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>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
84 #include <linux/uaccess.h>
86 static DEFINE_IDR(loop_index_idr
);
87 static DEFINE_MUTEX(loop_ctl_mutex
);
90 static int part_shift
;
92 static int transfer_xor(struct loop_device
*lo
, int cmd
,
93 struct page
*raw_page
, unsigned raw_off
,
94 struct page
*loop_page
, unsigned loop_off
,
95 int size
, sector_t real_block
)
97 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
98 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
110 key
= lo
->lo_encrypt_key
;
111 keysize
= lo
->lo_encrypt_key_size
;
112 for (i
= 0; i
< size
; i
++)
113 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
115 kunmap_atomic(loop_buf
);
116 kunmap_atomic(raw_buf
);
121 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
123 if (unlikely(info
->lo_encrypt_key_size
<= 0))
128 static struct loop_func_table none_funcs
= {
129 .number
= LO_CRYPT_NONE
,
132 static struct loop_func_table xor_funcs
= {
133 .number
= LO_CRYPT_XOR
,
134 .transfer
= transfer_xor
,
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
144 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
148 /* Compute loopsize in bytes */
149 loopsize
= i_size_read(file
->f_mapping
->host
);
152 /* offset is beyond i_size, weird but possible */
156 if (sizelimit
> 0 && sizelimit
< loopsize
)
157 loopsize
= sizelimit
;
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
162 return loopsize
>> 9;
165 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
167 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
170 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
172 struct file
*file
= lo
->lo_backing_file
;
173 struct address_space
*mapping
= file
->f_mapping
;
174 struct inode
*inode
= mapping
->host
;
175 unsigned short sb_bsize
= 0;
176 unsigned dio_align
= 0;
179 if (inode
->i_sb
->s_bdev
) {
180 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
181 dio_align
= sb_bsize
- 1;
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
195 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
196 !(lo
->lo_offset
& dio_align
) &&
197 mapping
->a_ops
->direct_IO
&&
206 if (lo
->use_dio
== use_dio
)
209 /* flush dirty pages before changing direct IO */
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
217 if (lo
->lo_state
== Lo_bound
)
218 blk_mq_freeze_queue(lo
->lo_queue
);
219 lo
->use_dio
= use_dio
;
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
222 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
225 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
227 if (lo
->lo_state
== Lo_bound
)
228 blk_mq_unfreeze_queue(lo
->lo_queue
);
232 * loop_set_size() - sets device size and notifies userspace
233 * @lo: struct loop_device to set the size for
234 * @size: new size of the loop device
236 * Callers must validate that the size passed into this function fits into
237 * a sector_t, eg using loop_validate_size()
239 static void loop_set_size(struct loop_device
*lo
, loff_t size
)
241 struct block_device
*bdev
= lo
->lo_device
;
243 bd_set_size(bdev
, size
<< SECTOR_SHIFT
);
245 set_capacity_revalidate_and_notify(lo
->lo_disk
, size
, false);
249 lo_do_transfer(struct loop_device
*lo
, int cmd
,
250 struct page
*rpage
, unsigned roffs
,
251 struct page
*lpage
, unsigned loffs
,
252 int size
, sector_t rblock
)
256 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock
<< 9, size
);
266 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
271 iov_iter_bvec(&i
, WRITE
, bvec
, 1, bvec
->bv_len
);
273 file_start_write(file
);
274 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
275 file_end_write(file
);
277 if (likely(bw
== bvec
->bv_len
))
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos
, bvec
->bv_len
);
288 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
292 struct req_iterator iter
;
295 rq_for_each_segment(bvec
, rq
, iter
) {
296 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
310 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
313 struct bio_vec bvec
, b
;
314 struct req_iterator iter
;
318 page
= alloc_page(GFP_NOIO
);
322 rq_for_each_segment(bvec
, rq
, iter
) {
323 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
324 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
330 b
.bv_len
= bvec
.bv_len
;
331 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
340 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
344 struct req_iterator iter
;
348 rq_for_each_segment(bvec
, rq
, iter
) {
349 iov_iter_bvec(&i
, READ
, &bvec
, 1, bvec
.bv_len
);
350 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
354 flush_dcache_page(bvec
.bv_page
);
356 if (len
!= bvec
.bv_len
) {
359 __rq_for_each_bio(bio
, rq
)
369 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
372 struct bio_vec bvec
, b
;
373 struct req_iterator iter
;
379 page
= alloc_page(GFP_NOIO
);
383 rq_for_each_segment(bvec
, rq
, iter
) {
388 b
.bv_len
= bvec
.bv_len
;
390 iov_iter_bvec(&i
, READ
, &b
, 1, b
.bv_len
);
391 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
397 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
398 bvec
.bv_offset
, len
, offset
>> 9);
402 flush_dcache_page(bvec
.bv_page
);
404 if (len
!= bvec
.bv_len
) {
407 __rq_for_each_bio(bio
, rq
)
419 static int lo_fallocate(struct loop_device
*lo
, struct request
*rq
, loff_t pos
,
423 * We use fallocate to manipulate the space mappings used by the image
424 * a.k.a. discard/zerorange. However we do not support this if
425 * encryption is enabled, because it may give an attacker useful
428 struct file
*file
= lo
->lo_backing_file
;
429 struct request_queue
*q
= lo
->lo_queue
;
432 mode
|= FALLOC_FL_KEEP_SIZE
;
434 if (!blk_queue_discard(q
)) {
439 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
440 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
446 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
448 struct file
*file
= lo
->lo_backing_file
;
449 int ret
= vfs_fsync(file
, 0);
450 if (unlikely(ret
&& ret
!= -EINVAL
))
456 static void lo_complete_rq(struct request
*rq
)
458 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
459 blk_status_t ret
= BLK_STS_OK
;
461 if (!cmd
->use_aio
|| cmd
->ret
< 0 || cmd
->ret
== blk_rq_bytes(rq
) ||
462 req_op(rq
) != REQ_OP_READ
) {
464 ret
= errno_to_blk_status(cmd
->ret
);
469 * Short READ - if we got some data, advance our request and
470 * retry it. If we got no data, end the rest with EIO.
473 blk_update_request(rq
, BLK_STS_OK
, cmd
->ret
);
475 blk_mq_requeue_request(rq
, true);
478 struct bio
*bio
= rq
->bio
;
487 blk_mq_end_request(rq
, ret
);
491 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
493 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
495 if (!atomic_dec_and_test(&cmd
->ref
))
499 blk_mq_complete_request(rq
);
502 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
504 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
509 lo_rw_aio_do_completion(cmd
);
512 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
515 struct iov_iter iter
;
516 struct req_iterator rq_iter
;
517 struct bio_vec
*bvec
;
518 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
519 struct bio
*bio
= rq
->bio
;
520 struct file
*file
= lo
->lo_backing_file
;
526 rq_for_each_bvec(tmp
, rq
, rq_iter
)
529 if (rq
->bio
!= rq
->biotail
) {
531 bvec
= kmalloc_array(nr_bvec
, sizeof(struct bio_vec
),
538 * The bios of the request may be started from the middle of
539 * the 'bvec' because of bio splitting, so we can't directly
540 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
541 * API will take care of all details for us.
543 rq_for_each_bvec(tmp
, rq
, rq_iter
) {
551 * Same here, this bio may be started from the middle of the
552 * 'bvec' because of bio splitting, so offset from the bvec
553 * must be passed to iov iterator
555 offset
= bio
->bi_iter
.bi_bvec_done
;
556 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
558 atomic_set(&cmd
->ref
, 2);
560 iov_iter_bvec(&iter
, rw
, bvec
, nr_bvec
, blk_rq_bytes(rq
));
561 iter
.iov_offset
= offset
;
563 cmd
->iocb
.ki_pos
= pos
;
564 cmd
->iocb
.ki_filp
= file
;
565 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
566 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
567 cmd
->iocb
.ki_ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
569 kthread_associate_blkcg(cmd
->css
);
572 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
574 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
576 lo_rw_aio_do_completion(cmd
);
577 kthread_associate_blkcg(NULL
);
579 if (ret
!= -EIOCBQUEUED
)
580 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
584 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
586 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
587 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
590 * lo_write_simple and lo_read_simple should have been covered
591 * by io submit style function like lo_rw_aio(), one blocker
592 * is that lo_read_simple() need to call flush_dcache_page after
593 * the page is written from kernel, and it isn't easy to handle
594 * this in io submit style function which submits all segments
595 * of the req at one time. And direct read IO doesn't need to
596 * run flush_dcache_page().
598 switch (req_op(rq
)) {
600 return lo_req_flush(lo
, rq
);
601 case REQ_OP_WRITE_ZEROES
:
603 * If the caller doesn't want deallocation, call zeroout to
604 * write zeroes the range. Otherwise, punch them out.
606 return lo_fallocate(lo
, rq
, pos
,
607 (rq
->cmd_flags
& REQ_NOUNMAP
) ?
608 FALLOC_FL_ZERO_RANGE
:
609 FALLOC_FL_PUNCH_HOLE
);
611 return lo_fallocate(lo
, rq
, pos
, FALLOC_FL_PUNCH_HOLE
);
614 return lo_write_transfer(lo
, rq
, pos
);
615 else if (cmd
->use_aio
)
616 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
618 return lo_write_simple(lo
, rq
, pos
);
621 return lo_read_transfer(lo
, rq
, pos
);
622 else if (cmd
->use_aio
)
623 return lo_rw_aio(lo
, cmd
, pos
, READ
);
625 return lo_read_simple(lo
, rq
, pos
);
632 static inline void loop_update_dio(struct loop_device
*lo
)
634 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
638 static void loop_reread_partitions(struct loop_device
*lo
,
639 struct block_device
*bdev
)
643 mutex_lock(&bdev
->bd_mutex
);
644 rc
= bdev_disk_changed(bdev
, false);
645 mutex_unlock(&bdev
->bd_mutex
);
647 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
648 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
651 static inline int is_loop_device(struct file
*file
)
653 struct inode
*i
= file
->f_mapping
->host
;
655 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
658 static int loop_validate_file(struct file
*file
, struct block_device
*bdev
)
660 struct inode
*inode
= file
->f_mapping
->host
;
661 struct file
*f
= file
;
663 /* Avoid recursion */
664 while (is_loop_device(f
)) {
665 struct loop_device
*l
;
667 if (f
->f_mapping
->host
->i_bdev
== bdev
)
670 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
671 if (l
->lo_state
!= Lo_bound
) {
674 f
= l
->lo_backing_file
;
676 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
682 * loop_change_fd switched the backing store of a loopback device to
683 * a new file. This is useful for operating system installers to free up
684 * the original file and in High Availability environments to switch to
685 * an alternative location for the content in case of server meltdown.
686 * This can only work if the loop device is used read-only, and if the
687 * new backing store is the same size and type as the old backing store.
689 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
692 struct file
*file
= NULL
, *old_file
;
696 error
= mutex_lock_killable(&loop_ctl_mutex
);
700 if (lo
->lo_state
!= Lo_bound
)
703 /* the loop device has to be read-only */
705 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
713 error
= loop_validate_file(file
, bdev
);
717 old_file
= lo
->lo_backing_file
;
721 /* size of the new backing store needs to be the same */
722 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
726 blk_mq_freeze_queue(lo
->lo_queue
);
727 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
728 lo
->lo_backing_file
= file
;
729 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
730 mapping_set_gfp_mask(file
->f_mapping
,
731 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
733 blk_mq_unfreeze_queue(lo
->lo_queue
);
734 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
735 mutex_unlock(&loop_ctl_mutex
);
737 * We must drop file reference outside of loop_ctl_mutex as dropping
738 * the file ref can take bd_mutex which creates circular locking
743 loop_reread_partitions(lo
, bdev
);
747 mutex_unlock(&loop_ctl_mutex
);
753 /* loop sysfs attributes */
755 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
756 ssize_t (*callback
)(struct loop_device
*, char *))
758 struct gendisk
*disk
= dev_to_disk(dev
);
759 struct loop_device
*lo
= disk
->private_data
;
761 return callback(lo
, page
);
764 #define LOOP_ATTR_RO(_name) \
765 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
766 static ssize_t loop_attr_do_show_##_name(struct device *d, \
767 struct device_attribute *attr, char *b) \
769 return loop_attr_show(d, b, loop_attr_##_name##_show); \
771 static struct device_attribute loop_attr_##_name = \
772 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
774 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
779 spin_lock_irq(&lo
->lo_lock
);
780 if (lo
->lo_backing_file
)
781 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
782 spin_unlock_irq(&lo
->lo_lock
);
784 if (IS_ERR_OR_NULL(p
))
788 memmove(buf
, p
, ret
);
796 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
798 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
801 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
803 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
806 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
808 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
810 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
813 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
815 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
817 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
820 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
822 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
824 return sprintf(buf
, "%s\n", dio
? "1" : "0");
827 LOOP_ATTR_RO(backing_file
);
828 LOOP_ATTR_RO(offset
);
829 LOOP_ATTR_RO(sizelimit
);
830 LOOP_ATTR_RO(autoclear
);
831 LOOP_ATTR_RO(partscan
);
834 static struct attribute
*loop_attrs
[] = {
835 &loop_attr_backing_file
.attr
,
836 &loop_attr_offset
.attr
,
837 &loop_attr_sizelimit
.attr
,
838 &loop_attr_autoclear
.attr
,
839 &loop_attr_partscan
.attr
,
844 static struct attribute_group loop_attribute_group
= {
849 static void loop_sysfs_init(struct loop_device
*lo
)
851 lo
->sysfs_inited
= !sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
852 &loop_attribute_group
);
855 static void loop_sysfs_exit(struct loop_device
*lo
)
857 if (lo
->sysfs_inited
)
858 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
859 &loop_attribute_group
);
862 static void loop_config_discard(struct loop_device
*lo
)
864 struct file
*file
= lo
->lo_backing_file
;
865 struct inode
*inode
= file
->f_mapping
->host
;
866 struct request_queue
*q
= lo
->lo_queue
;
869 * If the backing device is a block device, mirror its zeroing
870 * capability. Set the discard sectors to the block device's zeroing
871 * capabilities because loop discards result in blkdev_issue_zeroout(),
872 * not blkdev_issue_discard(). This maintains consistent behavior with
873 * file-backed loop devices: discarded regions read back as zero.
875 if (S_ISBLK(inode
->i_mode
) && !lo
->lo_encrypt_key_size
) {
876 struct request_queue
*backingq
;
878 backingq
= bdev_get_queue(inode
->i_bdev
);
879 blk_queue_max_discard_sectors(q
,
880 backingq
->limits
.max_write_zeroes_sectors
);
882 blk_queue_max_write_zeroes_sectors(q
,
883 backingq
->limits
.max_write_zeroes_sectors
);
886 * We use punch hole to reclaim the free space used by the
887 * image a.k.a. discard. However we do not support discard if
888 * encryption is enabled, because it may give an attacker
889 * useful information.
891 } else if (!file
->f_op
->fallocate
|| lo
->lo_encrypt_key_size
) {
892 q
->limits
.discard_granularity
= 0;
893 q
->limits
.discard_alignment
= 0;
894 blk_queue_max_discard_sectors(q
, 0);
895 blk_queue_max_write_zeroes_sectors(q
, 0);
898 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
899 q
->limits
.discard_alignment
= 0;
901 blk_queue_max_discard_sectors(q
, UINT_MAX
>> 9);
902 blk_queue_max_write_zeroes_sectors(q
, UINT_MAX
>> 9);
905 if (q
->limits
.max_write_zeroes_sectors
)
906 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, q
);
908 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, q
);
911 static void loop_unprepare_queue(struct loop_device
*lo
)
913 kthread_flush_worker(&lo
->worker
);
914 kthread_stop(lo
->worker_task
);
917 static int loop_kthread_worker_fn(void *worker_ptr
)
919 current
->flags
|= PF_LESS_THROTTLE
| PF_MEMALLOC_NOIO
;
920 return kthread_worker_fn(worker_ptr
);
923 static int loop_prepare_queue(struct loop_device
*lo
)
925 kthread_init_worker(&lo
->worker
);
926 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
927 &lo
->worker
, "loop%d", lo
->lo_number
);
928 if (IS_ERR(lo
->worker_task
))
930 set_user_nice(lo
->worker_task
, MIN_NICE
);
934 static void loop_update_rotational(struct loop_device
*lo
)
936 struct file
*file
= lo
->lo_backing_file
;
937 struct inode
*file_inode
= file
->f_mapping
->host
;
938 struct block_device
*file_bdev
= file_inode
->i_sb
->s_bdev
;
939 struct request_queue
*q
= lo
->lo_queue
;
942 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
944 nonrot
= blk_queue_nonrot(bdev_get_queue(file_bdev
));
947 blk_queue_flag_set(QUEUE_FLAG_NONROT
, q
);
949 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, q
);
953 loop_release_xfer(struct loop_device
*lo
)
956 struct loop_func_table
*xfer
= lo
->lo_encryption
;
960 err
= xfer
->release(lo
);
962 lo
->lo_encryption
= NULL
;
963 module_put(xfer
->owner
);
969 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
970 const struct loop_info64
*i
)
975 struct module
*owner
= xfer
->owner
;
977 if (!try_module_get(owner
))
980 err
= xfer
->init(lo
, i
);
984 lo
->lo_encryption
= xfer
;
990 * loop_set_status_from_info - configure device from loop_info
991 * @lo: struct loop_device to configure
992 * @info: struct loop_info64 to configure the device with
994 * Configures the loop device parameters according to the passed
995 * in loop_info64 configuration.
998 loop_set_status_from_info(struct loop_device
*lo
,
999 const struct loop_info64
*info
)
1002 struct loop_func_table
*xfer
;
1003 kuid_t uid
= current_uid();
1005 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1008 err
= loop_release_xfer(lo
);
1012 if (info
->lo_encrypt_type
) {
1013 unsigned int type
= info
->lo_encrypt_type
;
1015 if (type
>= MAX_LO_CRYPT
)
1017 xfer
= xfer_funcs
[type
];
1023 err
= loop_init_xfer(lo
, xfer
, info
);
1027 lo
->lo_offset
= info
->lo_offset
;
1028 lo
->lo_sizelimit
= info
->lo_sizelimit
;
1029 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1030 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1031 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1032 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1036 lo
->transfer
= xfer
->transfer
;
1037 lo
->ioctl
= xfer
->ioctl
;
1039 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1040 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1041 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1043 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1044 lo
->lo_init
[0] = info
->lo_init
[0];
1045 lo
->lo_init
[1] = info
->lo_init
[1];
1046 if (info
->lo_encrypt_key_size
) {
1047 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1048 info
->lo_encrypt_key_size
);
1049 lo
->lo_key_owner
= uid
;
1055 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
1056 struct block_device
*bdev
, unsigned int arg
)
1059 struct inode
*inode
;
1060 struct address_space
*mapping
;
1061 struct block_device
*claimed_bdev
= NULL
;
1067 /* This is safe, since we have a reference from open(). */
1068 __module_get(THIS_MODULE
);
1076 * If we don't hold exclusive handle for the device, upgrade to it
1077 * here to avoid changing device under exclusive owner.
1079 if (!(mode
& FMODE_EXCL
)) {
1080 claimed_bdev
= bd_start_claiming(bdev
, loop_set_fd
);
1081 if (IS_ERR(claimed_bdev
)) {
1082 error
= PTR_ERR(claimed_bdev
);
1087 error
= mutex_lock_killable(&loop_ctl_mutex
);
1092 if (lo
->lo_state
!= Lo_unbound
)
1095 error
= loop_validate_file(file
, bdev
);
1099 mapping
= file
->f_mapping
;
1100 inode
= mapping
->host
;
1102 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
1103 !file
->f_op
->write_iter
)
1104 lo_flags
|= LO_FLAGS_READ_ONLY
;
1106 size
= get_loop_size(lo
, file
);
1108 error
= loop_prepare_queue(lo
);
1114 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
1116 lo
->use_dio
= false;
1117 lo
->lo_device
= bdev
;
1118 lo
->lo_flags
= lo_flags
;
1119 lo
->lo_backing_file
= file
;
1120 lo
->transfer
= NULL
;
1122 lo
->lo_sizelimit
= 0;
1123 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
1124 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
1126 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
1127 blk_queue_write_cache(lo
->lo_queue
, true, false);
1129 if (io_is_direct(lo
->lo_backing_file
) && inode
->i_sb
->s_bdev
) {
1130 /* In case of direct I/O, match underlying block size */
1131 unsigned short bsize
= bdev_logical_block_size(
1132 inode
->i_sb
->s_bdev
);
1134 blk_queue_logical_block_size(lo
->lo_queue
, bsize
);
1135 blk_queue_physical_block_size(lo
->lo_queue
, bsize
);
1136 blk_queue_io_min(lo
->lo_queue
, bsize
);
1139 loop_update_rotational(lo
);
1140 loop_update_dio(lo
);
1141 loop_sysfs_init(lo
);
1142 loop_set_size(lo
, size
);
1144 set_blocksize(bdev
, S_ISBLK(inode
->i_mode
) ?
1145 block_size(inode
->i_bdev
) : PAGE_SIZE
);
1147 lo
->lo_state
= Lo_bound
;
1149 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1150 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
1152 /* Grab the block_device to prevent its destruction after we
1153 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1156 mutex_unlock(&loop_ctl_mutex
);
1158 loop_reread_partitions(lo
, bdev
);
1160 bd_abort_claiming(bdev
, claimed_bdev
, loop_set_fd
);
1164 mutex_unlock(&loop_ctl_mutex
);
1167 bd_abort_claiming(bdev
, claimed_bdev
, loop_set_fd
);
1171 /* This is safe: open() is still holding a reference. */
1172 module_put(THIS_MODULE
);
1176 static int __loop_clr_fd(struct loop_device
*lo
, bool release
)
1178 struct file
*filp
= NULL
;
1179 gfp_t gfp
= lo
->old_gfp_mask
;
1180 struct block_device
*bdev
= lo
->lo_device
;
1182 bool partscan
= false;
1185 mutex_lock(&loop_ctl_mutex
);
1186 if (WARN_ON_ONCE(lo
->lo_state
!= Lo_rundown
)) {
1191 filp
= lo
->lo_backing_file
;
1197 /* freeze request queue during the transition */
1198 blk_mq_freeze_queue(lo
->lo_queue
);
1200 spin_lock_irq(&lo
->lo_lock
);
1201 lo
->lo_backing_file
= NULL
;
1202 spin_unlock_irq(&lo
->lo_lock
);
1204 loop_release_xfer(lo
);
1205 lo
->transfer
= NULL
;
1207 lo
->lo_device
= NULL
;
1208 lo
->lo_encryption
= NULL
;
1210 lo
->lo_sizelimit
= 0;
1211 lo
->lo_encrypt_key_size
= 0;
1212 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1213 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1214 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1215 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1216 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1217 blk_queue_io_min(lo
->lo_queue
, 512);
1220 invalidate_bdev(bdev
);
1221 bdev
->bd_inode
->i_mapping
->wb_err
= 0;
1223 set_capacity(lo
->lo_disk
, 0);
1224 loop_sysfs_exit(lo
);
1226 bd_set_size(bdev
, 0);
1227 /* let user-space know about this change */
1228 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1230 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1231 /* This is safe: open() is still holding a reference. */
1232 module_put(THIS_MODULE
);
1233 blk_mq_unfreeze_queue(lo
->lo_queue
);
1235 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
;
1236 lo_number
= lo
->lo_number
;
1237 loop_unprepare_queue(lo
);
1239 mutex_unlock(&loop_ctl_mutex
);
1242 * bd_mutex has been held already in release path, so don't
1243 * acquire it if this function is called in such case.
1245 * If the reread partition isn't from release path, lo_refcnt
1246 * must be at least one and it can only become zero when the
1247 * current holder is released.
1250 mutex_lock(&bdev
->bd_mutex
);
1251 err
= bdev_disk_changed(bdev
, false);
1253 mutex_unlock(&bdev
->bd_mutex
);
1255 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1256 __func__
, lo_number
, err
);
1257 /* Device is gone, no point in returning error */
1262 * lo->lo_state is set to Lo_unbound here after above partscan has
1265 * There cannot be anybody else entering __loop_clr_fd() as
1266 * lo->lo_backing_file is already cleared and Lo_rundown state
1267 * protects us from all the other places trying to change the 'lo'
1270 mutex_lock(&loop_ctl_mutex
);
1273 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1274 lo
->lo_state
= Lo_unbound
;
1275 mutex_unlock(&loop_ctl_mutex
);
1278 * Need not hold loop_ctl_mutex to fput backing file.
1279 * Calling fput holding loop_ctl_mutex triggers a circular
1280 * lock dependency possibility warning as fput can take
1281 * bd_mutex which is usually taken before loop_ctl_mutex.
1288 static int loop_clr_fd(struct loop_device
*lo
)
1292 err
= mutex_lock_killable(&loop_ctl_mutex
);
1295 if (lo
->lo_state
!= Lo_bound
) {
1296 mutex_unlock(&loop_ctl_mutex
);
1300 * If we've explicitly asked to tear down the loop device,
1301 * and it has an elevated reference count, set it for auto-teardown when
1302 * the last reference goes away. This stops $!~#$@ udev from
1303 * preventing teardown because it decided that it needs to run blkid on
1304 * the loopback device whenever they appear. xfstests is notorious for
1305 * failing tests because blkid via udev races with a losetup
1306 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1307 * command to fail with EBUSY.
1309 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1310 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1311 mutex_unlock(&loop_ctl_mutex
);
1314 lo
->lo_state
= Lo_rundown
;
1315 mutex_unlock(&loop_ctl_mutex
);
1317 return __loop_clr_fd(lo
, false);
1321 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1324 struct block_device
*bdev
;
1325 kuid_t uid
= current_uid();
1326 bool partscan
= false;
1327 bool size_changed
= false;
1329 err
= mutex_lock_killable(&loop_ctl_mutex
);
1332 if (lo
->lo_encrypt_key_size
&&
1333 !uid_eq(lo
->lo_key_owner
, uid
) &&
1334 !capable(CAP_SYS_ADMIN
)) {
1338 if (lo
->lo_state
!= Lo_bound
) {
1343 if (lo
->lo_offset
!= info
->lo_offset
||
1344 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1345 size_changed
= true;
1346 sync_blockdev(lo
->lo_device
);
1347 kill_bdev(lo
->lo_device
);
1350 /* I/O need to be drained during transfer transition */
1351 blk_mq_freeze_queue(lo
->lo_queue
);
1353 if (size_changed
&& lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1354 /* If any pages were dirtied after kill_bdev(), try again */
1356 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1357 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1358 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1362 err
= loop_set_status_from_info(lo
, info
);
1367 loff_t new_size
= get_size(lo
->lo_offset
, lo
->lo_sizelimit
,
1368 lo
->lo_backing_file
);
1369 loop_set_size(lo
, new_size
);
1372 loop_config_discard(lo
);
1374 /* update dio if lo_offset or transfer is changed */
1375 __loop_update_dio(lo
, lo
->use_dio
);
1378 blk_mq_unfreeze_queue(lo
->lo_queue
);
1380 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1381 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1382 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1383 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1384 bdev
= lo
->lo_device
;
1388 mutex_unlock(&loop_ctl_mutex
);
1390 loop_reread_partitions(lo
, bdev
);
1396 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1402 ret
= mutex_lock_killable(&loop_ctl_mutex
);
1405 if (lo
->lo_state
!= Lo_bound
) {
1406 mutex_unlock(&loop_ctl_mutex
);
1410 memset(info
, 0, sizeof(*info
));
1411 info
->lo_number
= lo
->lo_number
;
1412 info
->lo_offset
= lo
->lo_offset
;
1413 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1414 info
->lo_flags
= lo
->lo_flags
;
1415 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1416 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1417 info
->lo_encrypt_type
=
1418 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1419 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1420 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1421 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1422 lo
->lo_encrypt_key_size
);
1425 /* Drop loop_ctl_mutex while we call into the filesystem. */
1426 path
= lo
->lo_backing_file
->f_path
;
1428 mutex_unlock(&loop_ctl_mutex
);
1429 ret
= vfs_getattr(&path
, &stat
, STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1431 info
->lo_device
= huge_encode_dev(stat
.dev
);
1432 info
->lo_inode
= stat
.ino
;
1433 info
->lo_rdevice
= huge_encode_dev(stat
.rdev
);
1440 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1442 memset(info64
, 0, sizeof(*info64
));
1443 info64
->lo_number
= info
->lo_number
;
1444 info64
->lo_device
= info
->lo_device
;
1445 info64
->lo_inode
= info
->lo_inode
;
1446 info64
->lo_rdevice
= info
->lo_rdevice
;
1447 info64
->lo_offset
= info
->lo_offset
;
1448 info64
->lo_sizelimit
= 0;
1449 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1450 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1451 info64
->lo_flags
= info
->lo_flags
;
1452 info64
->lo_init
[0] = info
->lo_init
[0];
1453 info64
->lo_init
[1] = info
->lo_init
[1];
1454 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1455 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1457 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1458 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1462 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1464 memset(info
, 0, sizeof(*info
));
1465 info
->lo_number
= info64
->lo_number
;
1466 info
->lo_device
= info64
->lo_device
;
1467 info
->lo_inode
= info64
->lo_inode
;
1468 info
->lo_rdevice
= info64
->lo_rdevice
;
1469 info
->lo_offset
= info64
->lo_offset
;
1470 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1471 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1472 info
->lo_flags
= info64
->lo_flags
;
1473 info
->lo_init
[0] = info64
->lo_init
[0];
1474 info
->lo_init
[1] = info64
->lo_init
[1];
1475 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1476 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1478 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1479 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1481 /* error in case values were truncated */
1482 if (info
->lo_device
!= info64
->lo_device
||
1483 info
->lo_rdevice
!= info64
->lo_rdevice
||
1484 info
->lo_inode
!= info64
->lo_inode
||
1485 info
->lo_offset
!= info64
->lo_offset
)
1492 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1494 struct loop_info info
;
1495 struct loop_info64 info64
;
1497 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1499 loop_info64_from_old(&info
, &info64
);
1500 return loop_set_status(lo
, &info64
);
1504 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1506 struct loop_info64 info64
;
1508 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1510 return loop_set_status(lo
, &info64
);
1514 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1515 struct loop_info info
;
1516 struct loop_info64 info64
;
1521 err
= loop_get_status(lo
, &info64
);
1523 err
= loop_info64_to_old(&info64
, &info
);
1524 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1531 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1532 struct loop_info64 info64
;
1537 err
= loop_get_status(lo
, &info64
);
1538 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1544 static int loop_set_capacity(struct loop_device
*lo
)
1548 if (unlikely(lo
->lo_state
!= Lo_bound
))
1551 size
= get_loop_size(lo
, lo
->lo_backing_file
);
1552 loop_set_size(lo
, size
);
1557 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1560 if (lo
->lo_state
!= Lo_bound
)
1563 __loop_update_dio(lo
, !!arg
);
1564 if (lo
->use_dio
== !!arg
)
1571 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1575 if (lo
->lo_state
!= Lo_bound
)
1578 if (arg
< 512 || arg
> PAGE_SIZE
|| !is_power_of_2(arg
))
1581 if (lo
->lo_queue
->limits
.logical_block_size
== arg
)
1584 sync_blockdev(lo
->lo_device
);
1585 kill_bdev(lo
->lo_device
);
1587 blk_mq_freeze_queue(lo
->lo_queue
);
1589 /* kill_bdev should have truncated all the pages */
1590 if (lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1592 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1593 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1594 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1598 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1599 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1600 blk_queue_io_min(lo
->lo_queue
, arg
);
1601 loop_update_dio(lo
);
1603 blk_mq_unfreeze_queue(lo
->lo_queue
);
1608 static int lo_simple_ioctl(struct loop_device
*lo
, unsigned int cmd
,
1613 err
= mutex_lock_killable(&loop_ctl_mutex
);
1617 case LOOP_SET_CAPACITY
:
1618 err
= loop_set_capacity(lo
);
1620 case LOOP_SET_DIRECT_IO
:
1621 err
= loop_set_dio(lo
, arg
);
1623 case LOOP_SET_BLOCK_SIZE
:
1624 err
= loop_set_block_size(lo
, arg
);
1627 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1629 mutex_unlock(&loop_ctl_mutex
);
1633 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1634 unsigned int cmd
, unsigned long arg
)
1636 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1641 return loop_set_fd(lo
, mode
, bdev
, arg
);
1642 case LOOP_CHANGE_FD
:
1643 return loop_change_fd(lo
, bdev
, arg
);
1645 return loop_clr_fd(lo
);
1646 case LOOP_SET_STATUS
:
1648 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1649 err
= loop_set_status_old(lo
,
1650 (struct loop_info __user
*)arg
);
1653 case LOOP_GET_STATUS
:
1654 return loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1655 case LOOP_SET_STATUS64
:
1657 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1658 err
= loop_set_status64(lo
,
1659 (struct loop_info64 __user
*) arg
);
1662 case LOOP_GET_STATUS64
:
1663 return loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1664 case LOOP_SET_CAPACITY
:
1665 case LOOP_SET_DIRECT_IO
:
1666 case LOOP_SET_BLOCK_SIZE
:
1667 if (!(mode
& FMODE_WRITE
) && !capable(CAP_SYS_ADMIN
))
1671 err
= lo_simple_ioctl(lo
, cmd
, arg
);
1678 #ifdef CONFIG_COMPAT
1679 struct compat_loop_info
{
1680 compat_int_t lo_number
; /* ioctl r/o */
1681 compat_dev_t lo_device
; /* ioctl r/o */
1682 compat_ulong_t lo_inode
; /* ioctl r/o */
1683 compat_dev_t lo_rdevice
; /* ioctl r/o */
1684 compat_int_t lo_offset
;
1685 compat_int_t lo_encrypt_type
;
1686 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1687 compat_int_t lo_flags
; /* ioctl r/o */
1688 char lo_name
[LO_NAME_SIZE
];
1689 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1690 compat_ulong_t lo_init
[2];
1695 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1696 * - noinlined to reduce stack space usage in main part of driver
1699 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1700 struct loop_info64
*info64
)
1702 struct compat_loop_info info
;
1704 if (copy_from_user(&info
, arg
, sizeof(info
)))
1707 memset(info64
, 0, sizeof(*info64
));
1708 info64
->lo_number
= info
.lo_number
;
1709 info64
->lo_device
= info
.lo_device
;
1710 info64
->lo_inode
= info
.lo_inode
;
1711 info64
->lo_rdevice
= info
.lo_rdevice
;
1712 info64
->lo_offset
= info
.lo_offset
;
1713 info64
->lo_sizelimit
= 0;
1714 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1715 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1716 info64
->lo_flags
= info
.lo_flags
;
1717 info64
->lo_init
[0] = info
.lo_init
[0];
1718 info64
->lo_init
[1] = info
.lo_init
[1];
1719 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1720 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1722 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1723 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1728 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1729 * - noinlined to reduce stack space usage in main part of driver
1732 loop_info64_to_compat(const struct loop_info64
*info64
,
1733 struct compat_loop_info __user
*arg
)
1735 struct compat_loop_info info
;
1737 memset(&info
, 0, sizeof(info
));
1738 info
.lo_number
= info64
->lo_number
;
1739 info
.lo_device
= info64
->lo_device
;
1740 info
.lo_inode
= info64
->lo_inode
;
1741 info
.lo_rdevice
= info64
->lo_rdevice
;
1742 info
.lo_offset
= info64
->lo_offset
;
1743 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1744 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1745 info
.lo_flags
= info64
->lo_flags
;
1746 info
.lo_init
[0] = info64
->lo_init
[0];
1747 info
.lo_init
[1] = info64
->lo_init
[1];
1748 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1749 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1751 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1752 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1754 /* error in case values were truncated */
1755 if (info
.lo_device
!= info64
->lo_device
||
1756 info
.lo_rdevice
!= info64
->lo_rdevice
||
1757 info
.lo_inode
!= info64
->lo_inode
||
1758 info
.lo_offset
!= info64
->lo_offset
||
1759 info
.lo_init
[0] != info64
->lo_init
[0] ||
1760 info
.lo_init
[1] != info64
->lo_init
[1])
1763 if (copy_to_user(arg
, &info
, sizeof(info
)))
1769 loop_set_status_compat(struct loop_device
*lo
,
1770 const struct compat_loop_info __user
*arg
)
1772 struct loop_info64 info64
;
1775 ret
= loop_info64_from_compat(arg
, &info64
);
1778 return loop_set_status(lo
, &info64
);
1782 loop_get_status_compat(struct loop_device
*lo
,
1783 struct compat_loop_info __user
*arg
)
1785 struct loop_info64 info64
;
1790 err
= loop_get_status(lo
, &info64
);
1792 err
= loop_info64_to_compat(&info64
, arg
);
1796 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1797 unsigned int cmd
, unsigned long arg
)
1799 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1803 case LOOP_SET_STATUS
:
1804 err
= loop_set_status_compat(lo
,
1805 (const struct compat_loop_info __user
*)arg
);
1807 case LOOP_GET_STATUS
:
1808 err
= loop_get_status_compat(lo
,
1809 (struct compat_loop_info __user
*)arg
);
1811 case LOOP_SET_CAPACITY
:
1813 case LOOP_GET_STATUS64
:
1814 case LOOP_SET_STATUS64
:
1815 arg
= (unsigned long) compat_ptr(arg
);
1818 case LOOP_CHANGE_FD
:
1819 case LOOP_SET_BLOCK_SIZE
:
1820 case LOOP_SET_DIRECT_IO
:
1821 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1831 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1833 struct loop_device
*lo
;
1836 err
= mutex_lock_killable(&loop_ctl_mutex
);
1839 lo
= bdev
->bd_disk
->private_data
;
1845 atomic_inc(&lo
->lo_refcnt
);
1847 mutex_unlock(&loop_ctl_mutex
);
1851 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1853 struct loop_device
*lo
;
1855 mutex_lock(&loop_ctl_mutex
);
1856 lo
= disk
->private_data
;
1857 if (atomic_dec_return(&lo
->lo_refcnt
))
1860 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1861 if (lo
->lo_state
!= Lo_bound
)
1863 lo
->lo_state
= Lo_rundown
;
1864 mutex_unlock(&loop_ctl_mutex
);
1866 * In autoclear mode, stop the loop thread
1867 * and remove configuration after last close.
1869 __loop_clr_fd(lo
, true);
1871 } else if (lo
->lo_state
== Lo_bound
) {
1873 * Otherwise keep thread (if running) and config,
1874 * but flush possible ongoing bios in thread.
1876 blk_mq_freeze_queue(lo
->lo_queue
);
1877 blk_mq_unfreeze_queue(lo
->lo_queue
);
1881 mutex_unlock(&loop_ctl_mutex
);
1884 static const struct block_device_operations lo_fops
= {
1885 .owner
= THIS_MODULE
,
1887 .release
= lo_release
,
1889 #ifdef CONFIG_COMPAT
1890 .compat_ioctl
= lo_compat_ioctl
,
1895 * And now the modules code and kernel interface.
1897 static int max_loop
;
1898 module_param(max_loop
, int, 0444);
1899 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1900 module_param(max_part
, int, 0444);
1901 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1902 MODULE_LICENSE("GPL");
1903 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1905 int loop_register_transfer(struct loop_func_table
*funcs
)
1907 unsigned int n
= funcs
->number
;
1909 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1911 xfer_funcs
[n
] = funcs
;
1915 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1917 struct loop_device
*lo
= ptr
;
1918 struct loop_func_table
*xfer
= data
;
1920 mutex_lock(&loop_ctl_mutex
);
1921 if (lo
->lo_encryption
== xfer
)
1922 loop_release_xfer(lo
);
1923 mutex_unlock(&loop_ctl_mutex
);
1927 int loop_unregister_transfer(int number
)
1929 unsigned int n
= number
;
1930 struct loop_func_table
*xfer
;
1932 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1935 xfer_funcs
[n
] = NULL
;
1936 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1940 EXPORT_SYMBOL(loop_register_transfer
);
1941 EXPORT_SYMBOL(loop_unregister_transfer
);
1943 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1944 const struct blk_mq_queue_data
*bd
)
1946 struct request
*rq
= bd
->rq
;
1947 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1948 struct loop_device
*lo
= rq
->q
->queuedata
;
1950 blk_mq_start_request(rq
);
1952 if (lo
->lo_state
!= Lo_bound
)
1953 return BLK_STS_IOERR
;
1955 switch (req_op(rq
)) {
1957 case REQ_OP_DISCARD
:
1958 case REQ_OP_WRITE_ZEROES
:
1959 cmd
->use_aio
= false;
1962 cmd
->use_aio
= lo
->use_dio
;
1966 /* always use the first bio's css */
1967 #ifdef CONFIG_BLK_CGROUP
1968 if (cmd
->use_aio
&& rq
->bio
&& rq
->bio
->bi_blkg
) {
1969 cmd
->css
= &bio_blkcg(rq
->bio
)->css
;
1974 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1979 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1981 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1982 const bool write
= op_is_write(req_op(rq
));
1983 struct loop_device
*lo
= rq
->q
->queuedata
;
1986 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1991 ret
= do_req_filebacked(lo
, rq
);
1993 /* complete non-aio request */
1994 if (!cmd
->use_aio
|| ret
) {
1995 if (ret
== -EOPNOTSUPP
)
1998 cmd
->ret
= ret
? -EIO
: 0;
1999 blk_mq_complete_request(rq
);
2003 static void loop_queue_work(struct kthread_work
*work
)
2005 struct loop_cmd
*cmd
=
2006 container_of(work
, struct loop_cmd
, work
);
2008 loop_handle_cmd(cmd
);
2011 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
2012 unsigned int hctx_idx
, unsigned int numa_node
)
2014 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
2016 kthread_init_work(&cmd
->work
, loop_queue_work
);
2020 static const struct blk_mq_ops loop_mq_ops
= {
2021 .queue_rq
= loop_queue_rq
,
2022 .init_request
= loop_init_request
,
2023 .complete
= lo_complete_rq
,
2026 static int loop_add(struct loop_device
**l
, int i
)
2028 struct loop_device
*lo
;
2029 struct gendisk
*disk
;
2033 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
2037 lo
->lo_state
= Lo_unbound
;
2039 /* allocate id, if @id >= 0, we're requesting that specific id */
2041 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
2045 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
2052 lo
->tag_set
.ops
= &loop_mq_ops
;
2053 lo
->tag_set
.nr_hw_queues
= 1;
2054 lo
->tag_set
.queue_depth
= 128;
2055 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
2056 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
2057 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
;
2058 lo
->tag_set
.driver_data
= lo
;
2060 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
2064 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
2065 if (IS_ERR(lo
->lo_queue
)) {
2066 err
= PTR_ERR(lo
->lo_queue
);
2067 goto out_cleanup_tags
;
2069 lo
->lo_queue
->queuedata
= lo
;
2071 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
2074 * By default, we do buffer IO, so it doesn't make sense to enable
2075 * merge because the I/O submitted to backing file is handled page by
2076 * page. For directio mode, merge does help to dispatch bigger request
2077 * to underlayer disk. We will enable merge once directio is enabled.
2079 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
2082 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
2084 goto out_free_queue
;
2087 * Disable partition scanning by default. The in-kernel partition
2088 * scanning can be requested individually per-device during its
2089 * setup. Userspace can always add and remove partitions from all
2090 * devices. The needed partition minors are allocated from the
2091 * extended minor space, the main loop device numbers will continue
2092 * to match the loop minors, regardless of the number of partitions
2095 * If max_part is given, partition scanning is globally enabled for
2096 * all loop devices. The minors for the main loop devices will be
2097 * multiples of max_part.
2099 * Note: Global-for-all-devices, set-only-at-init, read-only module
2100 * parameteters like 'max_loop' and 'max_part' make things needlessly
2101 * complicated, are too static, inflexible and may surprise
2102 * userspace tools. Parameters like this in general should be avoided.
2105 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
2106 disk
->flags
|= GENHD_FL_EXT_DEVT
;
2107 atomic_set(&lo
->lo_refcnt
, 0);
2109 spin_lock_init(&lo
->lo_lock
);
2110 disk
->major
= LOOP_MAJOR
;
2111 disk
->first_minor
= i
<< part_shift
;
2112 disk
->fops
= &lo_fops
;
2113 disk
->private_data
= lo
;
2114 disk
->queue
= lo
->lo_queue
;
2115 sprintf(disk
->disk_name
, "loop%d", i
);
2118 return lo
->lo_number
;
2121 blk_cleanup_queue(lo
->lo_queue
);
2123 blk_mq_free_tag_set(&lo
->tag_set
);
2125 idr_remove(&loop_index_idr
, i
);
2132 static void loop_remove(struct loop_device
*lo
)
2134 del_gendisk(lo
->lo_disk
);
2135 blk_cleanup_queue(lo
->lo_queue
);
2136 blk_mq_free_tag_set(&lo
->tag_set
);
2137 put_disk(lo
->lo_disk
);
2141 static int find_free_cb(int id
, void *ptr
, void *data
)
2143 struct loop_device
*lo
= ptr
;
2144 struct loop_device
**l
= data
;
2146 if (lo
->lo_state
== Lo_unbound
) {
2153 static int loop_lookup(struct loop_device
**l
, int i
)
2155 struct loop_device
*lo
;
2161 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
2164 ret
= lo
->lo_number
;
2169 /* lookup and return a specific i */
2170 lo
= idr_find(&loop_index_idr
, i
);
2173 ret
= lo
->lo_number
;
2179 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
2181 struct loop_device
*lo
;
2182 struct kobject
*kobj
;
2185 mutex_lock(&loop_ctl_mutex
);
2186 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
2188 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
2192 kobj
= get_disk_and_module(lo
->lo_disk
);
2193 mutex_unlock(&loop_ctl_mutex
);
2199 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
2202 struct loop_device
*lo
;
2205 ret
= mutex_lock_killable(&loop_ctl_mutex
);
2212 ret
= loop_lookup(&lo
, parm
);
2217 ret
= loop_add(&lo
, parm
);
2219 case LOOP_CTL_REMOVE
:
2220 ret
= loop_lookup(&lo
, parm
);
2223 if (lo
->lo_state
!= Lo_unbound
) {
2227 if (atomic_read(&lo
->lo_refcnt
) > 0) {
2231 lo
->lo_disk
->private_data
= NULL
;
2232 idr_remove(&loop_index_idr
, lo
->lo_number
);
2235 case LOOP_CTL_GET_FREE
:
2236 ret
= loop_lookup(&lo
, -1);
2239 ret
= loop_add(&lo
, -1);
2241 mutex_unlock(&loop_ctl_mutex
);
2246 static const struct file_operations loop_ctl_fops
= {
2247 .open
= nonseekable_open
,
2248 .unlocked_ioctl
= loop_control_ioctl
,
2249 .compat_ioctl
= loop_control_ioctl
,
2250 .owner
= THIS_MODULE
,
2251 .llseek
= noop_llseek
,
2254 static struct miscdevice loop_misc
= {
2255 .minor
= LOOP_CTRL_MINOR
,
2256 .name
= "loop-control",
2257 .fops
= &loop_ctl_fops
,
2260 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2261 MODULE_ALIAS("devname:loop-control");
2263 static int __init
loop_init(void)
2266 unsigned long range
;
2267 struct loop_device
*lo
;
2272 part_shift
= fls(max_part
);
2275 * Adjust max_part according to part_shift as it is exported
2276 * to user space so that user can decide correct minor number
2277 * if [s]he want to create more devices.
2279 * Note that -1 is required because partition 0 is reserved
2280 * for the whole disk.
2282 max_part
= (1UL << part_shift
) - 1;
2285 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2290 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2296 * If max_loop is specified, create that many devices upfront.
2297 * This also becomes a hard limit. If max_loop is not specified,
2298 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2299 * init time. Loop devices can be requested on-demand with the
2300 * /dev/loop-control interface, or be instantiated by accessing
2301 * a 'dead' device node.
2305 range
= max_loop
<< part_shift
;
2307 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2308 range
= 1UL << MINORBITS
;
2311 err
= misc_register(&loop_misc
);
2316 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2321 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2322 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2324 /* pre-create number of devices given by config or max_loop */
2325 mutex_lock(&loop_ctl_mutex
);
2326 for (i
= 0; i
< nr
; i
++)
2328 mutex_unlock(&loop_ctl_mutex
);
2330 printk(KERN_INFO
"loop: module loaded\n");
2334 misc_deregister(&loop_misc
);
2339 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2341 struct loop_device
*lo
= ptr
;
2347 static void __exit
loop_exit(void)
2349 unsigned long range
;
2351 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2353 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2354 idr_destroy(&loop_index_idr
);
2356 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2357 unregister_blkdev(LOOP_MAJOR
, "loop");
2359 misc_deregister(&loop_misc
);
2362 module_init(loop_init
);
2363 module_exit(loop_exit
);
2366 static int __init
max_loop_setup(char *str
)
2368 max_loop
= simple_strtol(str
, NULL
, 0);
2372 __setup("max_loop=", max_loop_setup
);