2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
81 #include <linux/uaccess.h>
83 static DEFINE_IDR(loop_index_idr
);
84 static DEFINE_MUTEX(loop_index_mutex
);
87 static int part_shift
;
89 static int transfer_xor(struct loop_device
*lo
, int cmd
,
90 struct page
*raw_page
, unsigned raw_off
,
91 struct page
*loop_page
, unsigned loop_off
,
92 int size
, sector_t real_block
)
94 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
95 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
107 key
= lo
->lo_encrypt_key
;
108 keysize
= lo
->lo_encrypt_key_size
;
109 for (i
= 0; i
< size
; i
++)
110 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
112 kunmap_atomic(loop_buf
);
113 kunmap_atomic(raw_buf
);
118 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
120 if (unlikely(info
->lo_encrypt_key_size
<= 0))
125 static struct loop_func_table none_funcs
= {
126 .number
= LO_CRYPT_NONE
,
129 static struct loop_func_table xor_funcs
= {
130 .number
= LO_CRYPT_XOR
,
131 .transfer
= transfer_xor
,
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
141 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
145 /* Compute loopsize in bytes */
146 loopsize
= i_size_read(file
->f_mapping
->host
);
149 /* offset is beyond i_size, weird but possible */
153 if (sizelimit
> 0 && sizelimit
< loopsize
)
154 loopsize
= sizelimit
;
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
159 return loopsize
>> 9;
162 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
164 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
167 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
169 struct file
*file
= lo
->lo_backing_file
;
170 struct address_space
*mapping
= file
->f_mapping
;
171 struct inode
*inode
= mapping
->host
;
172 unsigned short sb_bsize
= 0;
173 unsigned dio_align
= 0;
176 if (inode
->i_sb
->s_bdev
) {
177 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
178 dio_align
= sb_bsize
- 1;
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane applications should be PAGE_SIZE aligned
192 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
193 !(lo
->lo_offset
& dio_align
) &&
194 mapping
->a_ops
->direct_IO
&&
203 if (lo
->use_dio
== use_dio
)
206 /* flush dirty pages before changing direct IO */
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
214 blk_mq_freeze_queue(lo
->lo_queue
);
215 lo
->use_dio
= use_dio
;
217 queue_flag_clear_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
218 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
220 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
221 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
223 blk_mq_unfreeze_queue(lo
->lo_queue
);
227 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
229 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
230 sector_t x
= (sector_t
)size
;
231 struct block_device
*bdev
= lo
->lo_device
;
233 if (unlikely((loff_t
)x
!= size
))
235 if (lo
->lo_offset
!= offset
)
236 lo
->lo_offset
= offset
;
237 if (lo
->lo_sizelimit
!= sizelimit
)
238 lo
->lo_sizelimit
= sizelimit
;
239 set_capacity(lo
->lo_disk
, x
);
240 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
241 /* let user-space know about the new size */
242 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
247 lo_do_transfer(struct loop_device
*lo
, int cmd
,
248 struct page
*rpage
, unsigned roffs
,
249 struct page
*lpage
, unsigned loffs
,
250 int size
, sector_t rblock
)
254 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
258 printk_ratelimited(KERN_ERR
259 "loop: Transfer error at byte offset %llu, length %i.\n",
260 (unsigned long long)rblock
<< 9, size
);
264 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
269 iov_iter_bvec(&i
, ITER_BVEC
| WRITE
, bvec
, 1, bvec
->bv_len
);
271 file_start_write(file
);
272 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
273 file_end_write(file
);
275 if (likely(bw
== bvec
->bv_len
))
278 printk_ratelimited(KERN_ERR
279 "loop: Write error at byte offset %llu, length %i.\n",
280 (unsigned long long)*ppos
, bvec
->bv_len
);
286 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
290 struct req_iterator iter
;
293 rq_for_each_segment(bvec
, rq
, iter
) {
294 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
304 * This is the slow, transforming version that needs to double buffer the
305 * data as it cannot do the transformations in place without having direct
306 * access to the destination pages of the backing file.
308 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
311 struct bio_vec bvec
, b
;
312 struct req_iterator iter
;
316 page
= alloc_page(GFP_NOIO
);
320 rq_for_each_segment(bvec
, rq
, iter
) {
321 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
322 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
328 b
.bv_len
= bvec
.bv_len
;
329 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
338 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
342 struct req_iterator iter
;
346 rq_for_each_segment(bvec
, rq
, iter
) {
347 iov_iter_bvec(&i
, ITER_BVEC
, &bvec
, 1, bvec
.bv_len
);
348 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
352 flush_dcache_page(bvec
.bv_page
);
354 if (len
!= bvec
.bv_len
) {
357 __rq_for_each_bio(bio
, rq
)
367 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
370 struct bio_vec bvec
, b
;
371 struct req_iterator iter
;
377 page
= alloc_page(GFP_NOIO
);
381 rq_for_each_segment(bvec
, rq
, iter
) {
386 b
.bv_len
= bvec
.bv_len
;
388 iov_iter_bvec(&i
, ITER_BVEC
, &b
, 1, b
.bv_len
);
389 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
395 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
396 bvec
.bv_offset
, len
, offset
>> 9);
400 flush_dcache_page(bvec
.bv_page
);
402 if (len
!= bvec
.bv_len
) {
405 __rq_for_each_bio(bio
, rq
)
417 static int lo_discard(struct loop_device
*lo
, struct request
*rq
, loff_t pos
)
420 * We use punch hole to reclaim the free space used by the
421 * image a.k.a. discard. However we do not support discard if
422 * encryption is enabled, because it may give an attacker
423 * useful information.
425 struct file
*file
= lo
->lo_backing_file
;
426 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
429 if ((!file
->f_op
->fallocate
) || lo
->lo_encrypt_key_size
) {
434 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
435 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
441 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
443 struct file
*file
= lo
->lo_backing_file
;
444 int ret
= vfs_fsync(file
, 0);
445 if (unlikely(ret
&& ret
!= -EINVAL
))
451 static void lo_complete_rq(struct request
*rq
)
453 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
455 if (unlikely(req_op(cmd
->rq
) == REQ_OP_READ
&& cmd
->use_aio
&&
456 cmd
->ret
>= 0 && cmd
->ret
< blk_rq_bytes(cmd
->rq
))) {
457 struct bio
*bio
= cmd
->rq
->bio
;
459 bio_advance(bio
, cmd
->ret
);
463 blk_mq_end_request(rq
, cmd
->ret
< 0 ? BLK_STS_IOERR
: BLK_STS_OK
);
466 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
468 if (!atomic_dec_and_test(&cmd
->ref
))
472 blk_mq_complete_request(cmd
->rq
);
475 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
477 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
482 lo_rw_aio_do_completion(cmd
);
485 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
488 struct iov_iter iter
;
489 struct bio_vec
*bvec
;
490 struct request
*rq
= cmd
->rq
;
491 struct bio
*bio
= rq
->bio
;
492 struct file
*file
= lo
->lo_backing_file
;
497 if (rq
->bio
!= rq
->biotail
) {
498 struct req_iterator iter
;
501 __rq_for_each_bio(bio
, rq
)
502 segments
+= bio_segments(bio
);
503 bvec
= kmalloc(sizeof(struct bio_vec
) * segments
, GFP_NOIO
);
509 * The bios of the request may be started from the middle of
510 * the 'bvec' because of bio splitting, so we can't directly
511 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
512 * API will take care of all details for us.
514 rq_for_each_segment(tmp
, rq
, iter
) {
522 * Same here, this bio may be started from the middle of the
523 * 'bvec' because of bio splitting, so offset from the bvec
524 * must be passed to iov iterator
526 offset
= bio
->bi_iter
.bi_bvec_done
;
527 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
528 segments
= bio_segments(bio
);
530 atomic_set(&cmd
->ref
, 2);
532 iov_iter_bvec(&iter
, ITER_BVEC
| rw
, bvec
,
533 segments
, blk_rq_bytes(rq
));
534 iter
.iov_offset
= offset
;
536 cmd
->iocb
.ki_pos
= pos
;
537 cmd
->iocb
.ki_filp
= file
;
538 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
539 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
541 kthread_associate_blkcg(cmd
->css
);
544 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
546 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
548 lo_rw_aio_do_completion(cmd
);
549 kthread_associate_blkcg(NULL
);
551 if (ret
!= -EIOCBQUEUED
)
552 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
556 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
558 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
559 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
562 * lo_write_simple and lo_read_simple should have been covered
563 * by io submit style function like lo_rw_aio(), one blocker
564 * is that lo_read_simple() need to call flush_dcache_page after
565 * the page is written from kernel, and it isn't easy to handle
566 * this in io submit style function which submits all segments
567 * of the req at one time. And direct read IO doesn't need to
568 * run flush_dcache_page().
570 switch (req_op(rq
)) {
572 return lo_req_flush(lo
, rq
);
574 case REQ_OP_WRITE_ZEROES
:
575 return lo_discard(lo
, rq
, pos
);
578 return lo_write_transfer(lo
, rq
, pos
);
579 else if (cmd
->use_aio
)
580 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
582 return lo_write_simple(lo
, rq
, pos
);
585 return lo_read_transfer(lo
, rq
, pos
);
586 else if (cmd
->use_aio
)
587 return lo_rw_aio(lo
, cmd
, pos
, READ
);
589 return lo_read_simple(lo
, rq
, pos
);
597 static inline void loop_update_dio(struct loop_device
*lo
)
599 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
603 static struct file
*loop_real_file(struct file
*file
)
605 struct file
*f
= NULL
;
607 if (file
->f_path
.dentry
->d_sb
->s_op
->real_loop
)
608 f
= file
->f_path
.dentry
->d_sb
->s_op
->real_loop(file
);
612 static void loop_reread_partitions(struct loop_device
*lo
,
613 struct block_device
*bdev
)
618 * bd_mutex has been held already in release path, so don't
619 * acquire it if this function is called in such case.
621 * If the reread partition isn't from release path, lo_refcnt
622 * must be at least one and it can only become zero when the
623 * current holder is released.
625 if (!atomic_read(&lo
->lo_refcnt
))
626 rc
= __blkdev_reread_part(bdev
);
628 rc
= blkdev_reread_part(bdev
);
630 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
631 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
635 * loop_change_fd switched the backing store of a loopback device to
636 * a new file. This is useful for operating system installers to free up
637 * the original file and in High Availability environments to switch to
638 * an alternative location for the content in case of server meltdown.
639 * This can only work if the loop device is used read-only, and if the
640 * new backing store is the same size and type as the old backing store.
642 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
645 struct file
*file
, *old_file
;
646 struct file
*f
, *virt_file
= NULL
, *old_virt_file
;
651 if (lo
->lo_state
!= Lo_bound
)
654 /* the loop device has to be read-only */
656 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
663 f
= loop_real_file(file
);
670 inode
= file
->f_mapping
->host
;
671 old_file
= lo
->lo_backing_file
;
672 old_virt_file
= lo
->lo_backing_virt_file
;
676 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
679 /* size of the new backing store needs to be the same */
680 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
684 blk_mq_freeze_queue(lo
->lo_queue
);
685 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
686 lo
->lo_backing_file
= file
;
687 lo
->lo_backing_virt_file
= virt_file
;
688 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
689 mapping_set_gfp_mask(file
->f_mapping
,
690 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
692 blk_mq_unfreeze_queue(lo
->lo_queue
);
697 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
698 loop_reread_partitions(lo
, bdev
);
709 static inline int is_loop_device(struct file
*file
)
711 struct inode
*i
= file
->f_mapping
->host
;
713 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
718 * no get/put for file.
720 struct file
*loop_backing_file(struct super_block
*sb
)
723 struct loop_device
*l
;
726 if (MAJOR(sb
->s_dev
) == LOOP_MAJOR
) {
727 l
= sb
->s_bdev
->bd_disk
->private_data
;
728 ret
= l
->lo_backing_file
;
732 EXPORT_SYMBOL_GPL(loop_backing_file
);
734 /* loop sysfs attributes */
736 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
737 ssize_t (*callback
)(struct loop_device
*, char *))
739 struct gendisk
*disk
= dev_to_disk(dev
);
740 struct loop_device
*lo
= disk
->private_data
;
742 return callback(lo
, page
);
745 #define LOOP_ATTR_RO(_name) \
746 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
747 static ssize_t loop_attr_do_show_##_name(struct device *d, \
748 struct device_attribute *attr, char *b) \
750 return loop_attr_show(d, b, loop_attr_##_name##_show); \
752 static struct device_attribute loop_attr_##_name = \
753 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
755 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
760 spin_lock_irq(&lo
->lo_lock
);
761 if (lo
->lo_backing_file
)
762 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
763 spin_unlock_irq(&lo
->lo_lock
);
765 if (IS_ERR_OR_NULL(p
))
769 memmove(buf
, p
, ret
);
777 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
779 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
782 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
784 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
787 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
789 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
791 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
794 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
796 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
798 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
801 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
803 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
805 return sprintf(buf
, "%s\n", dio
? "1" : "0");
808 LOOP_ATTR_RO(backing_file
);
809 LOOP_ATTR_RO(offset
);
810 LOOP_ATTR_RO(sizelimit
);
811 LOOP_ATTR_RO(autoclear
);
812 LOOP_ATTR_RO(partscan
);
815 static struct attribute
*loop_attrs
[] = {
816 &loop_attr_backing_file
.attr
,
817 &loop_attr_offset
.attr
,
818 &loop_attr_sizelimit
.attr
,
819 &loop_attr_autoclear
.attr
,
820 &loop_attr_partscan
.attr
,
825 static struct attribute_group loop_attribute_group
= {
830 static int loop_sysfs_init(struct loop_device
*lo
)
832 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
833 &loop_attribute_group
);
836 static void loop_sysfs_exit(struct loop_device
*lo
)
838 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
839 &loop_attribute_group
);
842 static void loop_config_discard(struct loop_device
*lo
)
844 struct file
*file
= lo
->lo_backing_file
;
845 struct inode
*inode
= file
->f_mapping
->host
;
846 struct request_queue
*q
= lo
->lo_queue
;
849 * We use punch hole to reclaim the free space used by the
850 * image a.k.a. discard. However we do not support discard if
851 * encryption is enabled, because it may give an attacker
852 * useful information.
854 if ((!file
->f_op
->fallocate
) ||
855 lo
->lo_encrypt_key_size
) {
856 q
->limits
.discard_granularity
= 0;
857 q
->limits
.discard_alignment
= 0;
858 blk_queue_max_discard_sectors(q
, 0);
859 blk_queue_max_write_zeroes_sectors(q
, 0);
860 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
864 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
865 q
->limits
.discard_alignment
= 0;
867 blk_queue_max_discard_sectors(q
, UINT_MAX
>> 9);
868 blk_queue_max_write_zeroes_sectors(q
, UINT_MAX
>> 9);
869 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
872 static void loop_unprepare_queue(struct loop_device
*lo
)
874 kthread_flush_worker(&lo
->worker
);
875 kthread_stop(lo
->worker_task
);
878 static int loop_kthread_worker_fn(void *worker_ptr
)
880 current
->flags
|= PF_LESS_THROTTLE
;
881 return kthread_worker_fn(worker_ptr
);
884 static int loop_prepare_queue(struct loop_device
*lo
)
886 kthread_init_worker(&lo
->worker
);
887 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
888 &lo
->worker
, "loop%d", lo
->lo_number
);
889 if (IS_ERR(lo
->worker_task
))
891 set_user_nice(lo
->worker_task
, MIN_NICE
);
895 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
896 struct block_device
*bdev
, unsigned int arg
)
898 struct file
*file
, *f
, *virt_file
= NULL
;
900 struct address_space
*mapping
;
905 /* This is safe, since we have a reference from open(). */
906 __module_get(THIS_MODULE
);
912 f
= loop_real_file(file
);
920 if (lo
->lo_state
!= Lo_unbound
)
923 /* Avoid recursion */
925 while (is_loop_device(f
)) {
926 struct loop_device
*l
;
928 if (f
->f_mapping
->host
->i_bdev
== bdev
)
931 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
932 if (l
->lo_state
== Lo_unbound
) {
936 f
= l
->lo_backing_file
;
939 mapping
= file
->f_mapping
;
940 inode
= mapping
->host
;
943 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
946 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
947 !file
->f_op
->write_iter
)
948 lo_flags
|= LO_FLAGS_READ_ONLY
;
951 size
= get_loop_size(lo
, file
);
952 if ((loff_t
)(sector_t
)size
!= size
)
954 error
= loop_prepare_queue(lo
);
960 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
963 lo
->lo_device
= bdev
;
964 lo
->lo_flags
= lo_flags
;
965 lo
->lo_backing_file
= file
;
966 lo
->lo_backing_virt_file
= virt_file
;
969 lo
->lo_sizelimit
= 0;
970 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
971 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
973 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
974 blk_queue_write_cache(lo
->lo_queue
, true, false);
977 set_capacity(lo
->lo_disk
, size
);
978 bd_set_size(bdev
, size
<< 9);
980 /* let user-space know about the new size */
981 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
983 set_blocksize(bdev
, S_ISBLK(inode
->i_mode
) ?
984 block_size(inode
->i_bdev
) : PAGE_SIZE
);
986 lo
->lo_state
= Lo_bound
;
988 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
989 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
990 loop_reread_partitions(lo
, bdev
);
992 /* Grab the block_device to prevent its destruction after we
993 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
1003 /* This is safe: open() is still holding a reference. */
1004 module_put(THIS_MODULE
);
1009 loop_release_xfer(struct loop_device
*lo
)
1012 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1016 err
= xfer
->release(lo
);
1017 lo
->transfer
= NULL
;
1018 lo
->lo_encryption
= NULL
;
1019 module_put(xfer
->owner
);
1025 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1026 const struct loop_info64
*i
)
1031 struct module
*owner
= xfer
->owner
;
1033 if (!try_module_get(owner
))
1036 err
= xfer
->init(lo
, i
);
1040 lo
->lo_encryption
= xfer
;
1045 static int loop_clr_fd(struct loop_device
*lo
)
1047 struct file
*filp
= lo
->lo_backing_file
;
1048 struct file
*virt_filp
= lo
->lo_backing_virt_file
;
1049 gfp_t gfp
= lo
->old_gfp_mask
;
1050 struct block_device
*bdev
= lo
->lo_device
;
1052 if (lo
->lo_state
!= Lo_bound
)
1056 * If we've explicitly asked to tear down the loop device,
1057 * and it has an elevated reference count, set it for auto-teardown when
1058 * the last reference goes away. This stops $!~#$@ udev from
1059 * preventing teardown because it decided that it needs to run blkid on
1060 * the loopback device whenever they appear. xfstests is notorious for
1061 * failing tests because blkid via udev races with a losetup
1062 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1063 * command to fail with EBUSY.
1065 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1066 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1067 mutex_unlock(&lo
->lo_ctl_mutex
);
1074 /* freeze request queue during the transition */
1075 blk_mq_freeze_queue(lo
->lo_queue
);
1077 spin_lock_irq(&lo
->lo_lock
);
1078 lo
->lo_state
= Lo_rundown
;
1079 lo
->lo_backing_file
= NULL
;
1080 lo
->lo_backing_virt_file
= NULL
;
1081 spin_unlock_irq(&lo
->lo_lock
);
1083 loop_release_xfer(lo
);
1084 lo
->transfer
= NULL
;
1086 lo
->lo_device
= NULL
;
1087 lo
->lo_encryption
= NULL
;
1089 lo
->lo_sizelimit
= 0;
1090 lo
->lo_encrypt_key_size
= 0;
1091 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1092 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1093 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1094 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1095 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1096 blk_queue_io_min(lo
->lo_queue
, 512);
1099 invalidate_bdev(bdev
);
1101 set_capacity(lo
->lo_disk
, 0);
1102 loop_sysfs_exit(lo
);
1104 bd_set_size(bdev
, 0);
1105 /* let user-space know about this change */
1106 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1108 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1109 lo
->lo_state
= Lo_unbound
;
1110 /* This is safe: open() is still holding a reference. */
1111 module_put(THIS_MODULE
);
1112 blk_mq_unfreeze_queue(lo
->lo_queue
);
1114 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1115 loop_reread_partitions(lo
, bdev
);
1118 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1119 loop_unprepare_queue(lo
);
1120 mutex_unlock(&lo
->lo_ctl_mutex
);
1122 * Need not hold lo_ctl_mutex to fput backing file.
1123 * Calling fput holding lo_ctl_mutex triggers a circular
1124 * lock dependency possibility warning as fput can take
1125 * bd_mutex which is usually taken before lo_ctl_mutex.
1134 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1137 struct loop_func_table
*xfer
;
1138 kuid_t uid
= current_uid();
1140 if (lo
->lo_encrypt_key_size
&&
1141 !uid_eq(lo
->lo_key_owner
, uid
) &&
1142 !capable(CAP_SYS_ADMIN
))
1144 if (lo
->lo_state
!= Lo_bound
)
1146 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1149 /* I/O need to be drained during transfer transition */
1150 blk_mq_freeze_queue(lo
->lo_queue
);
1152 err
= loop_release_xfer(lo
);
1156 if (info
->lo_encrypt_type
) {
1157 unsigned int type
= info
->lo_encrypt_type
;
1159 if (type
>= MAX_LO_CRYPT
) {
1163 xfer
= xfer_funcs
[type
];
1171 err
= loop_init_xfer(lo
, xfer
, info
);
1175 if (lo
->lo_offset
!= info
->lo_offset
||
1176 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1177 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
)) {
1183 loop_config_discard(lo
);
1185 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1186 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1187 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1188 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1192 lo
->transfer
= xfer
->transfer
;
1193 lo
->ioctl
= xfer
->ioctl
;
1195 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1196 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1197 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1199 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1200 lo
->lo_init
[0] = info
->lo_init
[0];
1201 lo
->lo_init
[1] = info
->lo_init
[1];
1202 if (info
->lo_encrypt_key_size
) {
1203 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1204 info
->lo_encrypt_key_size
);
1205 lo
->lo_key_owner
= uid
;
1208 /* update dio if lo_offset or transfer is changed */
1209 __loop_update_dio(lo
, lo
->use_dio
);
1212 blk_mq_unfreeze_queue(lo
->lo_queue
);
1214 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1215 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1216 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1217 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1218 loop_reread_partitions(lo
, lo
->lo_device
);
1225 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1227 struct file
*file
= lo
->lo_backing_file
;
1231 if (lo
->lo_state
!= Lo_bound
)
1233 error
= vfs_getattr(&file
->f_path
, &stat
,
1234 STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1237 memset(info
, 0, sizeof(*info
));
1238 info
->lo_number
= lo
->lo_number
;
1239 info
->lo_device
= huge_encode_dev(stat
.dev
);
1240 info
->lo_inode
= stat
.ino
;
1241 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1242 info
->lo_offset
= lo
->lo_offset
;
1243 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1244 info
->lo_flags
= lo
->lo_flags
;
1245 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1246 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1247 info
->lo_encrypt_type
=
1248 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1249 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1250 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1251 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1252 lo
->lo_encrypt_key_size
);
1258 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1260 memset(info64
, 0, sizeof(*info64
));
1261 info64
->lo_number
= info
->lo_number
;
1262 info64
->lo_device
= info
->lo_device
;
1263 info64
->lo_inode
= info
->lo_inode
;
1264 info64
->lo_rdevice
= info
->lo_rdevice
;
1265 info64
->lo_offset
= info
->lo_offset
;
1266 info64
->lo_sizelimit
= 0;
1267 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1268 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1269 info64
->lo_flags
= info
->lo_flags
;
1270 info64
->lo_init
[0] = info
->lo_init
[0];
1271 info64
->lo_init
[1] = info
->lo_init
[1];
1272 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1273 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1275 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1276 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1280 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1282 memset(info
, 0, sizeof(*info
));
1283 info
->lo_number
= info64
->lo_number
;
1284 info
->lo_device
= info64
->lo_device
;
1285 info
->lo_inode
= info64
->lo_inode
;
1286 info
->lo_rdevice
= info64
->lo_rdevice
;
1287 info
->lo_offset
= info64
->lo_offset
;
1288 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1289 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1290 info
->lo_flags
= info64
->lo_flags
;
1291 info
->lo_init
[0] = info64
->lo_init
[0];
1292 info
->lo_init
[1] = info64
->lo_init
[1];
1293 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1294 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1296 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1297 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1299 /* error in case values were truncated */
1300 if (info
->lo_device
!= info64
->lo_device
||
1301 info
->lo_rdevice
!= info64
->lo_rdevice
||
1302 info
->lo_inode
!= info64
->lo_inode
||
1303 info
->lo_offset
!= info64
->lo_offset
)
1310 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1312 struct loop_info info
;
1313 struct loop_info64 info64
;
1315 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1317 loop_info64_from_old(&info
, &info64
);
1318 return loop_set_status(lo
, &info64
);
1322 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1324 struct loop_info64 info64
;
1326 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1328 return loop_set_status(lo
, &info64
);
1332 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1333 struct loop_info info
;
1334 struct loop_info64 info64
;
1340 err
= loop_get_status(lo
, &info64
);
1342 err
= loop_info64_to_old(&info64
, &info
);
1343 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1350 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1351 struct loop_info64 info64
;
1357 err
= loop_get_status(lo
, &info64
);
1358 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1364 static int loop_set_capacity(struct loop_device
*lo
)
1366 if (unlikely(lo
->lo_state
!= Lo_bound
))
1369 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1372 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1375 if (lo
->lo_state
!= Lo_bound
)
1378 __loop_update_dio(lo
, !!arg
);
1379 if (lo
->use_dio
== !!arg
)
1386 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1388 if (lo
->lo_state
!= Lo_bound
)
1391 if (arg
< 512 || arg
> PAGE_SIZE
|| !is_power_of_2(arg
))
1394 blk_mq_freeze_queue(lo
->lo_queue
);
1396 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1397 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1398 blk_queue_io_min(lo
->lo_queue
, arg
);
1399 loop_update_dio(lo
);
1401 blk_mq_unfreeze_queue(lo
->lo_queue
);
1406 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1407 unsigned int cmd
, unsigned long arg
)
1409 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1412 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1415 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1417 case LOOP_CHANGE_FD
:
1418 err
= loop_change_fd(lo
, bdev
, arg
);
1421 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1422 err
= loop_clr_fd(lo
);
1426 case LOOP_SET_STATUS
:
1428 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1429 err
= loop_set_status_old(lo
,
1430 (struct loop_info __user
*)arg
);
1432 case LOOP_GET_STATUS
:
1433 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1435 case LOOP_SET_STATUS64
:
1437 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1438 err
= loop_set_status64(lo
,
1439 (struct loop_info64 __user
*) arg
);
1441 case LOOP_GET_STATUS64
:
1442 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1444 case LOOP_SET_CAPACITY
:
1446 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1447 err
= loop_set_capacity(lo
);
1449 case LOOP_SET_DIRECT_IO
:
1451 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1452 err
= loop_set_dio(lo
, arg
);
1454 case LOOP_SET_BLOCK_SIZE
:
1456 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1457 err
= loop_set_block_size(lo
, arg
);
1460 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1462 mutex_unlock(&lo
->lo_ctl_mutex
);
1468 #ifdef CONFIG_COMPAT
1469 struct compat_loop_info
{
1470 compat_int_t lo_number
; /* ioctl r/o */
1471 compat_dev_t lo_device
; /* ioctl r/o */
1472 compat_ulong_t lo_inode
; /* ioctl r/o */
1473 compat_dev_t lo_rdevice
; /* ioctl r/o */
1474 compat_int_t lo_offset
;
1475 compat_int_t lo_encrypt_type
;
1476 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1477 compat_int_t lo_flags
; /* ioctl r/o */
1478 char lo_name
[LO_NAME_SIZE
];
1479 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1480 compat_ulong_t lo_init
[2];
1485 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1486 * - noinlined to reduce stack space usage in main part of driver
1489 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1490 struct loop_info64
*info64
)
1492 struct compat_loop_info info
;
1494 if (copy_from_user(&info
, arg
, sizeof(info
)))
1497 memset(info64
, 0, sizeof(*info64
));
1498 info64
->lo_number
= info
.lo_number
;
1499 info64
->lo_device
= info
.lo_device
;
1500 info64
->lo_inode
= info
.lo_inode
;
1501 info64
->lo_rdevice
= info
.lo_rdevice
;
1502 info64
->lo_offset
= info
.lo_offset
;
1503 info64
->lo_sizelimit
= 0;
1504 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1505 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1506 info64
->lo_flags
= info
.lo_flags
;
1507 info64
->lo_init
[0] = info
.lo_init
[0];
1508 info64
->lo_init
[1] = info
.lo_init
[1];
1509 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1510 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1512 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1513 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1518 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1519 * - noinlined to reduce stack space usage in main part of driver
1522 loop_info64_to_compat(const struct loop_info64
*info64
,
1523 struct compat_loop_info __user
*arg
)
1525 struct compat_loop_info info
;
1527 memset(&info
, 0, sizeof(info
));
1528 info
.lo_number
= info64
->lo_number
;
1529 info
.lo_device
= info64
->lo_device
;
1530 info
.lo_inode
= info64
->lo_inode
;
1531 info
.lo_rdevice
= info64
->lo_rdevice
;
1532 info
.lo_offset
= info64
->lo_offset
;
1533 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1534 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1535 info
.lo_flags
= info64
->lo_flags
;
1536 info
.lo_init
[0] = info64
->lo_init
[0];
1537 info
.lo_init
[1] = info64
->lo_init
[1];
1538 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1539 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1541 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1542 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1544 /* error in case values were truncated */
1545 if (info
.lo_device
!= info64
->lo_device
||
1546 info
.lo_rdevice
!= info64
->lo_rdevice
||
1547 info
.lo_inode
!= info64
->lo_inode
||
1548 info
.lo_offset
!= info64
->lo_offset
||
1549 info
.lo_init
[0] != info64
->lo_init
[0] ||
1550 info
.lo_init
[1] != info64
->lo_init
[1])
1553 if (copy_to_user(arg
, &info
, sizeof(info
)))
1559 loop_set_status_compat(struct loop_device
*lo
,
1560 const struct compat_loop_info __user
*arg
)
1562 struct loop_info64 info64
;
1565 ret
= loop_info64_from_compat(arg
, &info64
);
1568 return loop_set_status(lo
, &info64
);
1572 loop_get_status_compat(struct loop_device
*lo
,
1573 struct compat_loop_info __user
*arg
)
1575 struct loop_info64 info64
;
1581 err
= loop_get_status(lo
, &info64
);
1583 err
= loop_info64_to_compat(&info64
, arg
);
1587 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1588 unsigned int cmd
, unsigned long arg
)
1590 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1594 case LOOP_SET_STATUS
:
1595 mutex_lock(&lo
->lo_ctl_mutex
);
1596 err
= loop_set_status_compat(
1597 lo
, (const struct compat_loop_info __user
*) arg
);
1598 mutex_unlock(&lo
->lo_ctl_mutex
);
1600 case LOOP_GET_STATUS
:
1601 mutex_lock(&lo
->lo_ctl_mutex
);
1602 err
= loop_get_status_compat(
1603 lo
, (struct compat_loop_info __user
*) arg
);
1604 mutex_unlock(&lo
->lo_ctl_mutex
);
1606 case LOOP_SET_CAPACITY
:
1608 case LOOP_GET_STATUS64
:
1609 case LOOP_SET_STATUS64
:
1610 arg
= (unsigned long) compat_ptr(arg
);
1612 case LOOP_CHANGE_FD
:
1613 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1623 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1625 struct loop_device
*lo
;
1628 mutex_lock(&loop_index_mutex
);
1629 lo
= bdev
->bd_disk
->private_data
;
1635 atomic_inc(&lo
->lo_refcnt
);
1637 mutex_unlock(&loop_index_mutex
);
1641 static void __lo_release(struct loop_device
*lo
)
1645 if (atomic_dec_return(&lo
->lo_refcnt
))
1648 mutex_lock(&lo
->lo_ctl_mutex
);
1649 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1651 * In autoclear mode, stop the loop thread
1652 * and remove configuration after last close.
1654 err
= loop_clr_fd(lo
);
1657 } else if (lo
->lo_state
== Lo_bound
) {
1659 * Otherwise keep thread (if running) and config,
1660 * but flush possible ongoing bios in thread.
1662 blk_mq_freeze_queue(lo
->lo_queue
);
1663 blk_mq_unfreeze_queue(lo
->lo_queue
);
1666 mutex_unlock(&lo
->lo_ctl_mutex
);
1669 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1671 mutex_lock(&loop_index_mutex
);
1672 __lo_release(disk
->private_data
);
1673 mutex_unlock(&loop_index_mutex
);
1676 static const struct block_device_operations lo_fops
= {
1677 .owner
= THIS_MODULE
,
1679 .release
= lo_release
,
1681 #ifdef CONFIG_COMPAT
1682 .compat_ioctl
= lo_compat_ioctl
,
1687 * And now the modules code and kernel interface.
1689 static int max_loop
;
1690 module_param(max_loop
, int, S_IRUGO
);
1691 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1692 module_param(max_part
, int, S_IRUGO
);
1693 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1694 MODULE_LICENSE("GPL");
1695 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1697 int loop_register_transfer(struct loop_func_table
*funcs
)
1699 unsigned int n
= funcs
->number
;
1701 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1703 xfer_funcs
[n
] = funcs
;
1707 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1709 struct loop_device
*lo
= ptr
;
1710 struct loop_func_table
*xfer
= data
;
1712 mutex_lock(&lo
->lo_ctl_mutex
);
1713 if (lo
->lo_encryption
== xfer
)
1714 loop_release_xfer(lo
);
1715 mutex_unlock(&lo
->lo_ctl_mutex
);
1719 int loop_unregister_transfer(int number
)
1721 unsigned int n
= number
;
1722 struct loop_func_table
*xfer
;
1724 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1727 xfer_funcs
[n
] = NULL
;
1728 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1732 EXPORT_SYMBOL(loop_register_transfer
);
1733 EXPORT_SYMBOL(loop_unregister_transfer
);
1735 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1736 const struct blk_mq_queue_data
*bd
)
1738 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1739 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1741 blk_mq_start_request(bd
->rq
);
1743 if (lo
->lo_state
!= Lo_bound
)
1744 return BLK_STS_IOERR
;
1746 switch (req_op(cmd
->rq
)) {
1748 case REQ_OP_DISCARD
:
1749 case REQ_OP_WRITE_ZEROES
:
1750 cmd
->use_aio
= false;
1753 cmd
->use_aio
= lo
->use_dio
;
1757 /* always use the first bio's css */
1758 #ifdef CONFIG_BLK_CGROUP
1759 if (cmd
->use_aio
&& cmd
->rq
->bio
&& cmd
->rq
->bio
->bi_css
) {
1760 cmd
->css
= cmd
->rq
->bio
->bi_css
;
1765 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1770 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1772 const bool write
= op_is_write(req_op(cmd
->rq
));
1773 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1776 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1781 ret
= do_req_filebacked(lo
, cmd
->rq
);
1783 /* complete non-aio request */
1784 if (!cmd
->use_aio
|| ret
) {
1785 cmd
->ret
= ret
? -EIO
: 0;
1786 blk_mq_complete_request(cmd
->rq
);
1790 static void loop_queue_work(struct kthread_work
*work
)
1792 struct loop_cmd
*cmd
=
1793 container_of(work
, struct loop_cmd
, work
);
1795 loop_handle_cmd(cmd
);
1798 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1799 unsigned int hctx_idx
, unsigned int numa_node
)
1801 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1804 kthread_init_work(&cmd
->work
, loop_queue_work
);
1809 static const struct blk_mq_ops loop_mq_ops
= {
1810 .queue_rq
= loop_queue_rq
,
1811 .init_request
= loop_init_request
,
1812 .complete
= lo_complete_rq
,
1815 static int loop_add(struct loop_device
**l
, int i
)
1817 struct loop_device
*lo
;
1818 struct gendisk
*disk
;
1822 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1826 lo
->lo_state
= Lo_unbound
;
1828 /* allocate id, if @id >= 0, we're requesting that specific id */
1830 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1834 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1841 lo
->tag_set
.ops
= &loop_mq_ops
;
1842 lo
->tag_set
.nr_hw_queues
= 1;
1843 lo
->tag_set
.queue_depth
= 128;
1844 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1845 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1846 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1847 lo
->tag_set
.driver_data
= lo
;
1849 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1853 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1854 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1855 err
= PTR_ERR(lo
->lo_queue
);
1856 goto out_cleanup_tags
;
1858 lo
->lo_queue
->queuedata
= lo
;
1860 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
1863 * By default, we do buffer IO, so it doesn't make sense to enable
1864 * merge because the I/O submitted to backing file is handled page by
1865 * page. For directio mode, merge does help to dispatch bigger request
1866 * to underlayer disk. We will enable merge once directio is enabled.
1868 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
1871 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1873 goto out_free_queue
;
1876 * Disable partition scanning by default. The in-kernel partition
1877 * scanning can be requested individually per-device during its
1878 * setup. Userspace can always add and remove partitions from all
1879 * devices. The needed partition minors are allocated from the
1880 * extended minor space, the main loop device numbers will continue
1881 * to match the loop minors, regardless of the number of partitions
1884 * If max_part is given, partition scanning is globally enabled for
1885 * all loop devices. The minors for the main loop devices will be
1886 * multiples of max_part.
1888 * Note: Global-for-all-devices, set-only-at-init, read-only module
1889 * parameteters like 'max_loop' and 'max_part' make things needlessly
1890 * complicated, are too static, inflexible and may surprise
1891 * userspace tools. Parameters like this in general should be avoided.
1894 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1895 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1896 mutex_init(&lo
->lo_ctl_mutex
);
1897 atomic_set(&lo
->lo_refcnt
, 0);
1899 spin_lock_init(&lo
->lo_lock
);
1900 disk
->major
= LOOP_MAJOR
;
1901 disk
->first_minor
= i
<< part_shift
;
1902 disk
->fops
= &lo_fops
;
1903 disk
->private_data
= lo
;
1904 disk
->queue
= lo
->lo_queue
;
1905 sprintf(disk
->disk_name
, "loop%d", i
);
1908 return lo
->lo_number
;
1911 blk_cleanup_queue(lo
->lo_queue
);
1913 blk_mq_free_tag_set(&lo
->tag_set
);
1915 idr_remove(&loop_index_idr
, i
);
1922 static void loop_remove(struct loop_device
*lo
)
1924 blk_cleanup_queue(lo
->lo_queue
);
1925 del_gendisk(lo
->lo_disk
);
1926 blk_mq_free_tag_set(&lo
->tag_set
);
1927 put_disk(lo
->lo_disk
);
1931 static int find_free_cb(int id
, void *ptr
, void *data
)
1933 struct loop_device
*lo
= ptr
;
1934 struct loop_device
**l
= data
;
1936 if (lo
->lo_state
== Lo_unbound
) {
1943 static int loop_lookup(struct loop_device
**l
, int i
)
1945 struct loop_device
*lo
;
1951 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1954 ret
= lo
->lo_number
;
1959 /* lookup and return a specific i */
1960 lo
= idr_find(&loop_index_idr
, i
);
1963 ret
= lo
->lo_number
;
1969 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1971 struct loop_device
*lo
;
1972 struct kobject
*kobj
;
1975 mutex_lock(&loop_index_mutex
);
1976 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1978 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1982 kobj
= get_disk(lo
->lo_disk
);
1983 mutex_unlock(&loop_index_mutex
);
1989 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1992 struct loop_device
*lo
;
1995 mutex_lock(&loop_index_mutex
);
1998 ret
= loop_lookup(&lo
, parm
);
2003 ret
= loop_add(&lo
, parm
);
2005 case LOOP_CTL_REMOVE
:
2006 ret
= loop_lookup(&lo
, parm
);
2009 mutex_lock(&lo
->lo_ctl_mutex
);
2010 if (lo
->lo_state
!= Lo_unbound
) {
2012 mutex_unlock(&lo
->lo_ctl_mutex
);
2015 if (atomic_read(&lo
->lo_refcnt
) > 0) {
2017 mutex_unlock(&lo
->lo_ctl_mutex
);
2020 lo
->lo_disk
->private_data
= NULL
;
2021 mutex_unlock(&lo
->lo_ctl_mutex
);
2022 idr_remove(&loop_index_idr
, lo
->lo_number
);
2025 case LOOP_CTL_GET_FREE
:
2026 ret
= loop_lookup(&lo
, -1);
2029 ret
= loop_add(&lo
, -1);
2031 mutex_unlock(&loop_index_mutex
);
2036 static const struct file_operations loop_ctl_fops
= {
2037 .open
= nonseekable_open
,
2038 .unlocked_ioctl
= loop_control_ioctl
,
2039 .compat_ioctl
= loop_control_ioctl
,
2040 .owner
= THIS_MODULE
,
2041 .llseek
= noop_llseek
,
2044 static struct miscdevice loop_misc
= {
2045 .minor
= LOOP_CTRL_MINOR
,
2046 .name
= "loop-control",
2047 .fops
= &loop_ctl_fops
,
2050 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2051 MODULE_ALIAS("devname:loop-control");
2053 static int __init
loop_init(void)
2056 unsigned long range
;
2057 struct loop_device
*lo
;
2062 part_shift
= fls(max_part
);
2065 * Adjust max_part according to part_shift as it is exported
2066 * to user space so that user can decide correct minor number
2067 * if [s]he want to create more devices.
2069 * Note that -1 is required because partition 0 is reserved
2070 * for the whole disk.
2072 max_part
= (1UL << part_shift
) - 1;
2075 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2080 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2086 * If max_loop is specified, create that many devices upfront.
2087 * This also becomes a hard limit. If max_loop is not specified,
2088 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2089 * init time. Loop devices can be requested on-demand with the
2090 * /dev/loop-control interface, or be instantiated by accessing
2091 * a 'dead' device node.
2095 range
= max_loop
<< part_shift
;
2097 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2098 range
= 1UL << MINORBITS
;
2101 err
= misc_register(&loop_misc
);
2106 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2111 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2112 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2114 /* pre-create number of devices given by config or max_loop */
2115 mutex_lock(&loop_index_mutex
);
2116 for (i
= 0; i
< nr
; i
++)
2118 mutex_unlock(&loop_index_mutex
);
2120 printk(KERN_INFO
"loop: module loaded\n");
2124 misc_deregister(&loop_misc
);
2129 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2131 struct loop_device
*lo
= ptr
;
2137 static void __exit
loop_exit(void)
2139 unsigned long range
;
2141 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2143 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2144 idr_destroy(&loop_index_idr
);
2146 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2147 unregister_blkdev(LOOP_MAJOR
, "loop");
2149 misc_deregister(&loop_misc
);
2152 module_init(loop_init
);
2153 module_exit(loop_exit
);
2156 static int __init
max_loop_setup(char *str
)
2158 max_loop
= simple_strtol(str
, NULL
, 0);
2162 __setup("max_loop=", max_loop_setup
);