
) {
1589 disk_unblock_events(bdev
->bd_disk
);
1590 bdev
->bd_write_holder
= false;
1595 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1596 * event. This is to ensure detection of media removal commanded
1597 * from userland - e.g. eject(1).
1599 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1601 mutex_unlock(&bdev
->bd_mutex
);
1603 __blkdev_put(bdev
, mode
, 0);
1605 EXPORT_SYMBOL(blkdev_put
);
1607 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1609 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1610 blkdev_put(bdev
, filp
->f_mode
);
1614 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1616 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1617 fmode_t mode
= file
->f_mode
;
1620 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1621 * to updated it before every ioctl.
1623 if (file
->f_flags
& O_NDELAY
)
1624 mode
|= FMODE_NDELAY
;
1626 mode
&= ~FMODE_NDELAY
;
1628 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1632 * Write data to the block device. Only intended for the block device itself
1633 * and the raw driver which basically is a fake block device.
1635 * Does not take i_mutex for the write and thus is not for general purpose
1638 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1640 struct file
*file
= iocb
->ki_filp
;
1641 struct inode
*bd_inode
= bdev_file_inode(file
);
1642 loff_t size
= i_size_read(bd_inode
);
1643 struct blk_plug plug
;
1646 if (bdev_read_only(I_BDEV(bd_inode
)))
1649 if (!iov_iter_count(from
))
1652 if (iocb
->ki_pos
>= size
)
1655 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1657 blk_start_plug(&plug
);
1658 ret
= __generic_file_write_iter(iocb
, from
);
1661 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
1665 blk_finish_plug(&plug
);
1668 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
1670 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
1672 struct file
*file
= iocb
->ki_filp
;
1673 struct inode
*bd_inode
= bdev_file_inode(file
);
1674 loff_t size
= i_size_read(bd_inode
);
1675 loff_t pos
= iocb
->ki_pos
;
1681 iov_iter_truncate(to
, size
);
1682 return generic_file_read_iter(iocb
, to
);
1684 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
1687 * Try to release a page associated with block device when the system
1688 * is under memory pressure.
1690 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
1692 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
1694 if (super
&& super
->s_op
->bdev_try_to_free_page
)
1695 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
1697 return try_to_free_buffers(page
);
1700 static int blkdev_writepages(struct address_space
*mapping
,
1701 struct writeback_control
*wbc
)
1703 if (dax_mapping(mapping
)) {
1704 struct block_device
*bdev
= I_BDEV(mapping
->host
);
1706 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
1708 return generic_writepages(mapping
, wbc
);
1711 static const struct address_space_operations def_blk_aops
= {
1712 .readpage
= blkdev_readpage
,
1713 .readpages
= blkdev_readpages
,
1714 .writepage
= blkdev_writepage
,
1715 .write_begin
= blkdev_write_begin
,
1716 .write_end
= blkdev_write_end
,
1717 .writepages
= blkdev_writepages
,
1718 .releasepage
= blkdev_releasepage
,
1719 .direct_IO
= blkdev_direct_IO
,
1720 .is_dirty_writeback
= buffer_check_dirty_writeback
,
1723 #ifdef CONFIG_FS_DAX
1725 * In the raw block case we do not need to contend with truncation nor
1726 * unwritten file extents. Without those concerns there is no need for
1727 * additional locking beyond the mmap_sem context that these routines
1728 * are already executing under.
1730 * Note, there is no protection if the block device is dynamically
1731 * resized (partition grow/shrink) during a fault. A stable block device
1732 * size is already not enforced in the blkdev_direct_IO path.
1734 * For DAX, it is the responsibility of the block device driver to
1735 * ensure the whole-disk device size is stable while requests are in
1738 * Finally, unlike the filemap_page_mkwrite() case there is no
1739 * filesystem superblock to sync against freezing. We still include a
1740 * pfn_mkwrite callback for dax drivers to receive write fault
1743 static int blkdev_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1745 return __dax_fault(vma
, vmf
, blkdev_get_block
, NULL
);
1748 static int blkdev_dax_pfn_mkwrite(struct vm_area_struct
*vma
,
1749 struct vm_fault
*vmf
)
1751 return dax_pfn_mkwrite(vma
, vmf
);
1754 static int blkdev_dax_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
1755 pmd_t
*pmd
, unsigned int flags
)
1757 return __dax_pmd_fault(vma
, addr
, pmd
, flags
, blkdev_get_block
, NULL
);
1760 static const struct vm_operations_struct blkdev_dax_vm_ops
= {
1761 .fault
= blkdev_dax_fault
,
1762 .pmd_fault
= blkdev_dax_pmd_fault
,
1763 .pfn_mkwrite
= blkdev_dax_pfn_mkwrite
,
1766 static const struct vm_operations_struct blkdev_default_vm_ops
= {
1767 .fault
= filemap_fault
,
1768 .map_pages
= filemap_map_pages
,
1771 static int blkdev_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1773 struct inode
*bd_inode
= bdev_file_inode(file
);
1775 file_accessed(file
);
1776 if (IS_DAX(bd_inode
)) {
1777 vma
->vm_ops
= &blkdev_dax_vm_ops
;
1778 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
1780 vma
->vm_ops
= &blkdev_default_vm_ops
;
1786 #define blkdev_mmap generic_file_mmap
1789 const struct file_operations def_blk_fops
= {
1790 .open
= blkdev_open
,
1791 .release
= blkdev_close
,
1792 .llseek
= block_llseek
,
1793 .read_iter
= blkdev_read_iter
,
1794 .write_iter
= blkdev_write_iter
,
1795 .mmap
= blkdev_mmap
,
1796 .fsync
= blkdev_fsync
,
1797 .unlocked_ioctl
= block_ioctl
,
1798 #ifdef CONFIG_COMPAT
1799 .compat_ioctl
= compat_blkdev_ioctl
,
1801 .splice_read
= generic_file_splice_read
,
1802 .splice_write
= iter_file_splice_write
,
1805 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
1808 mm_segment_t old_fs
= get_fs();
1810 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
1815 EXPORT_SYMBOL(ioctl_by_bdev
);
1818 * lookup_bdev - lookup a struct block_device by name
1819 * @pathname: special file representing the block device
1821 * Get a reference to the blockdevice at @pathname in the current
1822 * namespace if possible and return it. Return ERR_PTR(error)
1825 struct block_device
*lookup_bdev(const char *pathname
)
1827 struct block_device
*bdev
;
1828 struct inode
*inode
;
1832 if (!pathname
|| !*pathname
)
1833 return ERR_PTR(-EINVAL
);
1835 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
1837 return ERR_PTR(error
);
1839 inode
= d_backing_inode(path
.dentry
);
1841 if (!S_ISBLK(inode
->i_mode
))
1844 if (path
.mnt
->mnt_flags
& MNT_NODEV
)
1847 bdev
= bd_acquire(inode
);
1854 bdev
= ERR_PTR(error
);
1857 EXPORT_SYMBOL(lookup_bdev
);
1859 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
1861 struct super_block
*sb
= get_super(bdev
);
1866 * no need to lock the super, get_super holds the
1867 * read mutex so the filesystem cannot go away
1868 * under us (->put_super runs with the write lock
1871 shrink_dcache_sb(sb
);
1872 res
= invalidate_inodes(sb
, kill_dirty
);
1875 invalidate_bdev(bdev
);
1878 EXPORT_SYMBOL(__invalidate_device
);
1880 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
1882 struct inode
*inode
, *old_inode
= NULL
;
1884 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1885 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
1886 struct address_space
*mapping
= inode
->i_mapping
;
1888 spin_lock(&inode
->i_lock
);
1889 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
1890 mapping
->nrpages
== 0) {
1891 spin_unlock(&inode
->i_lock
);
1895 spin_unlock(&inode
->i_lock
);
1896 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
1898 * We hold a reference to 'inode' so it couldn't have been
1899 * removed from s_inodes list while we dropped the
1900 * s_inode_list_lock We cannot iput the inode now as we can
1901 * be holding the last reference and we cannot iput it under
1902 * s_inode_list_lock. So we keep the reference and iput it
1908 func(I_BDEV(inode
), arg
);
1910 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
1912 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);