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>
83 #include <linux/uaccess.h>
85 static DEFINE_IDR(loop_index_idr
);
86 static DEFINE_MUTEX(loop_index_mutex
);
89 static int part_shift
;
91 static int transfer_xor(struct loop_device
*lo
, int cmd
,
92 struct page
*raw_page
, unsigned raw_off
,
93 struct page
*loop_page
, unsigned loop_off
,
94 int size
, sector_t real_block
)
96 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
97 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
109 key
= lo
->lo_encrypt_key
;
110 keysize
= lo
->lo_encrypt_key_size
;
111 for (i
= 0; i
< size
; i
++)
112 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
114 kunmap_atomic(loop_buf
);
115 kunmap_atomic(raw_buf
);
120 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
122 if (unlikely(info
->lo_encrypt_key_size
<= 0))
127 static struct loop_func_table none_funcs
= {
128 .number
= LO_CRYPT_NONE
,
131 static struct loop_func_table xor_funcs
= {
132 .number
= LO_CRYPT_XOR
,
133 .transfer
= transfer_xor
,
137 /* xfer_funcs[0] is special - its release function is never called */
138 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
143 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
147 /* Compute loopsize in bytes */
148 loopsize
= i_size_read(file
->f_mapping
->host
);
151 /* offset is beyond i_size, weird but possible */
155 if (sizelimit
> 0 && sizelimit
< loopsize
)
156 loopsize
= sizelimit
;
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
161 return loopsize
>> 9;
164 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
166 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
169 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
171 struct file
*file
= lo
->lo_backing_file
;
172 struct address_space
*mapping
= file
->f_mapping
;
173 struct inode
*inode
= mapping
->host
;
174 unsigned short sb_bsize
= 0;
175 unsigned dio_align
= 0;
178 if (inode
->i_sb
->s_bdev
) {
179 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
180 dio_align
= sb_bsize
- 1;
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
191 * of requests in sane applications should be PAGE_SIZE aligned
194 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
195 !(lo
->lo_offset
& dio_align
) &&
196 mapping
->a_ops
->direct_IO
&&
205 if (lo
->use_dio
== use_dio
)
208 /* flush dirty pages before changing direct IO */
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
216 blk_mq_freeze_queue(lo
->lo_queue
);
217 lo
->use_dio
= use_dio
;
219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
220 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
223 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
225 blk_mq_unfreeze_queue(lo
->lo_queue
);
229 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
231 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
232 sector_t x
= (sector_t
)size
;
233 struct block_device
*bdev
= lo
->lo_device
;
235 if (unlikely((loff_t
)x
!= size
))
237 if (lo
->lo_offset
!= offset
)
238 lo
->lo_offset
= offset
;
239 if (lo
->lo_sizelimit
!= sizelimit
)
240 lo
->lo_sizelimit
= sizelimit
;
241 set_capacity(lo
->lo_disk
, x
);
242 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
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
, ITER_BVEC
| 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
, ITER_BVEC
, &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
, ITER_BVEC
, &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_discard(struct loop_device
*lo
, struct request
*rq
, loff_t pos
)
422 * We use punch hole to reclaim the free space used by the
423 * image a.k.a. discard. However we do not support discard if
424 * encryption is enabled, because it may give an attacker
425 * useful information.
427 struct file
*file
= lo
->lo_backing_file
;
428 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
431 if ((!file
->f_op
->fallocate
) || lo
->lo_encrypt_key_size
) {
436 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
437 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
443 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
445 struct file
*file
= lo
->lo_backing_file
;
446 int ret
= vfs_fsync(file
, 0);
447 if (unlikely(ret
&& ret
!= -EINVAL
))
453 static void lo_complete_rq(struct request
*rq
)
455 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
456 blk_status_t ret
= BLK_STS_OK
;
458 if (!cmd
->use_aio
|| cmd
->ret
< 0 || cmd
->ret
== blk_rq_bytes(rq
) ||
459 req_op(rq
) != REQ_OP_READ
) {
466 * Short READ - if we got some data, advance our request and
467 * retry it. If we got no data, end the rest with EIO.
470 blk_update_request(rq
, BLK_STS_OK
, cmd
->ret
);
472 blk_mq_requeue_request(rq
, true);
475 struct bio
*bio
= rq
->bio
;
484 blk_mq_end_request(rq
, ret
);
488 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
490 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
492 if (!atomic_dec_and_test(&cmd
->ref
))
496 blk_mq_complete_request(rq
);
499 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
501 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
506 lo_rw_aio_do_completion(cmd
);
509 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
512 struct iov_iter iter
;
513 struct bio_vec
*bvec
;
514 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
515 struct bio
*bio
= rq
->bio
;
516 struct file
*file
= lo
->lo_backing_file
;
521 if (rq
->bio
!= rq
->biotail
) {
522 struct req_iterator iter
;
525 __rq_for_each_bio(bio
, rq
)
526 segments
+= bio_segments(bio
);
527 bvec
= kmalloc_array(segments
, sizeof(struct bio_vec
),
534 * The bios of the request may be started from the middle of
535 * the 'bvec' because of bio splitting, so we can't directly
536 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
537 * API will take care of all details for us.
539 rq_for_each_segment(tmp
, rq
, iter
) {
547 * Same here, this bio may be started from the middle of the
548 * 'bvec' because of bio splitting, so offset from the bvec
549 * must be passed to iov iterator
551 offset
= bio
->bi_iter
.bi_bvec_done
;
552 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
553 segments
= bio_segments(bio
);
555 atomic_set(&cmd
->ref
, 2);
557 iov_iter_bvec(&iter
, ITER_BVEC
| rw
, bvec
,
558 segments
, blk_rq_bytes(rq
));
559 iter
.iov_offset
= offset
;
561 cmd
->iocb
.ki_pos
= pos
;
562 cmd
->iocb
.ki_filp
= file
;
563 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
564 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
565 cmd
->iocb
.ki_ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
567 kthread_associate_blkcg(cmd
->css
);
570 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
572 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
574 lo_rw_aio_do_completion(cmd
);
575 kthread_associate_blkcg(NULL
);
577 if (ret
!= -EIOCBQUEUED
)
578 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
582 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
584 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
585 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
588 * lo_write_simple and lo_read_simple should have been covered
589 * by io submit style function like lo_rw_aio(), one blocker
590 * is that lo_read_simple() need to call flush_dcache_page after
591 * the page is written from kernel, and it isn't easy to handle
592 * this in io submit style function which submits all segments
593 * of the req at one time. And direct read IO doesn't need to
594 * run flush_dcache_page().
596 switch (req_op(rq
)) {
598 return lo_req_flush(lo
, rq
);
600 case REQ_OP_WRITE_ZEROES
:
601 return lo_discard(lo
, rq
, pos
);
604 return lo_write_transfer(lo
, rq
, pos
);
605 else if (cmd
->use_aio
)
606 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
608 return lo_write_simple(lo
, rq
, pos
);
611 return lo_read_transfer(lo
, rq
, pos
);
612 else if (cmd
->use_aio
)
613 return lo_rw_aio(lo
, cmd
, pos
, READ
);
615 return lo_read_simple(lo
, rq
, pos
);
623 static inline void loop_update_dio(struct loop_device
*lo
)
625 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
629 static void loop_reread_partitions(struct loop_device
*lo
,
630 struct block_device
*bdev
)
635 * bd_mutex has been held already in release path, so don't
636 * acquire it if this function is called in such case.
638 * If the reread partition isn't from release path, lo_refcnt
639 * must be at least one and it can only become zero when the
640 * current holder is released.
642 if (!atomic_read(&lo
->lo_refcnt
))
643 rc
= __blkdev_reread_part(bdev
);
645 rc
= blkdev_reread_part(bdev
);
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_unbound
) {
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
, *old_file
;
697 if (lo
->lo_state
!= Lo_bound
)
700 /* the loop device has to be read-only */
702 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
710 error
= loop_validate_file(file
, bdev
);
714 inode
= file
->f_mapping
->host
;
715 old_file
= lo
->lo_backing_file
;
719 /* size of the new backing store needs to be the same */
720 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
724 blk_mq_freeze_queue(lo
->lo_queue
);
725 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
726 lo
->lo_backing_file
= file
;
727 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
728 mapping_set_gfp_mask(file
->f_mapping
,
729 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
731 blk_mq_unfreeze_queue(lo
->lo_queue
);
734 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
735 loop_reread_partitions(lo
, bdev
);
744 /* loop sysfs attributes */
746 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
747 ssize_t (*callback
)(struct loop_device
*, char *))
749 struct gendisk
*disk
= dev_to_disk(dev
);
750 struct loop_device
*lo
= disk
->private_data
;
752 return callback(lo
, page
);
755 #define LOOP_ATTR_RO(_name) \
756 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
757 static ssize_t loop_attr_do_show_##_name(struct device *d, \
758 struct device_attribute *attr, char *b) \
760 return loop_attr_show(d, b, loop_attr_##_name##_show); \
762 static struct device_attribute loop_attr_##_name = \
763 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
765 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
770 spin_lock_irq(&lo
->lo_lock
);
771 if (lo
->lo_backing_file
)
772 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
773 spin_unlock_irq(&lo
->lo_lock
);
775 if (IS_ERR_OR_NULL(p
))
779 memmove(buf
, p
, ret
);
787 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
789 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
792 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
794 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
797 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
799 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
801 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
804 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
806 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
808 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
811 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
813 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
815 return sprintf(buf
, "%s\n", dio
? "1" : "0");
818 LOOP_ATTR_RO(backing_file
);
819 LOOP_ATTR_RO(offset
);
820 LOOP_ATTR_RO(sizelimit
);
821 LOOP_ATTR_RO(autoclear
);
822 LOOP_ATTR_RO(partscan
);
825 static struct attribute
*loop_attrs
[] = {
826 &loop_attr_backing_file
.attr
,
827 &loop_attr_offset
.attr
,
828 &loop_attr_sizelimit
.attr
,
829 &loop_attr_autoclear
.attr
,
830 &loop_attr_partscan
.attr
,
835 static struct attribute_group loop_attribute_group
= {
840 static void loop_sysfs_init(struct loop_device
*lo
)
842 lo
->sysfs_inited
= !sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
843 &loop_attribute_group
);
846 static void loop_sysfs_exit(struct loop_device
*lo
)
848 if (lo
->sysfs_inited
)
849 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
850 &loop_attribute_group
);
853 static void loop_config_discard(struct loop_device
*lo
)
855 struct file
*file
= lo
->lo_backing_file
;
856 struct inode
*inode
= file
->f_mapping
->host
;
857 struct request_queue
*q
= lo
->lo_queue
;
860 * We use punch hole to reclaim the free space used by the
861 * image a.k.a. discard. However we do not support discard if
862 * encryption is enabled, because it may give an attacker
863 * useful information.
865 if ((!file
->f_op
->fallocate
) ||
866 lo
->lo_encrypt_key_size
) {
867 q
->limits
.discard_granularity
= 0;
868 q
->limits
.discard_alignment
= 0;
869 blk_queue_max_discard_sectors(q
, 0);
870 blk_queue_max_write_zeroes_sectors(q
, 0);
871 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, q
);
875 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
876 q
->limits
.discard_alignment
= 0;
878 blk_queue_max_discard_sectors(q
, UINT_MAX
>> 9);
879 blk_queue_max_write_zeroes_sectors(q
, UINT_MAX
>> 9);
880 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, q
);
883 static void loop_unprepare_queue(struct loop_device
*lo
)
885 kthread_flush_worker(&lo
->worker
);
886 kthread_stop(lo
->worker_task
);
889 static int loop_kthread_worker_fn(void *worker_ptr
)
891 current
->flags
|= PF_LESS_THROTTLE
;
892 return kthread_worker_fn(worker_ptr
);
895 static int loop_prepare_queue(struct loop_device
*lo
)
897 kthread_init_worker(&lo
->worker
);
898 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
899 &lo
->worker
, "loop%d", lo
->lo_number
);
900 if (IS_ERR(lo
->worker_task
))
902 set_user_nice(lo
->worker_task
, MIN_NICE
);
906 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
907 struct block_device
*bdev
, unsigned int arg
)
911 struct address_space
*mapping
;
916 /* This is safe, since we have a reference from open(). */
917 __module_get(THIS_MODULE
);
925 if (lo
->lo_state
!= Lo_unbound
)
928 error
= loop_validate_file(file
, bdev
);
932 mapping
= file
->f_mapping
;
933 inode
= mapping
->host
;
935 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
936 !file
->f_op
->write_iter
)
937 lo_flags
|= LO_FLAGS_READ_ONLY
;
940 size
= get_loop_size(lo
, file
);
941 if ((loff_t
)(sector_t
)size
!= size
)
943 error
= loop_prepare_queue(lo
);
949 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
952 lo
->lo_device
= bdev
;
953 lo
->lo_flags
= lo_flags
;
954 lo
->lo_backing_file
= file
;
957 lo
->lo_sizelimit
= 0;
958 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
959 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
961 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
962 blk_queue_write_cache(lo
->lo_queue
, true, false);
965 set_capacity(lo
->lo_disk
, size
);
966 bd_set_size(bdev
, size
<< 9);
968 /* let user-space know about the new size */
969 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
971 set_blocksize(bdev
, S_ISBLK(inode
->i_mode
) ?
972 block_size(inode
->i_bdev
) : PAGE_SIZE
);
974 lo
->lo_state
= Lo_bound
;
976 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
977 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
978 loop_reread_partitions(lo
, bdev
);
980 /* Grab the block_device to prevent its destruction after we
981 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
989 /* This is safe: open() is still holding a reference. */
990 module_put(THIS_MODULE
);
995 loop_release_xfer(struct loop_device
*lo
)
998 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1002 err
= xfer
->release(lo
);
1003 lo
->transfer
= NULL
;
1004 lo
->lo_encryption
= NULL
;
1005 module_put(xfer
->owner
);
1011 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1012 const struct loop_info64
*i
)
1017 struct module
*owner
= xfer
->owner
;
1019 if (!try_module_get(owner
))
1022 err
= xfer
->init(lo
, i
);
1026 lo
->lo_encryption
= xfer
;
1031 static int loop_clr_fd(struct loop_device
*lo
)
1033 struct file
*filp
= lo
->lo_backing_file
;
1034 gfp_t gfp
= lo
->old_gfp_mask
;
1035 struct block_device
*bdev
= lo
->lo_device
;
1037 if (lo
->lo_state
!= Lo_bound
)
1041 * If we've explicitly asked to tear down the loop device,
1042 * and it has an elevated reference count, set it for auto-teardown when
1043 * the last reference goes away. This stops $!~#$@ udev from
1044 * preventing teardown because it decided that it needs to run blkid on
1045 * the loopback device whenever they appear. xfstests is notorious for
1046 * failing tests because blkid via udev races with a losetup
1047 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1048 * command to fail with EBUSY.
1050 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1051 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1052 mutex_unlock(&lo
->lo_ctl_mutex
);
1059 /* freeze request queue during the transition */
1060 blk_mq_freeze_queue(lo
->lo_queue
);
1062 spin_lock_irq(&lo
->lo_lock
);
1063 lo
->lo_state
= Lo_rundown
;
1064 lo
->lo_backing_file
= NULL
;
1065 spin_unlock_irq(&lo
->lo_lock
);
1067 loop_release_xfer(lo
);
1068 lo
->transfer
= NULL
;
1070 lo
->lo_device
= NULL
;
1071 lo
->lo_encryption
= NULL
;
1073 lo
->lo_sizelimit
= 0;
1074 lo
->lo_encrypt_key_size
= 0;
1075 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1076 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1077 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1078 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1079 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1080 blk_queue_io_min(lo
->lo_queue
, 512);
1083 invalidate_bdev(bdev
);
1084 bdev
->bd_inode
->i_mapping
->wb_err
= 0;
1086 set_capacity(lo
->lo_disk
, 0);
1087 loop_sysfs_exit(lo
);
1089 bd_set_size(bdev
, 0);
1090 /* let user-space know about this change */
1091 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1093 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1094 lo
->lo_state
= Lo_unbound
;
1095 /* This is safe: open() is still holding a reference. */
1096 module_put(THIS_MODULE
);
1097 blk_mq_unfreeze_queue(lo
->lo_queue
);
1099 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1100 loop_reread_partitions(lo
, bdev
);
1103 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1104 loop_unprepare_queue(lo
);
1105 mutex_unlock(&lo
->lo_ctl_mutex
);
1107 * Need not hold lo_ctl_mutex to fput backing file.
1108 * Calling fput holding lo_ctl_mutex triggers a circular
1109 * lock dependency possibility warning as fput can take
1110 * bd_mutex which is usually taken before lo_ctl_mutex.
1117 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1120 struct loop_func_table
*xfer
;
1121 kuid_t uid
= current_uid();
1123 if (lo
->lo_encrypt_key_size
&&
1124 !uid_eq(lo
->lo_key_owner
, uid
) &&
1125 !capable(CAP_SYS_ADMIN
))
1127 if (lo
->lo_state
!= Lo_bound
)
1129 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1132 /* I/O need to be drained during transfer transition */
1133 blk_mq_freeze_queue(lo
->lo_queue
);
1135 err
= loop_release_xfer(lo
);
1139 if (info
->lo_encrypt_type
) {
1140 unsigned int type
= info
->lo_encrypt_type
;
1142 if (type
>= MAX_LO_CRYPT
) {
1146 xfer
= xfer_funcs
[type
];
1154 err
= loop_init_xfer(lo
, xfer
, info
);
1158 if (lo
->lo_offset
!= info
->lo_offset
||
1159 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1160 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
)) {
1166 loop_config_discard(lo
);
1168 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1169 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1170 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1171 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1175 lo
->transfer
= xfer
->transfer
;
1176 lo
->ioctl
= xfer
->ioctl
;
1178 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1179 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1180 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1182 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1183 lo
->lo_init
[0] = info
->lo_init
[0];
1184 lo
->lo_init
[1] = info
->lo_init
[1];
1185 if (info
->lo_encrypt_key_size
) {
1186 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1187 info
->lo_encrypt_key_size
);
1188 lo
->lo_key_owner
= uid
;
1191 /* update dio if lo_offset or transfer is changed */
1192 __loop_update_dio(lo
, lo
->use_dio
);
1195 blk_mq_unfreeze_queue(lo
->lo_queue
);
1197 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1198 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1199 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1200 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1201 loop_reread_partitions(lo
, lo
->lo_device
);
1208 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1214 if (lo
->lo_state
!= Lo_bound
) {
1215 mutex_unlock(&lo
->lo_ctl_mutex
);
1219 memset(info
, 0, sizeof(*info
));
1220 info
->lo_number
= lo
->lo_number
;
1221 info
->lo_offset
= lo
->lo_offset
;
1222 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1223 info
->lo_flags
= lo
->lo_flags
;
1224 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1225 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1226 info
->lo_encrypt_type
=
1227 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1228 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1229 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1230 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1231 lo
->lo_encrypt_key_size
);
1234 /* Drop lo_ctl_mutex while we call into the filesystem. */
1235 file
= get_file(lo
->lo_backing_file
);
1236 mutex_unlock(&lo
->lo_ctl_mutex
);
1237 ret
= vfs_getattr(&file
->f_path
, &stat
, STATX_INO
,
1238 AT_STATX_SYNC_AS_STAT
);
1240 info
->lo_device
= huge_encode_dev(stat
.dev
);
1241 info
->lo_inode
= stat
.ino
;
1242 info
->lo_rdevice
= huge_encode_dev(stat
.rdev
);
1249 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1251 memset(info64
, 0, sizeof(*info64
));
1252 info64
->lo_number
= info
->lo_number
;
1253 info64
->lo_device
= info
->lo_device
;
1254 info64
->lo_inode
= info
->lo_inode
;
1255 info64
->lo_rdevice
= info
->lo_rdevice
;
1256 info64
->lo_offset
= info
->lo_offset
;
1257 info64
->lo_sizelimit
= 0;
1258 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1259 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1260 info64
->lo_flags
= info
->lo_flags
;
1261 info64
->lo_init
[0] = info
->lo_init
[0];
1262 info64
->lo_init
[1] = info
->lo_init
[1];
1263 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1264 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1266 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1267 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1271 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1273 memset(info
, 0, sizeof(*info
));
1274 info
->lo_number
= info64
->lo_number
;
1275 info
->lo_device
= info64
->lo_device
;
1276 info
->lo_inode
= info64
->lo_inode
;
1277 info
->lo_rdevice
= info64
->lo_rdevice
;
1278 info
->lo_offset
= info64
->lo_offset
;
1279 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1280 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1281 info
->lo_flags
= info64
->lo_flags
;
1282 info
->lo_init
[0] = info64
->lo_init
[0];
1283 info
->lo_init
[1] = info64
->lo_init
[1];
1284 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1285 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1287 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1288 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1290 /* error in case values were truncated */
1291 if (info
->lo_device
!= info64
->lo_device
||
1292 info
->lo_rdevice
!= info64
->lo_rdevice
||
1293 info
->lo_inode
!= info64
->lo_inode
||
1294 info
->lo_offset
!= info64
->lo_offset
)
1301 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1303 struct loop_info info
;
1304 struct loop_info64 info64
;
1306 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1308 loop_info64_from_old(&info
, &info64
);
1309 return loop_set_status(lo
, &info64
);
1313 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1315 struct loop_info64 info64
;
1317 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1319 return loop_set_status(lo
, &info64
);
1323 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1324 struct loop_info info
;
1325 struct loop_info64 info64
;
1329 mutex_unlock(&lo
->lo_ctl_mutex
);
1332 err
= loop_get_status(lo
, &info64
);
1334 err
= loop_info64_to_old(&info64
, &info
);
1335 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1342 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1343 struct loop_info64 info64
;
1347 mutex_unlock(&lo
->lo_ctl_mutex
);
1350 err
= loop_get_status(lo
, &info64
);
1351 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1357 static int loop_set_capacity(struct loop_device
*lo
)
1359 if (unlikely(lo
->lo_state
!= Lo_bound
))
1362 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1365 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1368 if (lo
->lo_state
!= Lo_bound
)
1371 __loop_update_dio(lo
, !!arg
);
1372 if (lo
->use_dio
== !!arg
)
1379 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1381 if (lo
->lo_state
!= Lo_bound
)
1384 if (arg
< 512 || arg
> PAGE_SIZE
|| !is_power_of_2(arg
))
1387 blk_mq_freeze_queue(lo
->lo_queue
);
1389 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1390 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1391 blk_queue_io_min(lo
->lo_queue
, arg
);
1392 loop_update_dio(lo
);
1394 blk_mq_unfreeze_queue(lo
->lo_queue
);
1399 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1400 unsigned int cmd
, unsigned long arg
)
1402 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1405 err
= mutex_lock_killable_nested(&lo
->lo_ctl_mutex
, 1);
1411 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1413 case LOOP_CHANGE_FD
:
1414 err
= loop_change_fd(lo
, bdev
, arg
);
1417 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1418 err
= loop_clr_fd(lo
);
1422 case LOOP_SET_STATUS
:
1424 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1425 err
= loop_set_status_old(lo
,
1426 (struct loop_info __user
*)arg
);
1428 case LOOP_GET_STATUS
:
1429 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1430 /* loop_get_status() unlocks lo_ctl_mutex */
1432 case LOOP_SET_STATUS64
:
1434 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1435 err
= loop_set_status64(lo
,
1436 (struct loop_info64 __user
*) arg
);
1438 case LOOP_GET_STATUS64
:
1439 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1440 /* loop_get_status() unlocks lo_ctl_mutex */
1442 case LOOP_SET_CAPACITY
:
1444 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1445 err
= loop_set_capacity(lo
);
1447 case LOOP_SET_DIRECT_IO
:
1449 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1450 err
= loop_set_dio(lo
, arg
);
1452 case LOOP_SET_BLOCK_SIZE
:
1454 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1455 err
= loop_set_block_size(lo
, arg
);
1458 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1460 mutex_unlock(&lo
->lo_ctl_mutex
);
1466 #ifdef CONFIG_COMPAT
1467 struct compat_loop_info
{
1468 compat_int_t lo_number
; /* ioctl r/o */
1469 compat_dev_t lo_device
; /* ioctl r/o */
1470 compat_ulong_t lo_inode
; /* ioctl r/o */
1471 compat_dev_t lo_rdevice
; /* ioctl r/o */
1472 compat_int_t lo_offset
;
1473 compat_int_t lo_encrypt_type
;
1474 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1475 compat_int_t lo_flags
; /* ioctl r/o */
1476 char lo_name
[LO_NAME_SIZE
];
1477 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1478 compat_ulong_t lo_init
[2];
1483 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1484 * - noinlined to reduce stack space usage in main part of driver
1487 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1488 struct loop_info64
*info64
)
1490 struct compat_loop_info info
;
1492 if (copy_from_user(&info
, arg
, sizeof(info
)))
1495 memset(info64
, 0, sizeof(*info64
));
1496 info64
->lo_number
= info
.lo_number
;
1497 info64
->lo_device
= info
.lo_device
;
1498 info64
->lo_inode
= info
.lo_inode
;
1499 info64
->lo_rdevice
= info
.lo_rdevice
;
1500 info64
->lo_offset
= info
.lo_offset
;
1501 info64
->lo_sizelimit
= 0;
1502 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1503 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1504 info64
->lo_flags
= info
.lo_flags
;
1505 info64
->lo_init
[0] = info
.lo_init
[0];
1506 info64
->lo_init
[1] = info
.lo_init
[1];
1507 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1508 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1510 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1511 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1516 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1517 * - noinlined to reduce stack space usage in main part of driver
1520 loop_info64_to_compat(const struct loop_info64
*info64
,
1521 struct compat_loop_info __user
*arg
)
1523 struct compat_loop_info info
;
1525 memset(&info
, 0, sizeof(info
));
1526 info
.lo_number
= info64
->lo_number
;
1527 info
.lo_device
= info64
->lo_device
;
1528 info
.lo_inode
= info64
->lo_inode
;
1529 info
.lo_rdevice
= info64
->lo_rdevice
;
1530 info
.lo_offset
= info64
->lo_offset
;
1531 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1532 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1533 info
.lo_flags
= info64
->lo_flags
;
1534 info
.lo_init
[0] = info64
->lo_init
[0];
1535 info
.lo_init
[1] = info64
->lo_init
[1];
1536 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1537 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1539 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1540 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1542 /* error in case values were truncated */
1543 if (info
.lo_device
!= info64
->lo_device
||
1544 info
.lo_rdevice
!= info64
->lo_rdevice
||
1545 info
.lo_inode
!= info64
->lo_inode
||
1546 info
.lo_offset
!= info64
->lo_offset
||
1547 info
.lo_init
[0] != info64
->lo_init
[0] ||
1548 info
.lo_init
[1] != info64
->lo_init
[1])
1551 if (copy_to_user(arg
, &info
, sizeof(info
)))
1557 loop_set_status_compat(struct loop_device
*lo
,
1558 const struct compat_loop_info __user
*arg
)
1560 struct loop_info64 info64
;
1563 ret
= loop_info64_from_compat(arg
, &info64
);
1566 return loop_set_status(lo
, &info64
);
1570 loop_get_status_compat(struct loop_device
*lo
,
1571 struct compat_loop_info __user
*arg
)
1573 struct loop_info64 info64
;
1577 mutex_unlock(&lo
->lo_ctl_mutex
);
1580 err
= loop_get_status(lo
, &info64
);
1582 err
= loop_info64_to_compat(&info64
, arg
);
1586 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1587 unsigned int cmd
, unsigned long arg
)
1589 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1593 case LOOP_SET_STATUS
:
1594 err
= mutex_lock_killable(&lo
->lo_ctl_mutex
);
1596 err
= loop_set_status_compat(lo
,
1597 (const struct compat_loop_info __user
*)arg
);
1598 mutex_unlock(&lo
->lo_ctl_mutex
);
1601 case LOOP_GET_STATUS
:
1602 err
= mutex_lock_killable(&lo
->lo_ctl_mutex
);
1604 err
= loop_get_status_compat(lo
,
1605 (struct compat_loop_info __user
*)arg
);
1606 /* loop_get_status() unlocks lo_ctl_mutex */
1609 case LOOP_SET_CAPACITY
:
1611 case LOOP_GET_STATUS64
:
1612 case LOOP_SET_STATUS64
:
1613 arg
= (unsigned long) compat_ptr(arg
);
1615 case LOOP_CHANGE_FD
:
1616 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1626 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1628 struct loop_device
*lo
;
1631 mutex_lock(&loop_index_mutex
);
1632 lo
= bdev
->bd_disk
->private_data
;
1638 atomic_inc(&lo
->lo_refcnt
);
1640 mutex_unlock(&loop_index_mutex
);
1644 static void __lo_release(struct loop_device
*lo
)
1648 if (atomic_dec_return(&lo
->lo_refcnt
))
1651 mutex_lock(&lo
->lo_ctl_mutex
);
1652 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1654 * In autoclear mode, stop the loop thread
1655 * and remove configuration after last close.
1657 err
= loop_clr_fd(lo
);
1660 } else if (lo
->lo_state
== Lo_bound
) {
1662 * Otherwise keep thread (if running) and config,
1663 * but flush possible ongoing bios in thread.
1665 blk_mq_freeze_queue(lo
->lo_queue
);
1666 blk_mq_unfreeze_queue(lo
->lo_queue
);
1669 mutex_unlock(&lo
->lo_ctl_mutex
);
1672 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1674 mutex_lock(&loop_index_mutex
);
1675 __lo_release(disk
->private_data
);
1676 mutex_unlock(&loop_index_mutex
);
1679 static const struct block_device_operations lo_fops
= {
1680 .owner
= THIS_MODULE
,
1682 .release
= lo_release
,
1684 #ifdef CONFIG_COMPAT
1685 .compat_ioctl
= lo_compat_ioctl
,
1690 * And now the modules code and kernel interface.
1692 static int max_loop
;
1693 module_param(max_loop
, int, 0444);
1694 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1695 module_param(max_part
, int, 0444);
1696 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1697 MODULE_LICENSE("GPL");
1698 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1700 int loop_register_transfer(struct loop_func_table
*funcs
)
1702 unsigned int n
= funcs
->number
;
1704 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1706 xfer_funcs
[n
] = funcs
;
1710 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1712 struct loop_device
*lo
= ptr
;
1713 struct loop_func_table
*xfer
= data
;
1715 mutex_lock(&lo
->lo_ctl_mutex
);
1716 if (lo
->lo_encryption
== xfer
)
1717 loop_release_xfer(lo
);
1718 mutex_unlock(&lo
->lo_ctl_mutex
);
1722 int loop_unregister_transfer(int number
)
1724 unsigned int n
= number
;
1725 struct loop_func_table
*xfer
;
1727 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1730 xfer_funcs
[n
] = NULL
;
1731 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1735 EXPORT_SYMBOL(loop_register_transfer
);
1736 EXPORT_SYMBOL(loop_unregister_transfer
);
1738 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1739 const struct blk_mq_queue_data
*bd
)
1741 struct request
*rq
= bd
->rq
;
1742 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1743 struct loop_device
*lo
= rq
->q
->queuedata
;
1745 blk_mq_start_request(rq
);
1747 if (lo
->lo_state
!= Lo_bound
)
1748 return BLK_STS_IOERR
;
1750 switch (req_op(rq
)) {
1752 case REQ_OP_DISCARD
:
1753 case REQ_OP_WRITE_ZEROES
:
1754 cmd
->use_aio
= false;
1757 cmd
->use_aio
= lo
->use_dio
;
1761 /* always use the first bio's css */
1762 #ifdef CONFIG_BLK_CGROUP
1763 if (cmd
->use_aio
&& rq
->bio
&& rq
->bio
->bi_css
) {
1764 cmd
->css
= rq
->bio
->bi_css
;
1769 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1774 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1776 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1777 const bool write
= op_is_write(req_op(rq
));
1778 struct loop_device
*lo
= rq
->q
->queuedata
;
1781 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1786 ret
= do_req_filebacked(lo
, rq
);
1788 /* complete non-aio request */
1789 if (!cmd
->use_aio
|| ret
) {
1790 cmd
->ret
= ret
? -EIO
: 0;
1791 blk_mq_complete_request(rq
);
1795 static void loop_queue_work(struct kthread_work
*work
)
1797 struct loop_cmd
*cmd
=
1798 container_of(work
, struct loop_cmd
, work
);
1800 loop_handle_cmd(cmd
);
1803 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1804 unsigned int hctx_idx
, unsigned int numa_node
)
1806 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1808 kthread_init_work(&cmd
->work
, loop_queue_work
);
1812 static const struct blk_mq_ops loop_mq_ops
= {
1813 .queue_rq
= loop_queue_rq
,
1814 .init_request
= loop_init_request
,
1815 .complete
= lo_complete_rq
,
1818 static int loop_add(struct loop_device
**l
, int i
)
1820 struct loop_device
*lo
;
1821 struct gendisk
*disk
;
1825 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1829 lo
->lo_state
= Lo_unbound
;
1831 /* allocate id, if @id >= 0, we're requesting that specific id */
1833 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1837 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1844 lo
->tag_set
.ops
= &loop_mq_ops
;
1845 lo
->tag_set
.nr_hw_queues
= 1;
1846 lo
->tag_set
.queue_depth
= 128;
1847 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1848 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1849 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1850 lo
->tag_set
.driver_data
= lo
;
1852 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1856 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1857 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1858 err
= PTR_ERR(lo
->lo_queue
);
1859 goto out_cleanup_tags
;
1861 lo
->lo_queue
->queuedata
= lo
;
1863 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
1866 * By default, we do buffer IO, so it doesn't make sense to enable
1867 * merge because the I/O submitted to backing file is handled page by
1868 * page. For directio mode, merge does help to dispatch bigger request
1869 * to underlayer disk. We will enable merge once directio is enabled.
1871 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
1874 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1876 goto out_free_queue
;
1879 * Disable partition scanning by default. The in-kernel partition
1880 * scanning can be requested individually per-device during its
1881 * setup. Userspace can always add and remove partitions from all
1882 * devices. The needed partition minors are allocated from the
1883 * extended minor space, the main loop device numbers will continue
1884 * to match the loop minors, regardless of the number of partitions
1887 * If max_part is given, partition scanning is globally enabled for
1888 * all loop devices. The minors for the main loop devices will be
1889 * multiples of max_part.
1891 * Note: Global-for-all-devices, set-only-at-init, read-only module
1892 * parameteters like 'max_loop' and 'max_part' make things needlessly
1893 * complicated, are too static, inflexible and may surprise
1894 * userspace tools. Parameters like this in general should be avoided.
1897 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1898 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1899 mutex_init(&lo
->lo_ctl_mutex
);
1900 atomic_set(&lo
->lo_refcnt
, 0);
1902 spin_lock_init(&lo
->lo_lock
);
1903 disk
->major
= LOOP_MAJOR
;
1904 disk
->first_minor
= i
<< part_shift
;
1905 disk
->fops
= &lo_fops
;
1906 disk
->private_data
= lo
;
1907 disk
->queue
= lo
->lo_queue
;
1908 sprintf(disk
->disk_name
, "loop%d", i
);
1911 return lo
->lo_number
;
1914 blk_cleanup_queue(lo
->lo_queue
);
1916 blk_mq_free_tag_set(&lo
->tag_set
);
1918 idr_remove(&loop_index_idr
, i
);
1925 static void loop_remove(struct loop_device
*lo
)
1927 del_gendisk(lo
->lo_disk
);
1928 blk_cleanup_queue(lo
->lo_queue
);
1929 blk_mq_free_tag_set(&lo
->tag_set
);
1930 put_disk(lo
->lo_disk
);
1934 static int find_free_cb(int id
, void *ptr
, void *data
)
1936 struct loop_device
*lo
= ptr
;
1937 struct loop_device
**l
= data
;
1939 if (lo
->lo_state
== Lo_unbound
) {
1946 static int loop_lookup(struct loop_device
**l
, int i
)
1948 struct loop_device
*lo
;
1954 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1957 ret
= lo
->lo_number
;
1962 /* lookup and return a specific i */
1963 lo
= idr_find(&loop_index_idr
, i
);
1966 ret
= lo
->lo_number
;
1972 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1974 struct loop_device
*lo
;
1975 struct kobject
*kobj
;
1978 mutex_lock(&loop_index_mutex
);
1979 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1981 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1985 kobj
= get_disk_and_module(lo
->lo_disk
);
1986 mutex_unlock(&loop_index_mutex
);
1992 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1995 struct loop_device
*lo
;
1998 mutex_lock(&loop_index_mutex
);
2001 ret
= loop_lookup(&lo
, parm
);
2006 ret
= loop_add(&lo
, parm
);
2008 case LOOP_CTL_REMOVE
:
2009 ret
= loop_lookup(&lo
, parm
);
2012 ret
= mutex_lock_killable(&lo
->lo_ctl_mutex
);
2015 if (lo
->lo_state
!= Lo_unbound
) {
2017 mutex_unlock(&lo
->lo_ctl_mutex
);
2020 if (atomic_read(&lo
->lo_refcnt
) > 0) {
2022 mutex_unlock(&lo
->lo_ctl_mutex
);
2025 lo
->lo_disk
->private_data
= NULL
;
2026 mutex_unlock(&lo
->lo_ctl_mutex
);
2027 idr_remove(&loop_index_idr
, lo
->lo_number
);
2030 case LOOP_CTL_GET_FREE
:
2031 ret
= loop_lookup(&lo
, -1);
2034 ret
= loop_add(&lo
, -1);
2036 mutex_unlock(&loop_index_mutex
);
2041 static const struct file_operations loop_ctl_fops
= {
2042 .open
= nonseekable_open
,
2043 .unlocked_ioctl
= loop_control_ioctl
,
2044 .compat_ioctl
= loop_control_ioctl
,
2045 .owner
= THIS_MODULE
,
2046 .llseek
= noop_llseek
,
2049 static struct miscdevice loop_misc
= {
2050 .minor
= LOOP_CTRL_MINOR
,
2051 .name
= "loop-control",
2052 .fops
= &loop_ctl_fops
,
2055 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2056 MODULE_ALIAS("devname:loop-control");
2058 static int __init
loop_init(void)
2061 unsigned long range
;
2062 struct loop_device
*lo
;
2067 part_shift
= fls(max_part
);
2070 * Adjust max_part according to part_shift as it is exported
2071 * to user space so that user can decide correct minor number
2072 * if [s]he want to create more devices.
2074 * Note that -1 is required because partition 0 is reserved
2075 * for the whole disk.
2077 max_part
= (1UL << part_shift
) - 1;
2080 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2085 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2091 * If max_loop is specified, create that many devices upfront.
2092 * This also becomes a hard limit. If max_loop is not specified,
2093 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2094 * init time. Loop devices can be requested on-demand with the
2095 * /dev/loop-control interface, or be instantiated by accessing
2096 * a 'dead' device node.
2100 range
= max_loop
<< part_shift
;
2102 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2103 range
= 1UL << MINORBITS
;
2106 err
= misc_register(&loop_misc
);
2111 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2116 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2117 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2119 /* pre-create number of devices given by config or max_loop */
2120 mutex_lock(&loop_index_mutex
);
2121 for (i
= 0; i
< nr
; i
++)
2123 mutex_unlock(&loop_index_mutex
);
2125 printk(KERN_INFO
"loop: module loaded\n");
2129 misc_deregister(&loop_misc
);
2134 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2136 struct loop_device
*lo
= ptr
;
2142 static void __exit
loop_exit(void)
2144 unsigned long range
;
2146 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2148 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2149 idr_destroy(&loop_index_idr
);
2151 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2152 unregister_blkdev(LOOP_MAJOR
, "loop");
2154 misc_deregister(&loop_misc
);
2157 module_init(loop_init
);
2158 module_exit(loop_exit
);
2161 static int __init
max_loop_setup(char *str
)
2163 max_loop
= simple_strtol(str
, NULL
, 0);
2167 __setup("max_loop=", max_loop_setup
);