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 blk_mq_freeze_queue(lo
->lo_queue
);
218 lo
->use_dio
= use_dio
;
220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
221 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
224 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
226 blk_mq_unfreeze_queue(lo
->lo_queue
);
230 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
232 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
233 sector_t x
= (sector_t
)size
;
234 struct block_device
*bdev
= lo
->lo_device
;
236 if (unlikely((loff_t
)x
!= size
))
238 if (lo
->lo_offset
!= offset
)
239 lo
->lo_offset
= offset
;
240 if (lo
->lo_sizelimit
!= sizelimit
)
241 lo
->lo_sizelimit
= sizelimit
;
242 set_capacity(lo
->lo_disk
, x
);
243 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
244 /* let user-space know about the new size */
245 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
250 lo_do_transfer(struct loop_device
*lo
, int cmd
,
251 struct page
*rpage
, unsigned roffs
,
252 struct page
*lpage
, unsigned loffs
,
253 int size
, sector_t rblock
)
257 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
261 printk_ratelimited(KERN_ERR
262 "loop: Transfer error at byte offset %llu, length %i.\n",
263 (unsigned long long)rblock
<< 9, size
);
267 static inline void loop_iov_iter_bvec(struct iov_iter
*i
,
268 unsigned int direction
, const struct bio_vec
*bvec
,
269 unsigned long nr_segs
, size_t count
)
271 iov_iter_bvec(i
, direction
, bvec
, nr_segs
, count
);
272 i
->type
|= ITER_BVEC_FLAG_NO_REF
;
275 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
280 loop_iov_iter_bvec(&i
, WRITE
, bvec
, 1, bvec
->bv_len
);
282 file_start_write(file
);
283 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
284 file_end_write(file
);
286 if (likely(bw
== bvec
->bv_len
))
289 printk_ratelimited(KERN_ERR
290 "loop: Write error at byte offset %llu, length %i.\n",
291 (unsigned long long)*ppos
, bvec
->bv_len
);
297 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
301 struct req_iterator iter
;
304 rq_for_each_segment(bvec
, rq
, iter
) {
305 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
315 * This is the slow, transforming version that needs to double buffer the
316 * data as it cannot do the transformations in place without having direct
317 * access to the destination pages of the backing file.
319 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
322 struct bio_vec bvec
, b
;
323 struct req_iterator iter
;
327 page
= alloc_page(GFP_NOIO
);
331 rq_for_each_segment(bvec
, rq
, iter
) {
332 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
333 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
339 b
.bv_len
= bvec
.bv_len
;
340 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
349 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
353 struct req_iterator iter
;
357 rq_for_each_segment(bvec
, rq
, iter
) {
358 loop_iov_iter_bvec(&i
, READ
, &bvec
, 1, bvec
.bv_len
);
359 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
363 flush_dcache_page(bvec
.bv_page
);
365 if (len
!= bvec
.bv_len
) {
368 __rq_for_each_bio(bio
, rq
)
378 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
381 struct bio_vec bvec
, b
;
382 struct req_iterator iter
;
388 page
= alloc_page(GFP_NOIO
);
392 rq_for_each_segment(bvec
, rq
, iter
) {
397 b
.bv_len
= bvec
.bv_len
;
399 loop_iov_iter_bvec(&i
, READ
, &b
, 1, b
.bv_len
);
400 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
406 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
407 bvec
.bv_offset
, len
, offset
>> 9);
411 flush_dcache_page(bvec
.bv_page
);
413 if (len
!= bvec
.bv_len
) {
416 __rq_for_each_bio(bio
, rq
)
428 static int lo_discard(struct loop_device
*lo
, struct request
*rq
, loff_t pos
)
431 * We use punch hole to reclaim the free space used by the
432 * image a.k.a. discard. However we do not support discard if
433 * encryption is enabled, because it may give an attacker
434 * useful information.
436 struct file
*file
= lo
->lo_backing_file
;
437 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
440 if ((!file
->f_op
->fallocate
) || lo
->lo_encrypt_key_size
) {
445 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
446 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
452 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
454 struct file
*file
= lo
->lo_backing_file
;
455 int ret
= vfs_fsync(file
, 0);
456 if (unlikely(ret
&& ret
!= -EINVAL
))
462 static void lo_complete_rq(struct request
*rq
)
464 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
465 blk_status_t ret
= BLK_STS_OK
;
467 if (!cmd
->use_aio
|| cmd
->ret
< 0 || cmd
->ret
== blk_rq_bytes(rq
) ||
468 req_op(rq
) != REQ_OP_READ
) {
475 * Short READ - if we got some data, advance our request and
476 * retry it. If we got no data, end the rest with EIO.
479 blk_update_request(rq
, BLK_STS_OK
, cmd
->ret
);
481 blk_mq_requeue_request(rq
, true);
484 struct bio
*bio
= rq
->bio
;
493 blk_mq_end_request(rq
, ret
);
497 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
499 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
501 if (!atomic_dec_and_test(&cmd
->ref
))
505 blk_mq_complete_request(rq
);
508 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
510 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
515 lo_rw_aio_do_completion(cmd
);
518 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
521 struct iov_iter iter
;
522 struct req_iterator rq_iter
;
523 struct bio_vec
*bvec
;
524 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
525 struct bio
*bio
= rq
->bio
;
526 struct file
*file
= lo
->lo_backing_file
;
532 rq_for_each_bvec(tmp
, rq
, rq_iter
)
535 if (rq
->bio
!= rq
->biotail
) {
537 bvec
= kmalloc_array(nr_bvec
, sizeof(struct bio_vec
),
544 * The bios of the request may be started from the middle of
545 * the 'bvec' because of bio splitting, so we can't directly
546 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
547 * API will take care of all details for us.
549 rq_for_each_bvec(tmp
, rq
, rq_iter
) {
557 * Same here, this bio may be started from the middle of the
558 * 'bvec' because of bio splitting, so offset from the bvec
559 * must be passed to iov iterator
561 offset
= bio
->bi_iter
.bi_bvec_done
;
562 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
564 atomic_set(&cmd
->ref
, 2);
566 loop_iov_iter_bvec(&iter
, rw
, bvec
, nr_bvec
, blk_rq_bytes(rq
));
567 iter
.iov_offset
= offset
;
569 cmd
->iocb
.ki_pos
= pos
;
570 cmd
->iocb
.ki_filp
= file
;
571 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
572 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
573 cmd
->iocb
.ki_ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
575 kthread_associate_blkcg(cmd
->css
);
578 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
580 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
582 lo_rw_aio_do_completion(cmd
);
583 kthread_associate_blkcg(NULL
);
585 if (ret
!= -EIOCBQUEUED
)
586 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
590 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
592 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
593 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
596 * lo_write_simple and lo_read_simple should have been covered
597 * by io submit style function like lo_rw_aio(), one blocker
598 * is that lo_read_simple() need to call flush_dcache_page after
599 * the page is written from kernel, and it isn't easy to handle
600 * this in io submit style function which submits all segments
601 * of the req at one time. And direct read IO doesn't need to
602 * run flush_dcache_page().
604 switch (req_op(rq
)) {
606 return lo_req_flush(lo
, rq
);
608 case REQ_OP_WRITE_ZEROES
:
609 return lo_discard(lo
, rq
, pos
);
612 return lo_write_transfer(lo
, rq
, pos
);
613 else if (cmd
->use_aio
)
614 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
616 return lo_write_simple(lo
, rq
, pos
);
619 return lo_read_transfer(lo
, rq
, pos
);
620 else if (cmd
->use_aio
)
621 return lo_rw_aio(lo
, cmd
, pos
, READ
);
623 return lo_read_simple(lo
, rq
, pos
);
630 static inline void loop_update_dio(struct loop_device
*lo
)
632 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
636 static void loop_reread_partitions(struct loop_device
*lo
,
637 struct block_device
*bdev
)
641 rc
= blkdev_reread_part(bdev
);
643 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
644 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
647 static inline int is_loop_device(struct file
*file
)
649 struct inode
*i
= file
->f_mapping
->host
;
651 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
654 static int loop_validate_file(struct file
*file
, struct block_device
*bdev
)
656 struct inode
*inode
= file
->f_mapping
->host
;
657 struct file
*f
= file
;
659 /* Avoid recursion */
660 while (is_loop_device(f
)) {
661 struct loop_device
*l
;
663 if (f
->f_mapping
->host
->i_bdev
== bdev
)
666 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
667 if (l
->lo_state
!= Lo_bound
) {
670 f
= l
->lo_backing_file
;
672 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
678 * loop_change_fd switched the backing store of a loopback device to
679 * a new file. This is useful for operating system installers to free up
680 * the original file and in High Availability environments to switch to
681 * an alternative location for the content in case of server meltdown.
682 * This can only work if the loop device is used read-only, and if the
683 * new backing store is the same size and type as the old backing store.
685 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
688 struct file
*file
= NULL
, *old_file
;
692 error
= mutex_lock_killable(&loop_ctl_mutex
);
696 if (lo
->lo_state
!= Lo_bound
)
699 /* the loop device has to be read-only */
701 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
709 error
= loop_validate_file(file
, bdev
);
713 old_file
= lo
->lo_backing_file
;
717 /* size of the new backing store needs to be the same */
718 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
722 blk_mq_freeze_queue(lo
->lo_queue
);
723 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
724 lo
->lo_backing_file
= file
;
725 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
726 mapping_set_gfp_mask(file
->f_mapping
,
727 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
729 blk_mq_unfreeze_queue(lo
->lo_queue
);
730 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
731 mutex_unlock(&loop_ctl_mutex
);
733 * We must drop file reference outside of loop_ctl_mutex as dropping
734 * the file ref can take bd_mutex which creates circular locking
739 loop_reread_partitions(lo
, bdev
);
743 mutex_unlock(&loop_ctl_mutex
);
749 /* loop sysfs attributes */
751 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
752 ssize_t (*callback
)(struct loop_device
*, char *))
754 struct gendisk
*disk
= dev_to_disk(dev
);
755 struct loop_device
*lo
= disk
->private_data
;
757 return callback(lo
, page
);
760 #define LOOP_ATTR_RO(_name) \
761 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
762 static ssize_t loop_attr_do_show_##_name(struct device *d, \
763 struct device_attribute *attr, char *b) \
765 return loop_attr_show(d, b, loop_attr_##_name##_show); \
767 static struct device_attribute loop_attr_##_name = \
768 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
770 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
775 spin_lock_irq(&lo
->lo_lock
);
776 if (lo
->lo_backing_file
)
777 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
778 spin_unlock_irq(&lo
->lo_lock
);
780 if (IS_ERR_OR_NULL(p
))
784 memmove(buf
, p
, ret
);
792 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
794 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
797 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
799 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
802 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
804 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
806 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
809 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
811 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
813 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
816 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
818 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
820 return sprintf(buf
, "%s\n", dio
? "1" : "0");
823 LOOP_ATTR_RO(backing_file
);
824 LOOP_ATTR_RO(offset
);
825 LOOP_ATTR_RO(sizelimit
);
826 LOOP_ATTR_RO(autoclear
);
827 LOOP_ATTR_RO(partscan
);
830 static struct attribute
*loop_attrs
[] = {
831 &loop_attr_backing_file
.attr
,
832 &loop_attr_offset
.attr
,
833 &loop_attr_sizelimit
.attr
,
834 &loop_attr_autoclear
.attr
,
835 &loop_attr_partscan
.attr
,
840 static struct attribute_group loop_attribute_group
= {
845 static void loop_sysfs_init(struct loop_device
*lo
)
847 lo
->sysfs_inited
= !sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
848 &loop_attribute_group
);
851 static void loop_sysfs_exit(struct loop_device
*lo
)
853 if (lo
->sysfs_inited
)
854 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
855 &loop_attribute_group
);
858 static void loop_config_discard(struct loop_device
*lo
)
860 struct file
*file
= lo
->lo_backing_file
;
861 struct inode
*inode
= file
->f_mapping
->host
;
862 struct request_queue
*q
= lo
->lo_queue
;
865 * We use punch hole to reclaim the free space used by the
866 * image a.k.a. discard. However we do not support discard if
867 * encryption is enabled, because it may give an attacker
868 * useful information.
870 if ((!file
->f_op
->fallocate
) ||
871 lo
->lo_encrypt_key_size
) {
872 q
->limits
.discard_granularity
= 0;
873 q
->limits
.discard_alignment
= 0;
874 blk_queue_max_discard_sectors(q
, 0);
875 blk_queue_max_write_zeroes_sectors(q
, 0);
876 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, q
);
880 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
881 q
->limits
.discard_alignment
= 0;
883 blk_queue_max_discard_sectors(q
, UINT_MAX
>> 9);
884 blk_queue_max_write_zeroes_sectors(q
, UINT_MAX
>> 9);
885 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, q
);
888 static void loop_unprepare_queue(struct loop_device
*lo
)
890 kthread_flush_worker(&lo
->worker
);
891 kthread_stop(lo
->worker_task
);
894 static int loop_kthread_worker_fn(void *worker_ptr
)
896 current
->flags
|= PF_LESS_THROTTLE
;
897 return kthread_worker_fn(worker_ptr
);
900 static int loop_prepare_queue(struct loop_device
*lo
)
902 kthread_init_worker(&lo
->worker
);
903 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
904 &lo
->worker
, "loop%d", lo
->lo_number
);
905 if (IS_ERR(lo
->worker_task
))
907 set_user_nice(lo
->worker_task
, MIN_NICE
);
911 static void loop_update_rotational(struct loop_device
*lo
)
913 struct file
*file
= lo
->lo_backing_file
;
914 struct inode
*file_inode
= file
->f_mapping
->host
;
915 struct block_device
*file_bdev
= file_inode
->i_sb
->s_bdev
;
916 struct request_queue
*q
= lo
->lo_queue
;
919 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
921 nonrot
= blk_queue_nonrot(bdev_get_queue(file_bdev
));
924 blk_queue_flag_set(QUEUE_FLAG_NONROT
, q
);
926 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, q
);
929 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
930 struct block_device
*bdev
, unsigned int arg
)
934 struct address_space
*mapping
;
940 /* This is safe, since we have a reference from open(). */
941 __module_get(THIS_MODULE
);
949 * If we don't hold exclusive handle for the device, upgrade to it
950 * here to avoid changing device under exclusive owner.
952 if (!(mode
& FMODE_EXCL
)) {
954 error
= blkdev_get(bdev
, mode
| FMODE_EXCL
, loop_set_fd
);
959 error
= mutex_lock_killable(&loop_ctl_mutex
);
964 if (lo
->lo_state
!= Lo_unbound
)
967 error
= loop_validate_file(file
, bdev
);
971 mapping
= file
->f_mapping
;
972 inode
= mapping
->host
;
974 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
975 !file
->f_op
->write_iter
)
976 lo_flags
|= LO_FLAGS_READ_ONLY
;
979 size
= get_loop_size(lo
, file
);
980 if ((loff_t
)(sector_t
)size
!= size
)
982 error
= loop_prepare_queue(lo
);
988 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
991 lo
->lo_device
= bdev
;
992 lo
->lo_flags
= lo_flags
;
993 lo
->lo_backing_file
= file
;
996 lo
->lo_sizelimit
= 0;
997 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
998 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
1000 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
1001 blk_queue_write_cache(lo
->lo_queue
, true, false);
1003 loop_update_rotational(lo
);
1004 loop_update_dio(lo
);
1005 set_capacity(lo
->lo_disk
, size
);
1006 bd_set_size(bdev
, size
<< 9);
1007 loop_sysfs_init(lo
);
1008 /* let user-space know about the new size */
1009 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1011 set_blocksize(bdev
, S_ISBLK(inode
->i_mode
) ?
1012 block_size(inode
->i_bdev
) : PAGE_SIZE
);
1014 lo
->lo_state
= Lo_bound
;
1016 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1017 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
1019 /* Grab the block_device to prevent its destruction after we
1020 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1023 mutex_unlock(&loop_ctl_mutex
);
1025 loop_reread_partitions(lo
, bdev
);
1026 if (!(mode
& FMODE_EXCL
))
1027 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1031 mutex_unlock(&loop_ctl_mutex
);
1033 if (!(mode
& FMODE_EXCL
))
1034 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1038 /* This is safe: open() is still holding a reference. */
1039 module_put(THIS_MODULE
);
1044 loop_release_xfer(struct loop_device
*lo
)
1047 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1051 err
= xfer
->release(lo
);
1052 lo
->transfer
= NULL
;
1053 lo
->lo_encryption
= NULL
;
1054 module_put(xfer
->owner
);
1060 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1061 const struct loop_info64
*i
)
1066 struct module
*owner
= xfer
->owner
;
1068 if (!try_module_get(owner
))
1071 err
= xfer
->init(lo
, i
);
1075 lo
->lo_encryption
= xfer
;
1080 static int __loop_clr_fd(struct loop_device
*lo
, bool release
)
1082 struct file
*filp
= NULL
;
1083 gfp_t gfp
= lo
->old_gfp_mask
;
1084 struct block_device
*bdev
= lo
->lo_device
;
1086 bool partscan
= false;
1089 mutex_lock(&loop_ctl_mutex
);
1090 if (WARN_ON_ONCE(lo
->lo_state
!= Lo_rundown
)) {
1095 filp
= lo
->lo_backing_file
;
1101 /* freeze request queue during the transition */
1102 blk_mq_freeze_queue(lo
->lo_queue
);
1104 spin_lock_irq(&lo
->lo_lock
);
1105 lo
->lo_backing_file
= NULL
;
1106 spin_unlock_irq(&lo
->lo_lock
);
1108 loop_release_xfer(lo
);
1109 lo
->transfer
= NULL
;
1111 lo
->lo_device
= NULL
;
1112 lo
->lo_encryption
= NULL
;
1114 lo
->lo_sizelimit
= 0;
1115 lo
->lo_encrypt_key_size
= 0;
1116 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1117 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1118 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1119 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1120 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1121 blk_queue_io_min(lo
->lo_queue
, 512);
1124 invalidate_bdev(bdev
);
1125 bdev
->bd_inode
->i_mapping
->wb_err
= 0;
1127 set_capacity(lo
->lo_disk
, 0);
1128 loop_sysfs_exit(lo
);
1130 bd_set_size(bdev
, 0);
1131 /* let user-space know about this change */
1132 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1134 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1135 /* This is safe: open() is still holding a reference. */
1136 module_put(THIS_MODULE
);
1137 blk_mq_unfreeze_queue(lo
->lo_queue
);
1139 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
;
1140 lo_number
= lo
->lo_number
;
1141 loop_unprepare_queue(lo
);
1143 mutex_unlock(&loop_ctl_mutex
);
1146 * bd_mutex has been held already in release path, so don't
1147 * acquire it if this function is called in such case.
1149 * If the reread partition isn't from release path, lo_refcnt
1150 * must be at least one and it can only become zero when the
1151 * current holder is released.
1154 err
= __blkdev_reread_part(bdev
);
1156 err
= blkdev_reread_part(bdev
);
1158 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1159 __func__
, lo_number
, err
);
1160 /* Device is gone, no point in returning error */
1165 * lo->lo_state is set to Lo_unbound here after above partscan has
1168 * There cannot be anybody else entering __loop_clr_fd() as
1169 * lo->lo_backing_file is already cleared and Lo_rundown state
1170 * protects us from all the other places trying to change the 'lo'
1173 mutex_lock(&loop_ctl_mutex
);
1176 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1177 lo
->lo_state
= Lo_unbound
;
1178 mutex_unlock(&loop_ctl_mutex
);
1181 * Need not hold loop_ctl_mutex to fput backing file.
1182 * Calling fput holding loop_ctl_mutex triggers a circular
1183 * lock dependency possibility warning as fput can take
1184 * bd_mutex which is usually taken before loop_ctl_mutex.
1191 static int loop_clr_fd(struct loop_device
*lo
)
1195 err
= mutex_lock_killable(&loop_ctl_mutex
);
1198 if (lo
->lo_state
!= Lo_bound
) {
1199 mutex_unlock(&loop_ctl_mutex
);
1203 * If we've explicitly asked to tear down the loop device,
1204 * and it has an elevated reference count, set it for auto-teardown when
1205 * the last reference goes away. This stops $!~#$@ udev from
1206 * preventing teardown because it decided that it needs to run blkid on
1207 * the loopback device whenever they appear. xfstests is notorious for
1208 * failing tests because blkid via udev races with a losetup
1209 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1210 * command to fail with EBUSY.
1212 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1213 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1214 mutex_unlock(&loop_ctl_mutex
);
1217 lo
->lo_state
= Lo_rundown
;
1218 mutex_unlock(&loop_ctl_mutex
);
1220 return __loop_clr_fd(lo
, false);
1224 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1227 struct loop_func_table
*xfer
;
1228 kuid_t uid
= current_uid();
1229 struct block_device
*bdev
;
1230 bool partscan
= false;
1232 err
= mutex_lock_killable(&loop_ctl_mutex
);
1235 if (lo
->lo_encrypt_key_size
&&
1236 !uid_eq(lo
->lo_key_owner
, uid
) &&
1237 !capable(CAP_SYS_ADMIN
)) {
1241 if (lo
->lo_state
!= Lo_bound
) {
1245 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
) {
1250 if (lo
->lo_offset
!= info
->lo_offset
||
1251 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1252 sync_blockdev(lo
->lo_device
);
1253 kill_bdev(lo
->lo_device
);
1256 /* I/O need to be drained during transfer transition */
1257 blk_mq_freeze_queue(lo
->lo_queue
);
1259 err
= loop_release_xfer(lo
);
1263 if (info
->lo_encrypt_type
) {
1264 unsigned int type
= info
->lo_encrypt_type
;
1266 if (type
>= MAX_LO_CRYPT
) {
1270 xfer
= xfer_funcs
[type
];
1278 err
= loop_init_xfer(lo
, xfer
, info
);
1282 if (lo
->lo_offset
!= info
->lo_offset
||
1283 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1284 /* kill_bdev should have truncated all the pages */
1285 if (lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1287 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1288 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1289 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1292 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
)) {
1298 loop_config_discard(lo
);
1300 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1301 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1302 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1303 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1307 lo
->transfer
= xfer
->transfer
;
1308 lo
->ioctl
= xfer
->ioctl
;
1310 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1311 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1312 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1314 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1315 lo
->lo_init
[0] = info
->lo_init
[0];
1316 lo
->lo_init
[1] = info
->lo_init
[1];
1317 if (info
->lo_encrypt_key_size
) {
1318 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1319 info
->lo_encrypt_key_size
);
1320 lo
->lo_key_owner
= uid
;
1323 /* update dio if lo_offset or transfer is changed */
1324 __loop_update_dio(lo
, lo
->use_dio
);
1327 blk_mq_unfreeze_queue(lo
->lo_queue
);
1329 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1330 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1331 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1332 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1333 bdev
= lo
->lo_device
;
1337 mutex_unlock(&loop_ctl_mutex
);
1339 loop_reread_partitions(lo
, bdev
);
1345 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1351 ret
= mutex_lock_killable(&loop_ctl_mutex
);
1354 if (lo
->lo_state
!= Lo_bound
) {
1355 mutex_unlock(&loop_ctl_mutex
);
1359 memset(info
, 0, sizeof(*info
));
1360 info
->lo_number
= lo
->lo_number
;
1361 info
->lo_offset
= lo
->lo_offset
;
1362 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1363 info
->lo_flags
= lo
->lo_flags
;
1364 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1365 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1366 info
->lo_encrypt_type
=
1367 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1368 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1369 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1370 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1371 lo
->lo_encrypt_key_size
);
1374 /* Drop loop_ctl_mutex while we call into the filesystem. */
1375 path
= lo
->lo_backing_file
->f_path
;
1377 mutex_unlock(&loop_ctl_mutex
);
1378 ret
= vfs_getattr(&path
, &stat
, STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1380 info
->lo_device
= huge_encode_dev(stat
.dev
);
1381 info
->lo_inode
= stat
.ino
;
1382 info
->lo_rdevice
= huge_encode_dev(stat
.rdev
);
1389 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1391 memset(info64
, 0, sizeof(*info64
));
1392 info64
->lo_number
= info
->lo_number
;
1393 info64
->lo_device
= info
->lo_device
;
1394 info64
->lo_inode
= info
->lo_inode
;
1395 info64
->lo_rdevice
= info
->lo_rdevice
;
1396 info64
->lo_offset
= info
->lo_offset
;
1397 info64
->lo_sizelimit
= 0;
1398 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1399 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1400 info64
->lo_flags
= info
->lo_flags
;
1401 info64
->lo_init
[0] = info
->lo_init
[0];
1402 info64
->lo_init
[1] = info
->lo_init
[1];
1403 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1404 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1406 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1407 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1411 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1413 memset(info
, 0, sizeof(*info
));
1414 info
->lo_number
= info64
->lo_number
;
1415 info
->lo_device
= info64
->lo_device
;
1416 info
->lo_inode
= info64
->lo_inode
;
1417 info
->lo_rdevice
= info64
->lo_rdevice
;
1418 info
->lo_offset
= info64
->lo_offset
;
1419 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1420 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1421 info
->lo_flags
= info64
->lo_flags
;
1422 info
->lo_init
[0] = info64
->lo_init
[0];
1423 info
->lo_init
[1] = info64
->lo_init
[1];
1424 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1425 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1427 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1428 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1430 /* error in case values were truncated */
1431 if (info
->lo_device
!= info64
->lo_device
||
1432 info
->lo_rdevice
!= info64
->lo_rdevice
||
1433 info
->lo_inode
!= info64
->lo_inode
||
1434 info
->lo_offset
!= info64
->lo_offset
)
1441 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1443 struct loop_info info
;
1444 struct loop_info64 info64
;
1446 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1448 loop_info64_from_old(&info
, &info64
);
1449 return loop_set_status(lo
, &info64
);
1453 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1455 struct loop_info64 info64
;
1457 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1459 return loop_set_status(lo
, &info64
);
1463 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1464 struct loop_info info
;
1465 struct loop_info64 info64
;
1470 err
= loop_get_status(lo
, &info64
);
1472 err
= loop_info64_to_old(&info64
, &info
);
1473 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1480 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1481 struct loop_info64 info64
;
1486 err
= loop_get_status(lo
, &info64
);
1487 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1493 static int loop_set_capacity(struct loop_device
*lo
)
1495 if (unlikely(lo
->lo_state
!= Lo_bound
))
1498 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1501 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1504 if (lo
->lo_state
!= Lo_bound
)
1507 __loop_update_dio(lo
, !!arg
);
1508 if (lo
->use_dio
== !!arg
)
1515 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1519 if (lo
->lo_state
!= Lo_bound
)
1522 if (arg
< 512 || arg
> PAGE_SIZE
|| !is_power_of_2(arg
))
1525 if (lo
->lo_queue
->limits
.logical_block_size
!= arg
) {
1526 sync_blockdev(lo
->lo_device
);
1527 kill_bdev(lo
->lo_device
);
1530 blk_mq_freeze_queue(lo
->lo_queue
);
1532 /* kill_bdev should have truncated all the pages */
1533 if (lo
->lo_queue
->limits
.logical_block_size
!= arg
&&
1534 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1536 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1537 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1538 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1542 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1543 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1544 blk_queue_io_min(lo
->lo_queue
, arg
);
1545 loop_update_dio(lo
);
1547 blk_mq_unfreeze_queue(lo
->lo_queue
);
1552 static int lo_simple_ioctl(struct loop_device
*lo
, unsigned int cmd
,
1557 err
= mutex_lock_killable(&loop_ctl_mutex
);
1561 case LOOP_SET_CAPACITY
:
1562 err
= loop_set_capacity(lo
);
1564 case LOOP_SET_DIRECT_IO
:
1565 err
= loop_set_dio(lo
, arg
);
1567 case LOOP_SET_BLOCK_SIZE
:
1568 err
= loop_set_block_size(lo
, arg
);
1571 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1573 mutex_unlock(&loop_ctl_mutex
);
1577 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1578 unsigned int cmd
, unsigned long arg
)
1580 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1585 return loop_set_fd(lo
, mode
, bdev
, arg
);
1586 case LOOP_CHANGE_FD
:
1587 return loop_change_fd(lo
, bdev
, arg
);
1589 return loop_clr_fd(lo
);
1590 case LOOP_SET_STATUS
:
1592 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1593 err
= loop_set_status_old(lo
,
1594 (struct loop_info __user
*)arg
);
1597 case LOOP_GET_STATUS
:
1598 return loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1599 case LOOP_SET_STATUS64
:
1601 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1602 err
= loop_set_status64(lo
,
1603 (struct loop_info64 __user
*) arg
);
1606 case LOOP_GET_STATUS64
:
1607 return loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1608 case LOOP_SET_CAPACITY
:
1609 case LOOP_SET_DIRECT_IO
:
1610 case LOOP_SET_BLOCK_SIZE
:
1611 if (!(mode
& FMODE_WRITE
) && !capable(CAP_SYS_ADMIN
))
1615 err
= lo_simple_ioctl(lo
, cmd
, arg
);
1622 #ifdef CONFIG_COMPAT
1623 struct compat_loop_info
{
1624 compat_int_t lo_number
; /* ioctl r/o */
1625 compat_dev_t lo_device
; /* ioctl r/o */
1626 compat_ulong_t lo_inode
; /* ioctl r/o */
1627 compat_dev_t lo_rdevice
; /* ioctl r/o */
1628 compat_int_t lo_offset
;
1629 compat_int_t lo_encrypt_type
;
1630 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1631 compat_int_t lo_flags
; /* ioctl r/o */
1632 char lo_name
[LO_NAME_SIZE
];
1633 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1634 compat_ulong_t lo_init
[2];
1639 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1640 * - noinlined to reduce stack space usage in main part of driver
1643 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1644 struct loop_info64
*info64
)
1646 struct compat_loop_info info
;
1648 if (copy_from_user(&info
, arg
, sizeof(info
)))
1651 memset(info64
, 0, sizeof(*info64
));
1652 info64
->lo_number
= info
.lo_number
;
1653 info64
->lo_device
= info
.lo_device
;
1654 info64
->lo_inode
= info
.lo_inode
;
1655 info64
->lo_rdevice
= info
.lo_rdevice
;
1656 info64
->lo_offset
= info
.lo_offset
;
1657 info64
->lo_sizelimit
= 0;
1658 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1659 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1660 info64
->lo_flags
= info
.lo_flags
;
1661 info64
->lo_init
[0] = info
.lo_init
[0];
1662 info64
->lo_init
[1] = info
.lo_init
[1];
1663 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1664 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1666 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1667 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1672 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1673 * - noinlined to reduce stack space usage in main part of driver
1676 loop_info64_to_compat(const struct loop_info64
*info64
,
1677 struct compat_loop_info __user
*arg
)
1679 struct compat_loop_info info
;
1681 memset(&info
, 0, sizeof(info
));
1682 info
.lo_number
= info64
->lo_number
;
1683 info
.lo_device
= info64
->lo_device
;
1684 info
.lo_inode
= info64
->lo_inode
;
1685 info
.lo_rdevice
= info64
->lo_rdevice
;
1686 info
.lo_offset
= info64
->lo_offset
;
1687 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1688 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1689 info
.lo_flags
= info64
->lo_flags
;
1690 info
.lo_init
[0] = info64
->lo_init
[0];
1691 info
.lo_init
[1] = info64
->lo_init
[1];
1692 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1693 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1695 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1696 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1698 /* error in case values were truncated */
1699 if (info
.lo_device
!= info64
->lo_device
||
1700 info
.lo_rdevice
!= info64
->lo_rdevice
||
1701 info
.lo_inode
!= info64
->lo_inode
||
1702 info
.lo_offset
!= info64
->lo_offset
||
1703 info
.lo_init
[0] != info64
->lo_init
[0] ||
1704 info
.lo_init
[1] != info64
->lo_init
[1])
1707 if (copy_to_user(arg
, &info
, sizeof(info
)))
1713 loop_set_status_compat(struct loop_device
*lo
,
1714 const struct compat_loop_info __user
*arg
)
1716 struct loop_info64 info64
;
1719 ret
= loop_info64_from_compat(arg
, &info64
);
1722 return loop_set_status(lo
, &info64
);
1726 loop_get_status_compat(struct loop_device
*lo
,
1727 struct compat_loop_info __user
*arg
)
1729 struct loop_info64 info64
;
1734 err
= loop_get_status(lo
, &info64
);
1736 err
= loop_info64_to_compat(&info64
, arg
);
1740 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1741 unsigned int cmd
, unsigned long arg
)
1743 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1747 case LOOP_SET_STATUS
:
1748 err
= loop_set_status_compat(lo
,
1749 (const struct compat_loop_info __user
*)arg
);
1751 case LOOP_GET_STATUS
:
1752 err
= loop_get_status_compat(lo
,
1753 (struct compat_loop_info __user
*)arg
);
1755 case LOOP_SET_CAPACITY
:
1757 case LOOP_GET_STATUS64
:
1758 case LOOP_SET_STATUS64
:
1759 arg
= (unsigned long) compat_ptr(arg
);
1762 case LOOP_CHANGE_FD
:
1763 case LOOP_SET_BLOCK_SIZE
:
1764 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1774 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1776 struct loop_device
*lo
;
1779 err
= mutex_lock_killable(&loop_ctl_mutex
);
1782 lo
= bdev
->bd_disk
->private_data
;
1788 atomic_inc(&lo
->lo_refcnt
);
1790 mutex_unlock(&loop_ctl_mutex
);
1794 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1796 struct loop_device
*lo
;
1798 mutex_lock(&loop_ctl_mutex
);
1799 lo
= disk
->private_data
;
1800 if (atomic_dec_return(&lo
->lo_refcnt
))
1803 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1804 if (lo
->lo_state
!= Lo_bound
)
1806 lo
->lo_state
= Lo_rundown
;
1807 mutex_unlock(&loop_ctl_mutex
);
1809 * In autoclear mode, stop the loop thread
1810 * and remove configuration after last close.
1812 __loop_clr_fd(lo
, true);
1814 } else if (lo
->lo_state
== Lo_bound
) {
1816 * Otherwise keep thread (if running) and config,
1817 * but flush possible ongoing bios in thread.
1819 blk_mq_freeze_queue(lo
->lo_queue
);
1820 blk_mq_unfreeze_queue(lo
->lo_queue
);
1824 mutex_unlock(&loop_ctl_mutex
);
1827 static const struct block_device_operations lo_fops
= {
1828 .owner
= THIS_MODULE
,
1830 .release
= lo_release
,
1832 #ifdef CONFIG_COMPAT
1833 .compat_ioctl
= lo_compat_ioctl
,
1838 * And now the modules code and kernel interface.
1840 static int max_loop
;
1841 module_param(max_loop
, int, 0444);
1842 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1843 module_param(max_part
, int, 0444);
1844 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1845 MODULE_LICENSE("GPL");
1846 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1848 int loop_register_transfer(struct loop_func_table
*funcs
)
1850 unsigned int n
= funcs
->number
;
1852 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1854 xfer_funcs
[n
] = funcs
;
1858 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1860 struct loop_device
*lo
= ptr
;
1861 struct loop_func_table
*xfer
= data
;
1863 mutex_lock(&loop_ctl_mutex
);
1864 if (lo
->lo_encryption
== xfer
)
1865 loop_release_xfer(lo
);
1866 mutex_unlock(&loop_ctl_mutex
);
1870 int loop_unregister_transfer(int number
)
1872 unsigned int n
= number
;
1873 struct loop_func_table
*xfer
;
1875 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1878 xfer_funcs
[n
] = NULL
;
1879 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1883 EXPORT_SYMBOL(loop_register_transfer
);
1884 EXPORT_SYMBOL(loop_unregister_transfer
);
1886 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1887 const struct blk_mq_queue_data
*bd
)
1889 struct request
*rq
= bd
->rq
;
1890 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1891 struct loop_device
*lo
= rq
->q
->queuedata
;
1893 blk_mq_start_request(rq
);
1895 if (lo
->lo_state
!= Lo_bound
)
1896 return BLK_STS_IOERR
;
1898 switch (req_op(rq
)) {
1900 case REQ_OP_DISCARD
:
1901 case REQ_OP_WRITE_ZEROES
:
1902 cmd
->use_aio
= false;
1905 cmd
->use_aio
= lo
->use_dio
;
1909 /* always use the first bio's css */
1910 #ifdef CONFIG_BLK_CGROUP
1911 if (cmd
->use_aio
&& rq
->bio
&& rq
->bio
->bi_blkg
) {
1912 cmd
->css
= &bio_blkcg(rq
->bio
)->css
;
1917 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1922 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1924 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1925 const bool write
= op_is_write(req_op(rq
));
1926 struct loop_device
*lo
= rq
->q
->queuedata
;
1929 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1934 ret
= do_req_filebacked(lo
, rq
);
1936 /* complete non-aio request */
1937 if (!cmd
->use_aio
|| ret
) {
1938 cmd
->ret
= ret
? -EIO
: 0;
1939 blk_mq_complete_request(rq
);
1943 static void loop_queue_work(struct kthread_work
*work
)
1945 struct loop_cmd
*cmd
=
1946 container_of(work
, struct loop_cmd
, work
);
1948 loop_handle_cmd(cmd
);
1951 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1952 unsigned int hctx_idx
, unsigned int numa_node
)
1954 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1956 kthread_init_work(&cmd
->work
, loop_queue_work
);
1960 static const struct blk_mq_ops loop_mq_ops
= {
1961 .queue_rq
= loop_queue_rq
,
1962 .init_request
= loop_init_request
,
1963 .complete
= lo_complete_rq
,
1966 static int loop_add(struct loop_device
**l
, int i
)
1968 struct loop_device
*lo
;
1969 struct gendisk
*disk
;
1973 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1977 lo
->lo_state
= Lo_unbound
;
1979 /* allocate id, if @id >= 0, we're requesting that specific id */
1981 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1985 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1992 lo
->tag_set
.ops
= &loop_mq_ops
;
1993 lo
->tag_set
.nr_hw_queues
= 1;
1994 lo
->tag_set
.queue_depth
= 128;
1995 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1996 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1997 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
;
1998 lo
->tag_set
.driver_data
= lo
;
2000 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
2004 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
2005 if (IS_ERR(lo
->lo_queue
)) {
2006 err
= PTR_ERR(lo
->lo_queue
);
2007 goto out_cleanup_tags
;
2009 lo
->lo_queue
->queuedata
= lo
;
2011 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
2014 * By default, we do buffer IO, so it doesn't make sense to enable
2015 * merge because the I/O submitted to backing file is handled page by
2016 * page. For directio mode, merge does help to dispatch bigger request
2017 * to underlayer disk. We will enable merge once directio is enabled.
2019 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
2022 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
2024 goto out_free_queue
;
2027 * Disable partition scanning by default. The in-kernel partition
2028 * scanning can be requested individually per-device during its
2029 * setup. Userspace can always add and remove partitions from all
2030 * devices. The needed partition minors are allocated from the
2031 * extended minor space, the main loop device numbers will continue
2032 * to match the loop minors, regardless of the number of partitions
2035 * If max_part is given, partition scanning is globally enabled for
2036 * all loop devices. The minors for the main loop devices will be
2037 * multiples of max_part.
2039 * Note: Global-for-all-devices, set-only-at-init, read-only module
2040 * parameteters like 'max_loop' and 'max_part' make things needlessly
2041 * complicated, are too static, inflexible and may surprise
2042 * userspace tools. Parameters like this in general should be avoided.
2045 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
2046 disk
->flags
|= GENHD_FL_EXT_DEVT
;
2047 atomic_set(&lo
->lo_refcnt
, 0);
2049 spin_lock_init(&lo
->lo_lock
);
2050 disk
->major
= LOOP_MAJOR
;
2051 disk
->first_minor
= i
<< part_shift
;
2052 disk
->fops
= &lo_fops
;
2053 disk
->private_data
= lo
;
2054 disk
->queue
= lo
->lo_queue
;
2055 sprintf(disk
->disk_name
, "loop%d", i
);
2058 return lo
->lo_number
;
2061 blk_cleanup_queue(lo
->lo_queue
);
2063 blk_mq_free_tag_set(&lo
->tag_set
);
2065 idr_remove(&loop_index_idr
, i
);
2072 static void loop_remove(struct loop_device
*lo
)
2074 del_gendisk(lo
->lo_disk
);
2075 blk_cleanup_queue(lo
->lo_queue
);
2076 blk_mq_free_tag_set(&lo
->tag_set
);
2077 put_disk(lo
->lo_disk
);
2081 static int find_free_cb(int id
, void *ptr
, void *data
)
2083 struct loop_device
*lo
= ptr
;
2084 struct loop_device
**l
= data
;
2086 if (lo
->lo_state
== Lo_unbound
) {
2093 static int loop_lookup(struct loop_device
**l
, int i
)
2095 struct loop_device
*lo
;
2101 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
2104 ret
= lo
->lo_number
;
2109 /* lookup and return a specific i */
2110 lo
= idr_find(&loop_index_idr
, i
);
2113 ret
= lo
->lo_number
;
2119 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
2121 struct loop_device
*lo
;
2122 struct kobject
*kobj
;
2125 mutex_lock(&loop_ctl_mutex
);
2126 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
2128 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
2132 kobj
= get_disk_and_module(lo
->lo_disk
);
2133 mutex_unlock(&loop_ctl_mutex
);
2139 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
2142 struct loop_device
*lo
;
2145 ret
= mutex_lock_killable(&loop_ctl_mutex
);
2152 ret
= loop_lookup(&lo
, parm
);
2157 ret
= loop_add(&lo
, parm
);
2159 case LOOP_CTL_REMOVE
:
2160 ret
= loop_lookup(&lo
, parm
);
2163 if (lo
->lo_state
!= Lo_unbound
) {
2167 if (atomic_read(&lo
->lo_refcnt
) > 0) {
2171 lo
->lo_disk
->private_data
= NULL
;
2172 idr_remove(&loop_index_idr
, lo
->lo_number
);
2175 case LOOP_CTL_GET_FREE
:
2176 ret
= loop_lookup(&lo
, -1);
2179 ret
= loop_add(&lo
, -1);
2181 mutex_unlock(&loop_ctl_mutex
);
2186 static const struct file_operations loop_ctl_fops
= {
2187 .open
= nonseekable_open
,
2188 .unlocked_ioctl
= loop_control_ioctl
,
2189 .compat_ioctl
= loop_control_ioctl
,
2190 .owner
= THIS_MODULE
,
2191 .llseek
= noop_llseek
,
2194 static struct miscdevice loop_misc
= {
2195 .minor
= LOOP_CTRL_MINOR
,
2196 .name
= "loop-control",
2197 .fops
= &loop_ctl_fops
,
2200 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2201 MODULE_ALIAS("devname:loop-control");
2203 static int __init
loop_init(void)
2206 unsigned long range
;
2207 struct loop_device
*lo
;
2212 part_shift
= fls(max_part
);
2215 * Adjust max_part according to part_shift as it is exported
2216 * to user space so that user can decide correct minor number
2217 * if [s]he want to create more devices.
2219 * Note that -1 is required because partition 0 is reserved
2220 * for the whole disk.
2222 max_part
= (1UL << part_shift
) - 1;
2225 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2230 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2236 * If max_loop is specified, create that many devices upfront.
2237 * This also becomes a hard limit. If max_loop is not specified,
2238 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2239 * init time. Loop devices can be requested on-demand with the
2240 * /dev/loop-control interface, or be instantiated by accessing
2241 * a 'dead' device node.
2245 range
= max_loop
<< part_shift
;
2247 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2248 range
= 1UL << MINORBITS
;
2251 err
= misc_register(&loop_misc
);
2256 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2261 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2262 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2264 /* pre-create number of devices given by config or max_loop */
2265 mutex_lock(&loop_ctl_mutex
);
2266 for (i
= 0; i
< nr
; i
++)
2268 mutex_unlock(&loop_ctl_mutex
);
2270 printk(KERN_INFO
"loop: module loaded\n");
2274 misc_deregister(&loop_misc
);
2279 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2281 struct loop_device
*lo
= ptr
;
2287 static void __exit
loop_exit(void)
2289 unsigned long range
;
2291 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2293 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2294 idr_destroy(&loop_index_idr
);
2296 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2297 unregister_blkdev(LOOP_MAJOR
, "loop");
2299 misc_deregister(&loop_misc
);
2302 module_init(loop_init
);
2303 module_exit(loop_exit
);
2306 static int __init
max_loop_setup(char *str
)
2308 max_loop
= simple_strtol(str
, NULL
, 0);
2312 __setup("max_loop=", max_loop_setup
);