4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <linux/badblocks.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
39 struct block_device bdev
;
40 struct inode vfs_inode
;
43 static const struct address_space_operations def_blk_aops
;
45 static inline struct bdev_inode
*BDEV_I(struct inode
*inode
)
47 return container_of(inode
, struct bdev_inode
, vfs_inode
);
50 struct block_device
*I_BDEV(struct inode
*inode
)
52 return &BDEV_I(inode
)->bdev
;
54 EXPORT_SYMBOL(I_BDEV
);
56 void __vfs_msg(struct super_block
*sb
, const char *prefix
, const char *fmt
, ...)
64 printk_ratelimited("%sVFS (%s): %pV\n", prefix
, sb
->s_id
, &vaf
);
68 static void bdev_write_inode(struct block_device
*bdev
)
70 struct inode
*inode
= bdev
->bd_inode
;
73 spin_lock(&inode
->i_lock
);
74 while (inode
->i_state
& I_DIRTY
) {
75 spin_unlock(&inode
->i_lock
);
76 ret
= write_inode_now(inode
, true);
78 char name
[BDEVNAME_SIZE
];
79 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
80 "for block device %s (err=%d).\n",
81 bdevname(bdev
, name
), ret
);
83 spin_lock(&inode
->i_lock
);
85 spin_unlock(&inode
->i_lock
);
88 /* Kill _all_ buffers and pagecache , dirty or not.. */
89 void kill_bdev(struct block_device
*bdev
)
91 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
93 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
97 truncate_inode_pages(mapping
, 0);
99 EXPORT_SYMBOL(kill_bdev
);
101 /* Invalidate clean unused buffers and pagecache. */
102 void invalidate_bdev(struct block_device
*bdev
)
104 struct address_space
*mapping
= bdev
->bd_inode
->i_mapping
;
106 if (mapping
->nrpages
) {
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping
, 0, -1);
111 /* 99% of the time, we don't need to flush the cleancache on the bdev.
112 * But, for the strange corners, lets be cautious
114 cleancache_invalidate_inode(mapping
);
116 EXPORT_SYMBOL(invalidate_bdev
);
118 int set_blocksize(struct block_device
*bdev
, int size
)
120 /* Size must be a power of two, and between 512 and PAGE_SIZE */
121 if (size
> PAGE_SIZE
|| size
< 512 || !is_power_of_2(size
))
124 /* Size cannot be smaller than the size supported by the device */
125 if (size
< bdev_logical_block_size(bdev
))
128 /* Don't change the size if it is same as current */
129 if (bdev
->bd_block_size
!= size
) {
131 bdev
->bd_block_size
= size
;
132 bdev
->bd_inode
->i_blkbits
= blksize_bits(size
);
138 EXPORT_SYMBOL(set_blocksize
);
140 int sb_set_blocksize(struct super_block
*sb
, int size
)
142 if (set_blocksize(sb
->s_bdev
, size
))
144 /* If we get here, we know size is power of two
145 * and it's value is between 512 and PAGE_SIZE */
146 sb
->s_blocksize
= size
;
147 sb
->s_blocksize_bits
= blksize_bits(size
);
148 return sb
->s_blocksize
;
151 EXPORT_SYMBOL(sb_set_blocksize
);
153 int sb_min_blocksize(struct super_block
*sb
, int size
)
155 int minsize
= bdev_logical_block_size(sb
->s_bdev
);
158 return sb_set_blocksize(sb
, size
);
161 EXPORT_SYMBOL(sb_min_blocksize
);
164 blkdev_get_block(struct inode
*inode
, sector_t iblock
,
165 struct buffer_head
*bh
, int create
)
167 bh
->b_bdev
= I_BDEV(inode
);
168 bh
->b_blocknr
= iblock
;
169 set_buffer_mapped(bh
);
173 static struct inode
*bdev_file_inode(struct file
*file
)
175 return file
->f_mapping
->host
;
178 static unsigned int dio_bio_write_op(struct kiocb
*iocb
)
180 unsigned int op
= REQ_OP_WRITE
| REQ_SYNC
| REQ_IDLE
;
182 /* avoid the need for a I/O completion work item */
183 if (iocb
->ki_flags
& IOCB_DSYNC
)
188 #define DIO_INLINE_BIO_VECS 4
190 static void blkdev_bio_end_io_simple(struct bio
*bio
)
192 struct task_struct
*waiter
= bio
->bi_private
;
194 WRITE_ONCE(bio
->bi_private
, NULL
);
195 wake_up_process(waiter
);
199 __blkdev_direct_IO_simple(struct kiocb
*iocb
, struct iov_iter
*iter
,
202 struct file
*file
= iocb
->ki_filp
;
203 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
204 struct bio_vec inline_vecs
[DIO_INLINE_BIO_VECS
], *vecs
, *bvec
;
205 loff_t pos
= iocb
->ki_pos
;
206 bool should_dirty
= false;
212 if ((pos
| iov_iter_alignment(iter
)) &
213 (bdev_logical_block_size(bdev
) - 1))
216 if (nr_pages
<= DIO_INLINE_BIO_VECS
)
219 vecs
= kmalloc(nr_pages
* sizeof(struct bio_vec
), GFP_KERNEL
);
224 bio_init(&bio
, vecs
, nr_pages
);
226 bio
.bi_iter
.bi_sector
= pos
>> 9;
227 bio
.bi_private
= current
;
228 bio
.bi_end_io
= blkdev_bio_end_io_simple
;
230 ret
= bio_iov_iter_get_pages(&bio
, iter
);
233 ret
= bio
.bi_iter
.bi_size
;
235 if (iov_iter_rw(iter
) == READ
) {
236 bio
.bi_opf
= REQ_OP_READ
;
237 if (iter_is_iovec(iter
))
240 bio
.bi_opf
= dio_bio_write_op(iocb
);
241 task_io_account_write(ret
);
244 qc
= submit_bio(&bio
);
246 set_current_state(TASK_UNINTERRUPTIBLE
);
247 if (!READ_ONCE(bio
.bi_private
))
249 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
250 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
253 __set_current_state(TASK_RUNNING
);
255 bio_for_each_segment_all(bvec
, &bio
, i
) {
256 if (should_dirty
&& !PageCompound(bvec
->bv_page
))
257 set_page_dirty_lock(bvec
->bv_page
);
258 put_page(bvec
->bv_page
);
261 if (vecs
!= inline_vecs
)
264 if (unlikely(bio
.bi_error
))
272 struct task_struct
*waiter
;
277 bool should_dirty
: 1;
282 static struct bio_set
*blkdev_dio_pool __read_mostly
;
284 static void blkdev_bio_end_io(struct bio
*bio
)
286 struct blkdev_dio
*dio
= bio
->bi_private
;
287 bool should_dirty
= dio
->should_dirty
;
289 if (dio
->multi_bio
&& !atomic_dec_and_test(&dio
->ref
)) {
290 if (bio
->bi_error
&& !dio
->bio
.bi_error
)
291 dio
->bio
.bi_error
= bio
->bi_error
;
294 struct kiocb
*iocb
= dio
->iocb
;
295 ssize_t ret
= dio
->bio
.bi_error
;
302 dio
->iocb
->ki_complete(iocb
, ret
, 0);
305 struct task_struct
*waiter
= dio
->waiter
;
307 WRITE_ONCE(dio
->waiter
, NULL
);
308 wake_up_process(waiter
);
313 bio_check_pages_dirty(bio
);
315 struct bio_vec
*bvec
;
318 bio_for_each_segment_all(bvec
, bio
, i
)
319 put_page(bvec
->bv_page
);
325 __blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
, int nr_pages
)
327 struct file
*file
= iocb
->ki_filp
;
328 struct inode
*inode
= bdev_file_inode(file
);
329 struct block_device
*bdev
= I_BDEV(inode
);
330 struct blk_plug plug
;
331 struct blkdev_dio
*dio
;
333 bool is_read
= (iov_iter_rw(iter
) == READ
), is_sync
;
334 loff_t pos
= iocb
->ki_pos
;
335 blk_qc_t qc
= BLK_QC_T_NONE
;
338 if ((pos
| iov_iter_alignment(iter
)) &
339 (bdev_logical_block_size(bdev
) - 1))
342 bio
= bio_alloc_bioset(GFP_KERNEL
, nr_pages
, blkdev_dio_pool
);
343 bio_get(bio
); /* extra ref for the completion handler */
345 dio
= container_of(bio
, struct blkdev_dio
, bio
);
346 dio
->is_sync
= is_sync
= is_sync_kiocb(iocb
);
348 dio
->waiter
= current
;
353 dio
->multi_bio
= false;
354 dio
->should_dirty
= is_read
&& (iter
->type
== ITER_IOVEC
);
356 blk_start_plug(&plug
);
359 bio
->bi_iter
.bi_sector
= pos
>> 9;
360 bio
->bi_private
= dio
;
361 bio
->bi_end_io
= blkdev_bio_end_io
;
363 ret
= bio_iov_iter_get_pages(bio
, iter
);
371 bio
->bi_opf
= REQ_OP_READ
;
372 if (dio
->should_dirty
)
373 bio_set_pages_dirty(bio
);
375 bio
->bi_opf
= dio_bio_write_op(iocb
);
376 task_io_account_write(bio
->bi_iter
.bi_size
);
379 dio
->size
+= bio
->bi_iter
.bi_size
;
380 pos
+= bio
->bi_iter
.bi_size
;
382 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
);
384 qc
= submit_bio(bio
);
388 if (!dio
->multi_bio
) {
389 dio
->multi_bio
= true;
390 atomic_set(&dio
->ref
, 2);
392 atomic_inc(&dio
->ref
);
396 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
398 blk_finish_plug(&plug
);
404 set_current_state(TASK_UNINTERRUPTIBLE
);
405 if (!READ_ONCE(dio
->waiter
))
408 if (!(iocb
->ki_flags
& IOCB_HIPRI
) ||
409 !blk_mq_poll(bdev_get_queue(bdev
), qc
))
412 __set_current_state(TASK_RUNNING
);
414 ret
= dio
->bio
.bi_error
;
423 blkdev_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
427 nr_pages
= iov_iter_npages(iter
, BIO_MAX_PAGES
+ 1);
430 if (is_sync_kiocb(iocb
) && nr_pages
<= BIO_MAX_PAGES
)
431 return __blkdev_direct_IO_simple(iocb
, iter
, nr_pages
);
433 return __blkdev_direct_IO(iocb
, iter
, min(nr_pages
, BIO_MAX_PAGES
));
436 static __init
int blkdev_init(void)
438 blkdev_dio_pool
= bioset_create(4, offsetof(struct blkdev_dio
, bio
));
439 if (!blkdev_dio_pool
)
443 module_init(blkdev_init
);
445 int __sync_blockdev(struct block_device
*bdev
, int wait
)
450 return filemap_flush(bdev
->bd_inode
->i_mapping
);
451 return filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
455 * Write out and wait upon all the dirty data associated with a block
456 * device via its mapping. Does not take the superblock lock.
458 int sync_blockdev(struct block_device
*bdev
)
460 return __sync_blockdev(bdev
, 1);
462 EXPORT_SYMBOL(sync_blockdev
);
465 * Write out and wait upon all dirty data associated with this
466 * device. Filesystem data as well as the underlying block
467 * device. Takes the superblock lock.
469 int fsync_bdev(struct block_device
*bdev
)
471 struct super_block
*sb
= get_super(bdev
);
473 int res
= sync_filesystem(sb
);
477 return sync_blockdev(bdev
);
479 EXPORT_SYMBOL(fsync_bdev
);
482 * freeze_bdev -- lock a filesystem and force it into a consistent state
483 * @bdev: blockdevice to lock
485 * If a superblock is found on this device, we take the s_umount semaphore
486 * on it to make sure nobody unmounts until the snapshot creation is done.
487 * The reference counter (bd_fsfreeze_count) guarantees that only the last
488 * unfreeze process can unfreeze the frozen filesystem actually when multiple
489 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
490 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
493 struct super_block
*freeze_bdev(struct block_device
*bdev
)
495 struct super_block
*sb
;
498 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
499 if (++bdev
->bd_fsfreeze_count
> 1) {
501 * We don't even need to grab a reference - the first call
502 * to freeze_bdev grab an active reference and only the last
503 * thaw_bdev drops it.
505 sb
= get_super(bdev
);
508 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
512 sb
= get_active_super(bdev
);
515 if (sb
->s_op
->freeze_super
)
516 error
= sb
->s_op
->freeze_super(sb
);
518 error
= freeze_super(sb
);
520 deactivate_super(sb
);
521 bdev
->bd_fsfreeze_count
--;
522 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
523 return ERR_PTR(error
);
525 deactivate_super(sb
);
528 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
529 return sb
; /* thaw_bdev releases s->s_umount */
531 EXPORT_SYMBOL(freeze_bdev
);
534 * thaw_bdev -- unlock filesystem
535 * @bdev: blockdevice to unlock
536 * @sb: associated superblock
538 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
540 int thaw_bdev(struct block_device
*bdev
, struct super_block
*sb
)
544 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
545 if (!bdev
->bd_fsfreeze_count
)
549 if (--bdev
->bd_fsfreeze_count
> 0)
555 if (sb
->s_op
->thaw_super
)
556 error
= sb
->s_op
->thaw_super(sb
);
558 error
= thaw_super(sb
);
560 bdev
->bd_fsfreeze_count
++;
562 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
565 EXPORT_SYMBOL(thaw_bdev
);
567 static int blkdev_writepage(struct page
*page
, struct writeback_control
*wbc
)
569 return block_write_full_page(page
, blkdev_get_block
, wbc
);
572 static int blkdev_readpage(struct file
* file
, struct page
* page
)
574 return block_read_full_page(page
, blkdev_get_block
);
577 static int blkdev_readpages(struct file
*file
, struct address_space
*mapping
,
578 struct list_head
*pages
, unsigned nr_pages
)
580 return mpage_readpages(mapping
, pages
, nr_pages
, blkdev_get_block
);
583 static int blkdev_write_begin(struct file
*file
, struct address_space
*mapping
,
584 loff_t pos
, unsigned len
, unsigned flags
,
585 struct page
**pagep
, void **fsdata
)
587 return block_write_begin(mapping
, pos
, len
, flags
, pagep
,
591 static int blkdev_write_end(struct file
*file
, struct address_space
*mapping
,
592 loff_t pos
, unsigned len
, unsigned copied
,
593 struct page
*page
, void *fsdata
)
596 ret
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
606 * for a block special file file_inode(file)->i_size is zero
607 * so we compute the size by hand (just as in block_read/write above)
609 static loff_t
block_llseek(struct file
*file
, loff_t offset
, int whence
)
611 struct inode
*bd_inode
= bdev_file_inode(file
);
614 inode_lock(bd_inode
);
615 retval
= fixed_size_llseek(file
, offset
, whence
, i_size_read(bd_inode
));
616 inode_unlock(bd_inode
);
620 int blkdev_fsync(struct file
*filp
, loff_t start
, loff_t end
, int datasync
)
622 struct inode
*bd_inode
= bdev_file_inode(filp
);
623 struct block_device
*bdev
= I_BDEV(bd_inode
);
626 error
= filemap_write_and_wait_range(filp
->f_mapping
, start
, end
);
631 * There is no need to serialise calls to blkdev_issue_flush with
632 * i_mutex and doing so causes performance issues with concurrent
633 * O_SYNC writers to a block device.
635 error
= blkdev_issue_flush(bdev
, GFP_KERNEL
, NULL
);
636 if (error
== -EOPNOTSUPP
)
641 EXPORT_SYMBOL(blkdev_fsync
);
644 * bdev_read_page() - Start reading a page from a block device
645 * @bdev: The device to read the page from
646 * @sector: The offset on the device to read the page to (need not be aligned)
647 * @page: The page to read
649 * On entry, the page should be locked. It will be unlocked when the page
650 * has been read. If the block driver implements rw_page synchronously,
651 * that will be true on exit from this function, but it need not be.
653 * Errors returned by this function are usually "soft", eg out of memory, or
654 * queue full; callers should try a different route to read this page rather
655 * than propagate an error back up the stack.
657 * Return: negative errno if an error occurs, 0 if submission was successful.
659 int bdev_read_page(struct block_device
*bdev
, sector_t sector
,
662 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
663 int result
= -EOPNOTSUPP
;
665 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
668 result
= blk_queue_enter(bdev
->bd_queue
, false);
671 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, false);
672 blk_queue_exit(bdev
->bd_queue
);
675 EXPORT_SYMBOL_GPL(bdev_read_page
);
678 * bdev_write_page() - Start writing a page to a block device
679 * @bdev: The device to write the page to
680 * @sector: The offset on the device to write the page to (need not be aligned)
681 * @page: The page to write
682 * @wbc: The writeback_control for the write
684 * On entry, the page should be locked and not currently under writeback.
685 * On exit, if the write started successfully, the page will be unlocked and
686 * under writeback. If the write failed already (eg the driver failed to
687 * queue the page to the device), the page will still be locked. If the
688 * caller is a ->writepage implementation, it will need to unlock the page.
690 * Errors returned by this function are usually "soft", eg out of memory, or
691 * queue full; callers should try a different route to write this page rather
692 * than propagate an error back up the stack.
694 * Return: negative errno if an error occurs, 0 if submission was successful.
696 int bdev_write_page(struct block_device
*bdev
, sector_t sector
,
697 struct page
*page
, struct writeback_control
*wbc
)
700 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
702 if (!ops
->rw_page
|| bdev_get_integrity(bdev
))
704 result
= blk_queue_enter(bdev
->bd_queue
, false);
708 set_page_writeback(page
);
709 result
= ops
->rw_page(bdev
, sector
+ get_start_sect(bdev
), page
, true);
711 end_page_writeback(page
);
714 blk_queue_exit(bdev
->bd_queue
);
717 EXPORT_SYMBOL_GPL(bdev_write_page
);
720 * bdev_direct_access() - Get the address for directly-accessibly memory
721 * @bdev: The device containing the memory
722 * @dax: control and output parameters for ->direct_access
724 * If a block device is made up of directly addressable memory, this function
725 * will tell the caller the PFN and the address of the memory. The address
726 * may be directly dereferenced within the kernel without the need to call
727 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
730 * Return: negative errno if an error occurs, otherwise the number of bytes
731 * accessible at this address.
733 long bdev_direct_access(struct block_device
*bdev
, struct blk_dax_ctl
*dax
)
735 sector_t sector
= dax
->sector
;
736 long avail
, size
= dax
->size
;
737 const struct block_device_operations
*ops
= bdev
->bd_disk
->fops
;
740 * The device driver is allowed to sleep, in order to make the
741 * memory directly accessible.
747 if (!blk_queue_dax(bdev_get_queue(bdev
)) || !ops
->direct_access
)
749 if ((sector
+ DIV_ROUND_UP(size
, 512)) >
750 part_nr_sects_read(bdev
->bd_part
))
752 sector
+= get_start_sect(bdev
);
753 if (sector
% (PAGE_SIZE
/ 512))
755 avail
= ops
->direct_access(bdev
, sector
, &dax
->addr
, &dax
->pfn
, size
);
758 if (avail
> 0 && avail
& ~PAGE_MASK
)
760 return min(avail
, size
);
762 EXPORT_SYMBOL_GPL(bdev_direct_access
);
765 * bdev_dax_supported() - Check if the device supports dax for filesystem
766 * @sb: The superblock of the device
767 * @blocksize: The block size of the device
769 * This is a library function for filesystems to check if the block device
770 * can be mounted with dax option.
772 * Return: negative errno if unsupported, 0 if supported.
774 int bdev_dax_supported(struct super_block
*sb
, int blocksize
)
776 struct blk_dax_ctl dax
= {
782 if (blocksize
!= PAGE_SIZE
) {
783 vfs_msg(sb
, KERN_ERR
, "error: unsupported blocksize for dax");
787 err
= bdev_direct_access(sb
->s_bdev
, &dax
);
791 vfs_msg(sb
, KERN_ERR
,
792 "error: device does not support dax");
795 vfs_msg(sb
, KERN_ERR
,
796 "error: unaligned partition for dax");
799 vfs_msg(sb
, KERN_ERR
,
800 "error: dax access failed (%d)", err
);
807 EXPORT_SYMBOL_GPL(bdev_dax_supported
);
810 * bdev_dax_capable() - Return if the raw device is capable for dax
811 * @bdev: The device for raw block device access
813 bool bdev_dax_capable(struct block_device
*bdev
)
815 struct blk_dax_ctl dax
= {
819 if (!IS_ENABLED(CONFIG_FS_DAX
))
823 if (bdev_direct_access(bdev
, &dax
) < 0)
826 dax
.sector
= bdev
->bd_part
->nr_sects
- (PAGE_SIZE
/ 512);
827 if (bdev_direct_access(bdev
, &dax
) < 0)
837 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(bdev_lock
);
838 static struct kmem_cache
* bdev_cachep __read_mostly
;
840 static struct inode
*bdev_alloc_inode(struct super_block
*sb
)
842 struct bdev_inode
*ei
= kmem_cache_alloc(bdev_cachep
, GFP_KERNEL
);
845 return &ei
->vfs_inode
;
848 static void bdev_i_callback(struct rcu_head
*head
)
850 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
851 struct bdev_inode
*bdi
= BDEV_I(inode
);
853 kmem_cache_free(bdev_cachep
, bdi
);
856 static void bdev_destroy_inode(struct inode
*inode
)
858 call_rcu(&inode
->i_rcu
, bdev_i_callback
);
861 static void init_once(void *foo
)
863 struct bdev_inode
*ei
= (struct bdev_inode
*) foo
;
864 struct block_device
*bdev
= &ei
->bdev
;
866 memset(bdev
, 0, sizeof(*bdev
));
867 mutex_init(&bdev
->bd_mutex
);
868 INIT_LIST_HEAD(&bdev
->bd_list
);
870 INIT_LIST_HEAD(&bdev
->bd_holder_disks
);
872 inode_init_once(&ei
->vfs_inode
);
873 /* Initialize mutex for freeze. */
874 mutex_init(&bdev
->bd_fsfreeze_mutex
);
877 static void bdev_evict_inode(struct inode
*inode
)
879 struct block_device
*bdev
= &BDEV_I(inode
)->bdev
;
880 truncate_inode_pages_final(&inode
->i_data
);
881 invalidate_inode_buffers(inode
); /* is it needed here? */
883 spin_lock(&bdev_lock
);
884 list_del_init(&bdev
->bd_list
);
885 spin_unlock(&bdev_lock
);
888 static const struct super_operations bdev_sops
= {
889 .statfs
= simple_statfs
,
890 .alloc_inode
= bdev_alloc_inode
,
891 .destroy_inode
= bdev_destroy_inode
,
892 .drop_inode
= generic_delete_inode
,
893 .evict_inode
= bdev_evict_inode
,
896 static struct dentry
*bd_mount(struct file_system_type
*fs_type
,
897 int flags
, const char *dev_name
, void *data
)
900 dent
= mount_pseudo(fs_type
, "bdev:", &bdev_sops
, NULL
, BDEVFS_MAGIC
);
902 dent
->d_sb
->s_iflags
|= SB_I_CGROUPWB
;
906 static struct file_system_type bd_type
= {
909 .kill_sb
= kill_anon_super
,
912 struct super_block
*blockdev_superblock __read_mostly
;
913 EXPORT_SYMBOL_GPL(blockdev_superblock
);
915 void __init
bdev_cache_init(void)
918 static struct vfsmount
*bd_mnt
;
920 bdev_cachep
= kmem_cache_create("bdev_cache", sizeof(struct bdev_inode
),
921 0, (SLAB_HWCACHE_ALIGN
|SLAB_RECLAIM_ACCOUNT
|
922 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
|SLAB_PANIC
),
924 err
= register_filesystem(&bd_type
);
926 panic("Cannot register bdev pseudo-fs");
927 bd_mnt
= kern_mount(&bd_type
);
929 panic("Cannot create bdev pseudo-fs");
930 blockdev_superblock
= bd_mnt
->mnt_sb
; /* For writeback */
934 * Most likely _very_ bad one - but then it's hardly critical for small
935 * /dev and can be fixed when somebody will need really large one.
936 * Keep in mind that it will be fed through icache hash function too.
938 static inline unsigned long hash(dev_t dev
)
940 return MAJOR(dev
)+MINOR(dev
);
943 static int bdev_test(struct inode
*inode
, void *data
)
945 return BDEV_I(inode
)->bdev
.bd_dev
== *(dev_t
*)data
;
948 static int bdev_set(struct inode
*inode
, void *data
)
950 BDEV_I(inode
)->bdev
.bd_dev
= *(dev_t
*)data
;
954 static LIST_HEAD(all_bdevs
);
956 struct block_device
*bdget(dev_t dev
)
958 struct block_device
*bdev
;
961 inode
= iget5_locked(blockdev_superblock
, hash(dev
),
962 bdev_test
, bdev_set
, &dev
);
967 bdev
= &BDEV_I(inode
)->bdev
;
969 if (inode
->i_state
& I_NEW
) {
970 bdev
->bd_contains
= NULL
;
971 bdev
->bd_super
= NULL
;
972 bdev
->bd_inode
= inode
;
973 bdev
->bd_block_size
= (1 << inode
->i_blkbits
);
974 bdev
->bd_part_count
= 0;
975 bdev
->bd_invalidated
= 0;
976 inode
->i_mode
= S_IFBLK
;
978 inode
->i_bdev
= bdev
;
979 inode
->i_data
.a_ops
= &def_blk_aops
;
980 mapping_set_gfp_mask(&inode
->i_data
, GFP_USER
);
981 spin_lock(&bdev_lock
);
982 list_add(&bdev
->bd_list
, &all_bdevs
);
983 spin_unlock(&bdev_lock
);
984 unlock_new_inode(inode
);
989 EXPORT_SYMBOL(bdget
);
992 * bdgrab -- Grab a reference to an already referenced block device
993 * @bdev: Block device to grab a reference to.
995 struct block_device
*bdgrab(struct block_device
*bdev
)
997 ihold(bdev
->bd_inode
);
1000 EXPORT_SYMBOL(bdgrab
);
1002 long nr_blockdev_pages(void)
1004 struct block_device
*bdev
;
1006 spin_lock(&bdev_lock
);
1007 list_for_each_entry(bdev
, &all_bdevs
, bd_list
) {
1008 ret
+= bdev
->bd_inode
->i_mapping
->nrpages
;
1010 spin_unlock(&bdev_lock
);
1014 void bdput(struct block_device
*bdev
)
1016 iput(bdev
->bd_inode
);
1019 EXPORT_SYMBOL(bdput
);
1021 static struct block_device
*bd_acquire(struct inode
*inode
)
1023 struct block_device
*bdev
;
1025 spin_lock(&bdev_lock
);
1026 bdev
= inode
->i_bdev
;
1029 spin_unlock(&bdev_lock
);
1032 spin_unlock(&bdev_lock
);
1034 bdev
= bdget(inode
->i_rdev
);
1036 spin_lock(&bdev_lock
);
1037 if (!inode
->i_bdev
) {
1039 * We take an additional reference to bd_inode,
1040 * and it's released in clear_inode() of inode.
1041 * So, we can access it via ->i_mapping always
1045 inode
->i_bdev
= bdev
;
1046 inode
->i_mapping
= bdev
->bd_inode
->i_mapping
;
1048 spin_unlock(&bdev_lock
);
1053 /* Call when you free inode */
1055 void bd_forget(struct inode
*inode
)
1057 struct block_device
*bdev
= NULL
;
1059 spin_lock(&bdev_lock
);
1060 if (!sb_is_blkdev_sb(inode
->i_sb
))
1061 bdev
= inode
->i_bdev
;
1062 inode
->i_bdev
= NULL
;
1063 inode
->i_mapping
= &inode
->i_data
;
1064 spin_unlock(&bdev_lock
);
1071 * bd_may_claim - test whether a block device can be claimed
1072 * @bdev: block device of interest
1073 * @whole: whole block device containing @bdev, may equal @bdev
1074 * @holder: holder trying to claim @bdev
1076 * Test whether @bdev can be claimed by @holder.
1079 * spin_lock(&bdev_lock).
1082 * %true if @bdev can be claimed, %false otherwise.
1084 static bool bd_may_claim(struct block_device
*bdev
, struct block_device
*whole
,
1087 if (bdev
->bd_holder
== holder
)
1088 return true; /* already a holder */
1089 else if (bdev
->bd_holder
!= NULL
)
1090 return false; /* held by someone else */
1091 else if (whole
== bdev
)
1092 return true; /* is a whole device which isn't held */
1094 else if (whole
->bd_holder
== bd_may_claim
)
1095 return true; /* is a partition of a device that is being partitioned */
1096 else if (whole
->bd_holder
!= NULL
)
1097 return false; /* is a partition of a held device */
1099 return true; /* is a partition of an un-held device */
1103 * bd_prepare_to_claim - prepare to claim a block device
1104 * @bdev: block device of interest
1105 * @whole: the whole device containing @bdev, may equal @bdev
1106 * @holder: holder trying to claim @bdev
1108 * Prepare to claim @bdev. This function fails if @bdev is already
1109 * claimed by another holder and waits if another claiming is in
1110 * progress. This function doesn't actually claim. On successful
1111 * return, the caller has ownership of bd_claiming and bd_holder[s].
1114 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1115 * it multiple times.
1118 * 0 if @bdev can be claimed, -EBUSY otherwise.
1120 static int bd_prepare_to_claim(struct block_device
*bdev
,
1121 struct block_device
*whole
, void *holder
)
1124 /* if someone else claimed, fail */
1125 if (!bd_may_claim(bdev
, whole
, holder
))
1128 /* if claiming is already in progress, wait for it to finish */
1129 if (whole
->bd_claiming
) {
1130 wait_queue_head_t
*wq
= bit_waitqueue(&whole
->bd_claiming
, 0);
1133 prepare_to_wait(wq
, &wait
, TASK_UNINTERRUPTIBLE
);
1134 spin_unlock(&bdev_lock
);
1136 finish_wait(wq
, &wait
);
1137 spin_lock(&bdev_lock
);
1146 * bd_start_claiming - start claiming a block device
1147 * @bdev: block device of interest
1148 * @holder: holder trying to claim @bdev
1150 * @bdev is about to be opened exclusively. Check @bdev can be opened
1151 * exclusively and mark that an exclusive open is in progress. Each
1152 * successful call to this function must be matched with a call to
1153 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1156 * This function is used to gain exclusive access to the block device
1157 * without actually causing other exclusive open attempts to fail. It
1158 * should be used when the open sequence itself requires exclusive
1159 * access but may subsequently fail.
1165 * Pointer to the block device containing @bdev on success, ERR_PTR()
1168 static struct block_device
*bd_start_claiming(struct block_device
*bdev
,
1171 struct gendisk
*disk
;
1172 struct block_device
*whole
;
1178 * @bdev might not have been initialized properly yet, look up
1179 * and grab the outer block device the hard way.
1181 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1183 return ERR_PTR(-ENXIO
);
1186 * Normally, @bdev should equal what's returned from bdget_disk()
1187 * if partno is 0; however, some drivers (floppy) use multiple
1188 * bdev's for the same physical device and @bdev may be one of the
1189 * aliases. Keep @bdev if partno is 0. This means claimer
1190 * tracking is broken for those devices but it has always been that
1194 whole
= bdget_disk(disk
, 0);
1196 whole
= bdgrab(bdev
);
1198 module_put(disk
->fops
->owner
);
1201 return ERR_PTR(-ENOMEM
);
1203 /* prepare to claim, if successful, mark claiming in progress */
1204 spin_lock(&bdev_lock
);
1206 err
= bd_prepare_to_claim(bdev
, whole
, holder
);
1208 whole
->bd_claiming
= holder
;
1209 spin_unlock(&bdev_lock
);
1212 spin_unlock(&bdev_lock
);
1214 return ERR_PTR(err
);
1219 struct bd_holder_disk
{
1220 struct list_head list
;
1221 struct gendisk
*disk
;
1225 static struct bd_holder_disk
*bd_find_holder_disk(struct block_device
*bdev
,
1226 struct gendisk
*disk
)
1228 struct bd_holder_disk
*holder
;
1230 list_for_each_entry(holder
, &bdev
->bd_holder_disks
, list
)
1231 if (holder
->disk
== disk
)
1236 static int add_symlink(struct kobject
*from
, struct kobject
*to
)
1238 return sysfs_create_link(from
, to
, kobject_name(to
));
1241 static void del_symlink(struct kobject
*from
, struct kobject
*to
)
1243 sysfs_remove_link(from
, kobject_name(to
));
1247 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1248 * @bdev: the claimed slave bdev
1249 * @disk: the holding disk
1251 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1253 * This functions creates the following sysfs symlinks.
1255 * - from "slaves" directory of the holder @disk to the claimed @bdev
1256 * - from "holders" directory of the @bdev to the holder @disk
1258 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1259 * passed to bd_link_disk_holder(), then:
1261 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1262 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1264 * The caller must have claimed @bdev before calling this function and
1265 * ensure that both @bdev and @disk are valid during the creation and
1266 * lifetime of these symlinks.
1272 * 0 on success, -errno on failure.
1274 int bd_link_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1276 struct bd_holder_disk
*holder
;
1279 mutex_lock(&bdev
->bd_mutex
);
1281 WARN_ON_ONCE(!bdev
->bd_holder
);
1283 /* FIXME: remove the following once add_disk() handles errors */
1284 if (WARN_ON(!disk
->slave_dir
|| !bdev
->bd_part
->holder_dir
))
1287 holder
= bd_find_holder_disk(bdev
, disk
);
1293 holder
= kzalloc(sizeof(*holder
), GFP_KERNEL
);
1299 INIT_LIST_HEAD(&holder
->list
);
1300 holder
->disk
= disk
;
1303 ret
= add_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1307 ret
= add_symlink(bdev
->bd_part
->holder_dir
, &disk_to_dev(disk
)->kobj
);
1311 * bdev could be deleted beneath us which would implicitly destroy
1312 * the holder directory. Hold on to it.
1314 kobject_get(bdev
->bd_part
->holder_dir
);
1316 list_add(&holder
->list
, &bdev
->bd_holder_disks
);
1320 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1324 mutex_unlock(&bdev
->bd_mutex
);
1327 EXPORT_SYMBOL_GPL(bd_link_disk_holder
);
1330 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1331 * @bdev: the calimed slave bdev
1332 * @disk: the holding disk
1334 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1339 void bd_unlink_disk_holder(struct block_device
*bdev
, struct gendisk
*disk
)
1341 struct bd_holder_disk
*holder
;
1343 mutex_lock(&bdev
->bd_mutex
);
1345 holder
= bd_find_holder_disk(bdev
, disk
);
1347 if (!WARN_ON_ONCE(holder
== NULL
) && !--holder
->refcnt
) {
1348 del_symlink(disk
->slave_dir
, &part_to_dev(bdev
->bd_part
)->kobj
);
1349 del_symlink(bdev
->bd_part
->holder_dir
,
1350 &disk_to_dev(disk
)->kobj
);
1351 kobject_put(bdev
->bd_part
->holder_dir
);
1352 list_del_init(&holder
->list
);
1356 mutex_unlock(&bdev
->bd_mutex
);
1358 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder
);
1362 * flush_disk - invalidates all buffer-cache entries on a disk
1364 * @bdev: struct block device to be flushed
1365 * @kill_dirty: flag to guide handling of dirty inodes
1367 * Invalidates all buffer-cache entries on a disk. It should be called
1368 * when a disk has been changed -- either by a media change or online
1371 static void flush_disk(struct block_device
*bdev
, bool kill_dirty
)
1373 if (__invalidate_device(bdev
, kill_dirty
)) {
1374 printk(KERN_WARNING
"VFS: busy inodes on changed media or "
1375 "resized disk %s\n",
1376 bdev
->bd_disk
? bdev
->bd_disk
->disk_name
: "");
1381 if (disk_part_scan_enabled(bdev
->bd_disk
))
1382 bdev
->bd_invalidated
= 1;
1386 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1387 * @disk: struct gendisk to check
1388 * @bdev: struct bdev to adjust.
1390 * This routine checks to see if the bdev size does not match the disk size
1391 * and adjusts it if it differs.
1393 void check_disk_size_change(struct gendisk
*disk
, struct block_device
*bdev
)
1395 loff_t disk_size
, bdev_size
;
1397 disk_size
= (loff_t
)get_capacity(disk
) << 9;
1398 bdev_size
= i_size_read(bdev
->bd_inode
);
1399 if (disk_size
!= bdev_size
) {
1401 "%s: detected capacity change from %lld to %lld\n",
1402 disk
->disk_name
, bdev_size
, disk_size
);
1403 i_size_write(bdev
->bd_inode
, disk_size
);
1404 flush_disk(bdev
, false);
1407 EXPORT_SYMBOL(check_disk_size_change
);
1410 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1411 * @disk: struct gendisk to be revalidated
1413 * This routine is a wrapper for lower-level driver's revalidate_disk
1414 * call-backs. It is used to do common pre and post operations needed
1415 * for all revalidate_disk operations.
1417 int revalidate_disk(struct gendisk
*disk
)
1419 struct block_device
*bdev
;
1422 if (disk
->fops
->revalidate_disk
)
1423 ret
= disk
->fops
->revalidate_disk(disk
);
1424 bdev
= bdget_disk(disk
, 0);
1428 mutex_lock(&bdev
->bd_mutex
);
1429 check_disk_size_change(disk
, bdev
);
1430 bdev
->bd_invalidated
= 0;
1431 mutex_unlock(&bdev
->bd_mutex
);
1435 EXPORT_SYMBOL(revalidate_disk
);
1438 * This routine checks whether a removable media has been changed,
1439 * and invalidates all buffer-cache-entries in that case. This
1440 * is a relatively slow routine, so we have to try to minimize using
1441 * it. Thus it is called only upon a 'mount' or 'open'. This
1442 * is the best way of combining speed and utility, I think.
1443 * People changing diskettes in the middle of an operation deserve
1446 int check_disk_change(struct block_device
*bdev
)
1448 struct gendisk
*disk
= bdev
->bd_disk
;
1449 const struct block_device_operations
*bdops
= disk
->fops
;
1450 unsigned int events
;
1452 events
= disk_clear_events(disk
, DISK_EVENT_MEDIA_CHANGE
|
1453 DISK_EVENT_EJECT_REQUEST
);
1454 if (!(events
& DISK_EVENT_MEDIA_CHANGE
))
1457 flush_disk(bdev
, true);
1458 if (bdops
->revalidate_disk
)
1459 bdops
->revalidate_disk(bdev
->bd_disk
);
1463 EXPORT_SYMBOL(check_disk_change
);
1465 void bd_set_size(struct block_device
*bdev
, loff_t size
)
1467 unsigned bsize
= bdev_logical_block_size(bdev
);
1469 inode_lock(bdev
->bd_inode
);
1470 i_size_write(bdev
->bd_inode
, size
);
1471 inode_unlock(bdev
->bd_inode
);
1472 while (bsize
< PAGE_SIZE
) {
1477 bdev
->bd_block_size
= bsize
;
1478 bdev
->bd_inode
->i_blkbits
= blksize_bits(bsize
);
1480 EXPORT_SYMBOL(bd_set_size
);
1482 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
);
1487 * mutex_lock(part->bd_mutex)
1488 * mutex_lock_nested(whole->bd_mutex, 1)
1491 static int __blkdev_get(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1493 struct gendisk
*disk
;
1494 struct module
*owner
;
1499 if (mode
& FMODE_READ
)
1501 if (mode
& FMODE_WRITE
)
1504 * hooks: /n/, see "layering violations".
1507 ret
= devcgroup_inode_permission(bdev
->bd_inode
, perm
);
1517 disk
= get_gendisk(bdev
->bd_dev
, &partno
);
1520 owner
= disk
->fops
->owner
;
1522 disk_block_events(disk
);
1523 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1524 if (!bdev
->bd_openers
) {
1525 bdev
->bd_disk
= disk
;
1526 bdev
->bd_queue
= disk
->queue
;
1527 bdev
->bd_contains
= bdev
;
1531 bdev
->bd_part
= disk_get_part(disk
, partno
);
1536 if (disk
->fops
->open
) {
1537 ret
= disk
->fops
->open(bdev
, mode
);
1538 if (ret
== -ERESTARTSYS
) {
1539 /* Lost a race with 'disk' being
1540 * deleted, try again.
1543 disk_put_part(bdev
->bd_part
);
1544 bdev
->bd_part
= NULL
;
1545 bdev
->bd_disk
= NULL
;
1546 bdev
->bd_queue
= NULL
;
1547 mutex_unlock(&bdev
->bd_mutex
);
1548 disk_unblock_events(disk
);
1556 bd_set_size(bdev
,(loff_t
)get_capacity(disk
)<<9);
1559 * If the device is invalidated, rescan partition
1560 * if open succeeded or failed with -ENOMEDIUM.
1561 * The latter is necessary to prevent ghost
1562 * partitions on a removed medium.
1564 if (bdev
->bd_invalidated
) {
1566 rescan_partitions(disk
, bdev
);
1567 else if (ret
== -ENOMEDIUM
)
1568 invalidate_partitions(disk
, bdev
);
1574 struct block_device
*whole
;
1575 whole
= bdget_disk(disk
, 0);
1580 ret
= __blkdev_get(whole
, mode
, 1);
1583 bdev
->bd_contains
= whole
;
1584 bdev
->bd_part
= disk_get_part(disk
, partno
);
1585 if (!(disk
->flags
& GENHD_FL_UP
) ||
1586 !bdev
->bd_part
|| !bdev
->bd_part
->nr_sects
) {
1590 bd_set_size(bdev
, (loff_t
)bdev
->bd_part
->nr_sects
<< 9);
1593 if (bdev
->bd_contains
== bdev
) {
1595 if (bdev
->bd_disk
->fops
->open
)
1596 ret
= bdev
->bd_disk
->fops
->open(bdev
, mode
);
1597 /* the same as first opener case, read comment there */
1598 if (bdev
->bd_invalidated
) {
1600 rescan_partitions(bdev
->bd_disk
, bdev
);
1601 else if (ret
== -ENOMEDIUM
)
1602 invalidate_partitions(bdev
->bd_disk
, bdev
);
1605 goto out_unlock_bdev
;
1607 /* only one opener holds refs to the module and disk */
1613 bdev
->bd_part_count
++;
1614 mutex_unlock(&bdev
->bd_mutex
);
1615 disk_unblock_events(disk
);
1619 disk_put_part(bdev
->bd_part
);
1620 bdev
->bd_disk
= NULL
;
1621 bdev
->bd_part
= NULL
;
1622 bdev
->bd_queue
= NULL
;
1623 if (bdev
!= bdev
->bd_contains
)
1624 __blkdev_put(bdev
->bd_contains
, mode
, 1);
1625 bdev
->bd_contains
= NULL
;
1627 mutex_unlock(&bdev
->bd_mutex
);
1628 disk_unblock_events(disk
);
1638 * blkdev_get - open a block device
1639 * @bdev: block_device to open
1640 * @mode: FMODE_* mask
1641 * @holder: exclusive holder identifier
1643 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1644 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1645 * @holder is invalid. Exclusive opens may nest for the same @holder.
1647 * On success, the reference count of @bdev is unchanged. On failure,
1654 * 0 on success, -errno on failure.
1656 int blkdev_get(struct block_device
*bdev
, fmode_t mode
, void *holder
)
1658 struct block_device
*whole
= NULL
;
1661 WARN_ON_ONCE((mode
& FMODE_EXCL
) && !holder
);
1663 if ((mode
& FMODE_EXCL
) && holder
) {
1664 whole
= bd_start_claiming(bdev
, holder
);
1665 if (IS_ERR(whole
)) {
1667 return PTR_ERR(whole
);
1671 res
= __blkdev_get(bdev
, mode
, 0);
1674 struct gendisk
*disk
= whole
->bd_disk
;
1676 /* finish claiming */
1677 mutex_lock(&bdev
->bd_mutex
);
1678 spin_lock(&bdev_lock
);
1681 BUG_ON(!bd_may_claim(bdev
, whole
, holder
));
1683 * Note that for a whole device bd_holders
1684 * will be incremented twice, and bd_holder
1685 * will be set to bd_may_claim before being
1688 whole
->bd_holders
++;
1689 whole
->bd_holder
= bd_may_claim
;
1691 bdev
->bd_holder
= holder
;
1694 /* tell others that we're done */
1695 BUG_ON(whole
->bd_claiming
!= holder
);
1696 whole
->bd_claiming
= NULL
;
1697 wake_up_bit(&whole
->bd_claiming
, 0);
1699 spin_unlock(&bdev_lock
);
1702 * Block event polling for write claims if requested. Any
1703 * write holder makes the write_holder state stick until
1704 * all are released. This is good enough and tracking
1705 * individual writeable reference is too fragile given the
1706 * way @mode is used in blkdev_get/put().
1708 if (!res
&& (mode
& FMODE_WRITE
) && !bdev
->bd_write_holder
&&
1709 (disk
->flags
& GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE
)) {
1710 bdev
->bd_write_holder
= true;
1711 disk_block_events(disk
);
1714 mutex_unlock(&bdev
->bd_mutex
);
1720 EXPORT_SYMBOL(blkdev_get
);
1723 * blkdev_get_by_path - open a block device by name
1724 * @path: path to the block device to open
1725 * @mode: FMODE_* mask
1726 * @holder: exclusive holder identifier
1728 * Open the blockdevice described by the device file at @path. @mode
1729 * and @holder are identical to blkdev_get().
1731 * On success, the returned block_device has reference count of one.
1737 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1739 struct block_device
*blkdev_get_by_path(const char *path
, fmode_t mode
,
1742 struct block_device
*bdev
;
1746 if (mode
& FMODE_READ
)
1748 if (mode
& FMODE_WRITE
)
1750 bdev
= lookup_bdev(path
, perm
);
1754 err
= blkdev_get(bdev
, mode
, holder
);
1756 return ERR_PTR(err
);
1758 if ((mode
& FMODE_WRITE
) && bdev_read_only(bdev
)) {
1759 blkdev_put(bdev
, mode
);
1760 return ERR_PTR(-EACCES
);
1765 EXPORT_SYMBOL(blkdev_get_by_path
);
1768 * blkdev_get_by_dev - open a block device by device number
1769 * @dev: device number of block device to open
1770 * @mode: FMODE_* mask
1771 * @holder: exclusive holder identifier
1773 * Open the blockdevice described by device number @dev. @mode and
1774 * @holder are identical to blkdev_get().
1776 * Use it ONLY if you really do not have anything better - i.e. when
1777 * you are behind a truly sucky interface and all you are given is a
1778 * device number. _Never_ to be used for internal purposes. If you
1779 * ever need it - reconsider your API.
1781 * On success, the returned block_device has reference count of one.
1787 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1789 struct block_device
*blkdev_get_by_dev(dev_t dev
, fmode_t mode
, void *holder
)
1791 struct block_device
*bdev
;
1796 return ERR_PTR(-ENOMEM
);
1798 err
= blkdev_get(bdev
, mode
, holder
);
1800 return ERR_PTR(err
);
1804 EXPORT_SYMBOL(blkdev_get_by_dev
);
1806 static int blkdev_open(struct inode
* inode
, struct file
* filp
)
1808 struct block_device
*bdev
;
1811 * Preserve backwards compatibility and allow large file access
1812 * even if userspace doesn't ask for it explicitly. Some mkfs
1813 * binary needs it. We might want to drop this workaround
1814 * during an unstable branch.
1816 filp
->f_flags
|= O_LARGEFILE
;
1818 if (filp
->f_flags
& O_NDELAY
)
1819 filp
->f_mode
|= FMODE_NDELAY
;
1820 if (filp
->f_flags
& O_EXCL
)
1821 filp
->f_mode
|= FMODE_EXCL
;
1822 if ((filp
->f_flags
& O_ACCMODE
) == 3)
1823 filp
->f_mode
|= FMODE_WRITE_IOCTL
;
1825 bdev
= bd_acquire(inode
);
1830 * A negative i_writecount for bdev->bd_inode means that the bdev
1831 * or one of its paritions is mounted in a user namespace. Deny
1832 * writing for non-root in this case, otherwise an unprivileged
1833 * user can attack the kernel by modifying the backing store of a
1834 * mounted filesystem.
1836 if ((filp
->f_mode
& FMODE_WRITE
) &&
1837 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
) &&
1838 !atomic_inc_unless_negative(&bdev
->bd_inode
->i_writecount
)) {
1843 filp
->f_mapping
= bdev
->bd_inode
->i_mapping
;
1845 return blkdev_get(bdev
, filp
->f_mode
, filp
);
1848 static void __blkdev_put(struct block_device
*bdev
, fmode_t mode
, int for_part
)
1850 struct gendisk
*disk
= bdev
->bd_disk
;
1851 struct block_device
*victim
= NULL
;
1853 mutex_lock_nested(&bdev
->bd_mutex
, for_part
);
1855 bdev
->bd_part_count
--;
1857 if (!--bdev
->bd_openers
) {
1858 WARN_ON_ONCE(bdev
->bd_holders
);
1859 sync_blockdev(bdev
);
1862 bdev_write_inode(bdev
);
1864 * Detaching bdev inode from its wb in __destroy_inode()
1865 * is too late: the queue which embeds its bdi (along with
1866 * root wb) can be gone as soon as we put_disk() below.
1868 inode_detach_wb(bdev
->bd_inode
);
1870 if (bdev
->bd_contains
== bdev
) {
1871 if (disk
->fops
->release
)
1872 disk
->fops
->release(disk
, mode
);
1874 if (!bdev
->bd_openers
) {
1875 struct module
*owner
= disk
->fops
->owner
;
1877 disk_put_part(bdev
->bd_part
);
1878 bdev
->bd_part
= NULL
;
1879 bdev
->bd_disk
= NULL
;
1880 if (bdev
!= bdev
->bd_contains
)
1881 victim
= bdev
->bd_contains
;
1882 bdev
->bd_contains
= NULL
;
1887 mutex_unlock(&bdev
->bd_mutex
);
1890 __blkdev_put(victim
, mode
, 1);
1893 void blkdev_put(struct block_device
*bdev
, fmode_t mode
)
1895 mutex_lock(&bdev
->bd_mutex
);
1897 if (mode
& FMODE_EXCL
) {
1901 * Release a claim on the device. The holder fields
1902 * are protected with bdev_lock. bd_mutex is to
1903 * synchronize disk_holder unlinking.
1905 spin_lock(&bdev_lock
);
1907 WARN_ON_ONCE(--bdev
->bd_holders
< 0);
1908 WARN_ON_ONCE(--bdev
->bd_contains
->bd_holders
< 0);
1910 /* bd_contains might point to self, check in a separate step */
1911 if ((bdev_free
= !bdev
->bd_holders
))
1912 bdev
->bd_holder
= NULL
;
1913 if (!bdev
->bd_contains
->bd_holders
)
1914 bdev
->bd_contains
->bd_holder
= NULL
;
1916 spin_unlock(&bdev_lock
);
1919 * If this was the last claim, remove holder link and
1920 * unblock evpoll if it was a write holder.
1922 if (bdev_free
&& bdev
->bd_write_holder
) {
1923 disk_unblock_events(bdev
->bd_disk
);
1924 bdev
->bd_write_holder
= false;
1929 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1930 * event. This is to ensure detection of media removal commanded
1931 * from userland - e.g. eject(1).
1933 disk_flush_events(bdev
->bd_disk
, DISK_EVENT_MEDIA_CHANGE
);
1935 mutex_unlock(&bdev
->bd_mutex
);
1937 __blkdev_put(bdev
, mode
, 0);
1939 EXPORT_SYMBOL(blkdev_put
);
1941 static int blkdev_close(struct inode
* inode
, struct file
* filp
)
1943 struct block_device
*bdev
= I_BDEV(bdev_file_inode(filp
));
1944 if (filp
->f_mode
& FMODE_WRITE
&&
1945 !file_ns_capable(filp
, &init_user_ns
, CAP_SYS_ADMIN
))
1946 atomic_dec(&bdev
->bd_inode
->i_writecount
);
1947 blkdev_put(bdev
, filp
->f_mode
);
1951 static long block_ioctl(struct file
*file
, unsigned cmd
, unsigned long arg
)
1953 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
1954 fmode_t mode
= file
->f_mode
;
1957 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1958 * to updated it before every ioctl.
1960 if (file
->f_flags
& O_NDELAY
)
1961 mode
|= FMODE_NDELAY
;
1963 mode
&= ~FMODE_NDELAY
;
1965 return blkdev_ioctl(bdev
, mode
, cmd
, arg
);
1969 * Write data to the block device. Only intended for the block device itself
1970 * and the raw driver which basically is a fake block device.
1972 * Does not take i_mutex for the write and thus is not for general purpose
1975 ssize_t
blkdev_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1977 struct file
*file
= iocb
->ki_filp
;
1978 struct inode
*bd_inode
= bdev_file_inode(file
);
1979 loff_t size
= i_size_read(bd_inode
);
1980 struct blk_plug plug
;
1983 if (bdev_read_only(I_BDEV(bd_inode
)))
1986 if (!iov_iter_count(from
))
1989 if (iocb
->ki_pos
>= size
)
1992 iov_iter_truncate(from
, size
- iocb
->ki_pos
);
1994 blk_start_plug(&plug
);
1995 ret
= __generic_file_write_iter(iocb
, from
);
1997 ret
= generic_write_sync(iocb
, ret
);
1998 blk_finish_plug(&plug
);
2001 EXPORT_SYMBOL_GPL(blkdev_write_iter
);
2003 ssize_t
blkdev_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
2005 struct file
*file
= iocb
->ki_filp
;
2006 struct inode
*bd_inode
= bdev_file_inode(file
);
2007 loff_t size
= i_size_read(bd_inode
);
2008 loff_t pos
= iocb
->ki_pos
;
2014 iov_iter_truncate(to
, size
);
2015 return generic_file_read_iter(iocb
, to
);
2017 EXPORT_SYMBOL_GPL(blkdev_read_iter
);
2020 * Try to release a page associated with block device when the system
2021 * is under memory pressure.
2023 static int blkdev_releasepage(struct page
*page
, gfp_t wait
)
2025 struct super_block
*super
= BDEV_I(page
->mapping
->host
)->bdev
.bd_super
;
2027 if (super
&& super
->s_op
->bdev_try_to_free_page
)
2028 return super
->s_op
->bdev_try_to_free_page(super
, page
, wait
);
2030 return try_to_free_buffers(page
);
2033 static int blkdev_writepages(struct address_space
*mapping
,
2034 struct writeback_control
*wbc
)
2036 if (dax_mapping(mapping
)) {
2037 struct block_device
*bdev
= I_BDEV(mapping
->host
);
2039 return dax_writeback_mapping_range(mapping
, bdev
, wbc
);
2041 return generic_writepages(mapping
, wbc
);
2044 static const struct address_space_operations def_blk_aops
= {
2045 .readpage
= blkdev_readpage
,
2046 .readpages
= blkdev_readpages
,
2047 .writepage
= blkdev_writepage
,
2048 .write_begin
= blkdev_write_begin
,
2049 .write_end
= blkdev_write_end
,
2050 .writepages
= blkdev_writepages
,
2051 .releasepage
= blkdev_releasepage
,
2052 .direct_IO
= blkdev_direct_IO
,
2053 .is_dirty_writeback
= buffer_check_dirty_writeback
,
2056 #define BLKDEV_FALLOC_FL_SUPPORTED \
2057 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2058 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2060 static long blkdev_fallocate(struct file
*file
, int mode
, loff_t start
,
2063 struct block_device
*bdev
= I_BDEV(bdev_file_inode(file
));
2064 struct request_queue
*q
= bdev_get_queue(bdev
);
2065 struct address_space
*mapping
;
2066 loff_t end
= start
+ len
- 1;
2070 /* Fail if we don't recognize the flags. */
2071 if (mode
& ~BLKDEV_FALLOC_FL_SUPPORTED
)
2074 /* Don't go off the end of the device. */
2075 isize
= i_size_read(bdev
->bd_inode
);
2079 if (mode
& FALLOC_FL_KEEP_SIZE
) {
2080 len
= isize
- start
;
2081 end
= start
+ len
- 1;
2087 * Don't allow IO that isn't aligned to logical block size.
2089 if ((start
| len
) & (bdev_logical_block_size(bdev
) - 1))
2092 /* Invalidate the page cache, including dirty pages. */
2093 mapping
= bdev
->bd_inode
->i_mapping
;
2094 truncate_inode_pages_range(mapping
, start
, end
);
2097 case FALLOC_FL_ZERO_RANGE
:
2098 case FALLOC_FL_ZERO_RANGE
| FALLOC_FL_KEEP_SIZE
:
2099 error
= blkdev_issue_zeroout(bdev
, start
>> 9, len
>> 9,
2102 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
:
2103 /* Only punch if the device can do zeroing discard. */
2104 if (!blk_queue_discard(q
) || !q
->limits
.discard_zeroes_data
)
2106 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2109 case FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
| FALLOC_FL_NO_HIDE_STALE
:
2110 if (!blk_queue_discard(q
))
2112 error
= blkdev_issue_discard(bdev
, start
>> 9, len
>> 9,
2122 * Invalidate again; if someone wandered in and dirtied a page,
2123 * the caller will be given -EBUSY. The third argument is
2124 * inclusive, so the rounding here is safe.
2126 return invalidate_inode_pages2_range(mapping
,
2127 start
>> PAGE_SHIFT
,
2131 const struct file_operations def_blk_fops
= {
2132 .open
= blkdev_open
,
2133 .release
= blkdev_close
,
2134 .llseek
= block_llseek
,
2135 .read_iter
= blkdev_read_iter
,
2136 .write_iter
= blkdev_write_iter
,
2137 .mmap
= generic_file_mmap
,
2138 .fsync
= blkdev_fsync
,
2139 .unlocked_ioctl
= block_ioctl
,
2140 #ifdef CONFIG_COMPAT
2141 .compat_ioctl
= compat_blkdev_ioctl
,
2143 .splice_read
= generic_file_splice_read
,
2144 .splice_write
= iter_file_splice_write
,
2145 .fallocate
= blkdev_fallocate
,
2148 int ioctl_by_bdev(struct block_device
*bdev
, unsigned cmd
, unsigned long arg
)
2151 mm_segment_t old_fs
= get_fs();
2153 res
= blkdev_ioctl(bdev
, 0, cmd
, arg
);
2158 EXPORT_SYMBOL(ioctl_by_bdev
);
2161 * lookup_bdev - lookup a struct block_device by name
2162 * @pathname: special file representing the block device
2163 * @mask: rights to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
2165 * Get a reference to the blockdevice at @pathname in the current
2166 * namespace if possible and return it. Return ERR_PTR(error)
2167 * otherwise. If @mask is non-zero, check for access rights to the
2168 * inode at @pathname.
2170 struct block_device
*lookup_bdev(const char *pathname
, int mask
)
2172 struct block_device
*bdev
;
2173 struct inode
*inode
;
2177 if (!pathname
|| !*pathname
)
2178 return ERR_PTR(-EINVAL
);
2180 error
= kern_path(pathname
, LOOKUP_FOLLOW
, &path
);
2182 return ERR_PTR(error
);
2184 inode
= d_backing_inode(path
.dentry
);
2185 if (mask
!= 0 && !capable(CAP_SYS_ADMIN
)) {
2186 error
= __inode_permission(inode
, mask
);
2191 if (!S_ISBLK(inode
->i_mode
))
2194 if (!may_open_dev(&path
))
2197 bdev
= bd_acquire(inode
);
2204 bdev
= ERR_PTR(error
);
2207 EXPORT_SYMBOL(lookup_bdev
);
2209 int __invalidate_device(struct block_device
*bdev
, bool kill_dirty
)
2211 struct super_block
*sb
= get_super(bdev
);
2216 * no need to lock the super, get_super holds the
2217 * read mutex so the filesystem cannot go away
2218 * under us (->put_super runs with the write lock
2221 shrink_dcache_sb(sb
);
2222 res
= invalidate_inodes(sb
, kill_dirty
);
2225 invalidate_bdev(bdev
);
2228 EXPORT_SYMBOL(__invalidate_device
);
2230 void iterate_bdevs(void (*func
)(struct block_device
*, void *), void *arg
)
2232 struct inode
*inode
, *old_inode
= NULL
;
2234 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2235 list_for_each_entry(inode
, &blockdev_superblock
->s_inodes
, i_sb_list
) {
2236 struct address_space
*mapping
= inode
->i_mapping
;
2237 struct block_device
*bdev
;
2239 spin_lock(&inode
->i_lock
);
2240 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
) ||
2241 mapping
->nrpages
== 0) {
2242 spin_unlock(&inode
->i_lock
);
2246 spin_unlock(&inode
->i_lock
);
2247 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);
2249 * We hold a reference to 'inode' so it couldn't have been
2250 * removed from s_inodes list while we dropped the
2251 * s_inode_list_lock We cannot iput the inode now as we can
2252 * be holding the last reference and we cannot iput it under
2253 * s_inode_list_lock. So we keep the reference and iput it
2258 bdev
= I_BDEV(inode
);
2260 mutex_lock(&bdev
->bd_mutex
);
2261 if (bdev
->bd_openers
)
2263 mutex_unlock(&bdev
->bd_mutex
);
2265 spin_lock(&blockdev_superblock
->s_inode_list_lock
);
2267 spin_unlock(&blockdev_superblock
->s_inode_list_lock
);