2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
84 #include <linux/uaccess.h>
86 static DEFINE_IDR(loop_index_idr
);
87 static DEFINE_MUTEX(loop_ctl_mutex
);
90 static int part_shift
;
92 static int transfer_xor(struct loop_device
*lo
, int cmd
,
93 struct page
*raw_page
, unsigned raw_off
,
94 struct page
*loop_page
, unsigned loop_off
,
95 int size
, sector_t real_block
)
97 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
98 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
110 key
= lo
->lo_encrypt_key
;
111 keysize
= lo
->lo_encrypt_key_size
;
112 for (i
= 0; i
< size
; i
++)
113 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
115 kunmap_atomic(loop_buf
);
116 kunmap_atomic(raw_buf
);
121 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
123 if (unlikely(info
->lo_encrypt_key_size
<= 0))
128 static struct loop_func_table none_funcs
= {
129 .number
= LO_CRYPT_NONE
,
132 static struct loop_func_table xor_funcs
= {
133 .number
= LO_CRYPT_XOR
,
134 .transfer
= transfer_xor
,
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
144 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
148 /* Compute loopsize in bytes */
149 loopsize
= i_size_read(file
->f_mapping
->host
);
152 /* offset is beyond i_size, weird but possible */
156 if (sizelimit
> 0 && sizelimit
< loopsize
)
157 loopsize
= sizelimit
;
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
162 return loopsize
>> 9;
165 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
167 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
170 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
172 struct file
*file
= lo
->lo_backing_file
;
173 struct address_space
*mapping
= file
->f_mapping
;
174 struct inode
*inode
= mapping
->host
;
175 unsigned short sb_bsize
= 0;
176 unsigned dio_align
= 0;
179 if (inode
->i_sb
->s_bdev
) {
180 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
181 dio_align
= sb_bsize
- 1;
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
195 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
196 !(lo
->lo_offset
& dio_align
) &&
197 mapping
->a_ops
->direct_IO
&&
206 if (lo
->use_dio
== use_dio
)
209 /* flush dirty pages before changing direct IO */
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
217 if (lo
->lo_state
== Lo_bound
)
218 blk_mq_freeze_queue(lo
->lo_queue
);
219 lo
->use_dio
= use_dio
;
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
222 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
225 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
227 if (lo
->lo_state
== Lo_bound
)
228 blk_mq_unfreeze_queue(lo
->lo_queue
);
232 * loop_validate_block_size() - validates the passed in block size
233 * @bsize: size to validate
236 loop_validate_block_size(unsigned short bsize
)
238 if (bsize
< 512 || bsize
> PAGE_SIZE
|| !is_power_of_2(bsize
))
245 * loop_set_size() - sets device size and notifies userspace
246 * @lo: struct loop_device to set the size for
247 * @size: new size of the loop device
249 * Callers must validate that the size passed into this function fits into
250 * a sector_t, eg using loop_validate_size()
252 static void loop_set_size(struct loop_device
*lo
, loff_t size
)
254 if (!set_capacity_and_notify(lo
->lo_disk
, size
))
255 kobject_uevent(&disk_to_dev(lo
->lo_disk
)->kobj
, KOBJ_CHANGE
);
259 lo_do_transfer(struct loop_device
*lo
, int cmd
,
260 struct page
*rpage
, unsigned roffs
,
261 struct page
*lpage
, unsigned loffs
,
262 int size
, sector_t rblock
)
266 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
270 printk_ratelimited(KERN_ERR
271 "loop: Transfer error at byte offset %llu, length %i.\n",
272 (unsigned long long)rblock
<< 9, size
);
276 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
281 iov_iter_bvec(&i
, WRITE
, bvec
, 1, bvec
->bv_len
);
283 file_start_write(file
);
284 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
285 file_end_write(file
);
287 if (likely(bw
== bvec
->bv_len
))
290 printk_ratelimited(KERN_ERR
291 "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)*ppos
, bvec
->bv_len
);
298 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
302 struct req_iterator iter
;
305 rq_for_each_segment(bvec
, rq
, iter
) {
306 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
316 * This is the slow, transforming version that needs to double buffer the
317 * data as it cannot do the transformations in place without having direct
318 * access to the destination pages of the backing file.
320 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
323 struct bio_vec bvec
, b
;
324 struct req_iterator iter
;
328 page
= alloc_page(GFP_NOIO
);
332 rq_for_each_segment(bvec
, rq
, iter
) {
333 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
334 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
340 b
.bv_len
= bvec
.bv_len
;
341 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
350 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
354 struct req_iterator iter
;
358 rq_for_each_segment(bvec
, rq
, iter
) {
359 iov_iter_bvec(&i
, READ
, &bvec
, 1, bvec
.bv_len
);
360 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
364 flush_dcache_page(bvec
.bv_page
);
366 if (len
!= bvec
.bv_len
) {
369 __rq_for_each_bio(bio
, rq
)
379 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
382 struct bio_vec bvec
, b
;
383 struct req_iterator iter
;
389 page
= alloc_page(GFP_NOIO
);
393 rq_for_each_segment(bvec
, rq
, iter
) {
398 b
.bv_len
= bvec
.bv_len
;
400 iov_iter_bvec(&i
, READ
, &b
, 1, b
.bv_len
);
401 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
407 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
408 bvec
.bv_offset
, len
, offset
>> 9);
412 flush_dcache_page(bvec
.bv_page
);
414 if (len
!= bvec
.bv_len
) {
417 __rq_for_each_bio(bio
, rq
)
429 static int lo_fallocate(struct loop_device
*lo
, struct request
*rq
, loff_t pos
,
433 * We use fallocate to manipulate the space mappings used by the image
434 * a.k.a. discard/zerorange. However we do not support this if
435 * encryption is enabled, because it may give an attacker useful
438 struct file
*file
= lo
->lo_backing_file
;
439 struct request_queue
*q
= lo
->lo_queue
;
442 mode
|= FALLOC_FL_KEEP_SIZE
;
444 if (!blk_queue_discard(q
)) {
449 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
450 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
456 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
458 struct file
*file
= lo
->lo_backing_file
;
459 int ret
= vfs_fsync(file
, 0);
460 if (unlikely(ret
&& ret
!= -EINVAL
))
466 static void lo_complete_rq(struct request
*rq
)
468 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
469 blk_status_t ret
= BLK_STS_OK
;
471 if (!cmd
->use_aio
|| cmd
->ret
< 0 || cmd
->ret
== blk_rq_bytes(rq
) ||
472 req_op(rq
) != REQ_OP_READ
) {
474 ret
= errno_to_blk_status(cmd
->ret
);
479 * Short READ - if we got some data, advance our request and
480 * retry it. If we got no data, end the rest with EIO.
483 blk_update_request(rq
, BLK_STS_OK
, cmd
->ret
);
485 blk_mq_requeue_request(rq
, true);
488 struct bio
*bio
= rq
->bio
;
497 blk_mq_end_request(rq
, ret
);
501 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
503 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
505 if (!atomic_dec_and_test(&cmd
->ref
))
509 if (likely(!blk_should_fake_timeout(rq
->q
)))
510 blk_mq_complete_request(rq
);
513 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
515 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
520 lo_rw_aio_do_completion(cmd
);
523 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
526 struct iov_iter iter
;
527 struct req_iterator rq_iter
;
528 struct bio_vec
*bvec
;
529 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
530 struct bio
*bio
= rq
->bio
;
531 struct file
*file
= lo
->lo_backing_file
;
537 rq_for_each_bvec(tmp
, rq
, rq_iter
)
540 if (rq
->bio
!= rq
->biotail
) {
542 bvec
= kmalloc_array(nr_bvec
, sizeof(struct bio_vec
),
549 * The bios of the request may be started from the middle of
550 * the 'bvec' because of bio splitting, so we can't directly
551 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
552 * API will take care of all details for us.
554 rq_for_each_bvec(tmp
, rq
, rq_iter
) {
562 * Same here, this bio may be started from the middle of the
563 * 'bvec' because of bio splitting, so offset from the bvec
564 * must be passed to iov iterator
566 offset
= bio
->bi_iter
.bi_bvec_done
;
567 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
569 atomic_set(&cmd
->ref
, 2);
571 iov_iter_bvec(&iter
, rw
, bvec
, nr_bvec
, blk_rq_bytes(rq
));
572 iter
.iov_offset
= offset
;
574 cmd
->iocb
.ki_pos
= pos
;
575 cmd
->iocb
.ki_filp
= file
;
576 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
577 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
578 cmd
->iocb
.ki_ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
580 kthread_associate_blkcg(cmd
->css
);
583 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
585 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
587 lo_rw_aio_do_completion(cmd
);
588 kthread_associate_blkcg(NULL
);
590 if (ret
!= -EIOCBQUEUED
)
591 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
595 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
597 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
598 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
601 * lo_write_simple and lo_read_simple should have been covered
602 * by io submit style function like lo_rw_aio(), one blocker
603 * is that lo_read_simple() need to call flush_dcache_page after
604 * the page is written from kernel, and it isn't easy to handle
605 * this in io submit style function which submits all segments
606 * of the req at one time. And direct read IO doesn't need to
607 * run flush_dcache_page().
609 switch (req_op(rq
)) {
611 return lo_req_flush(lo
, rq
);
612 case REQ_OP_WRITE_ZEROES
:
614 * If the caller doesn't want deallocation, call zeroout to
615 * write zeroes the range. Otherwise, punch them out.
617 return lo_fallocate(lo
, rq
, pos
,
618 (rq
->cmd_flags
& REQ_NOUNMAP
) ?
619 FALLOC_FL_ZERO_RANGE
:
620 FALLOC_FL_PUNCH_HOLE
);
622 return lo_fallocate(lo
, rq
, pos
, FALLOC_FL_PUNCH_HOLE
);
625 return lo_write_transfer(lo
, rq
, pos
);
626 else if (cmd
->use_aio
)
627 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
629 return lo_write_simple(lo
, rq
, pos
);
632 return lo_read_transfer(lo
, rq
, pos
);
633 else if (cmd
->use_aio
)
634 return lo_rw_aio(lo
, cmd
, pos
, READ
);
636 return lo_read_simple(lo
, rq
, pos
);
643 static inline void loop_update_dio(struct loop_device
*lo
)
645 __loop_update_dio(lo
, (lo
->lo_backing_file
->f_flags
& O_DIRECT
) |
649 static struct file
*loop_real_file(struct file
*file
)
651 struct file
*f
= NULL
;
653 if (file
->f_path
.dentry
->d_sb
->s_op
->real_loop
)
654 f
= file
->f_path
.dentry
->d_sb
->s_op
->real_loop(file
);
658 static void loop_reread_partitions(struct loop_device
*lo
,
659 struct block_device
*bdev
)
663 mutex_lock(&bdev
->bd_mutex
);
664 rc
= bdev_disk_changed(bdev
, false);
665 mutex_unlock(&bdev
->bd_mutex
);
667 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
668 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
671 static inline int is_loop_device(struct file
*file
)
673 struct inode
*i
= file
->f_mapping
->host
;
675 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
678 static int loop_validate_file(struct file
*file
, struct block_device
*bdev
)
680 struct inode
*inode
= file
->f_mapping
->host
;
681 struct file
*f
= file
;
683 /* Avoid recursion */
684 while (is_loop_device(f
)) {
685 struct loop_device
*l
;
687 if (f
->f_mapping
->host
->i_rdev
== bdev
->bd_dev
)
690 l
= I_BDEV(f
->f_mapping
->host
)->bd_disk
->private_data
;
691 if (l
->lo_state
!= Lo_bound
) {
694 f
= l
->lo_backing_file
;
696 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
702 * loop_change_fd switched the backing store of a loopback device to
703 * a new file. This is useful for operating system installers to free up
704 * the original file and in High Availability environments to switch to
705 * an alternative location for the content in case of server meltdown.
706 * This can only work if the loop device is used read-only, and if the
707 * new backing store is the same size and type as the old backing store.
709 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
712 struct file
*file
= NULL
, *old_file
;
713 struct file
*f
, *virt_file
= NULL
, *old_virt_file
;
717 error
= mutex_lock_killable(&loop_ctl_mutex
);
721 if (lo
->lo_state
!= Lo_bound
)
724 /* the loop device has to be read-only */
726 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
733 f
= loop_real_file(file
);
740 error
= loop_validate_file(file
, bdev
);
744 old_file
= lo
->lo_backing_file
;
745 old_virt_file
= lo
->lo_backing_virt_file
;
749 /* size of the new backing store needs to be the same */
750 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
754 blk_mq_freeze_queue(lo
->lo_queue
);
755 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
756 lo
->lo_backing_file
= file
;
757 lo
->lo_backing_virt_file
= virt_file
;
758 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
759 mapping_set_gfp_mask(file
->f_mapping
,
760 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
762 blk_mq_unfreeze_queue(lo
->lo_queue
);
763 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
764 mutex_unlock(&loop_ctl_mutex
);
766 * We must drop file reference outside of loop_ctl_mutex as dropping
767 * the file ref can take bd_mutex which creates circular locking
774 loop_reread_partitions(lo
, bdev
);
778 mutex_unlock(&loop_ctl_mutex
);
788 * no get/put for file.
790 struct file
*loop_backing_file(struct super_block
*sb
)
793 struct loop_device
*l
;
796 if (MAJOR(sb
->s_dev
) == LOOP_MAJOR
) {
797 l
= sb
->s_bdev
->bd_disk
->private_data
;
798 ret
= l
->lo_backing_file
;
802 EXPORT_SYMBOL_GPL(loop_backing_file
);
804 /* loop sysfs attributes */
806 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
807 ssize_t (*callback
)(struct loop_device
*, char *))
809 struct gendisk
*disk
= dev_to_disk(dev
);
810 struct loop_device
*lo
= disk
->private_data
;
812 return callback(lo
, page
);
815 #define LOOP_ATTR_RO(_name) \
816 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
817 static ssize_t loop_attr_do_show_##_name(struct device *d, \
818 struct device_attribute *attr, char *b) \
820 return loop_attr_show(d, b, loop_attr_##_name##_show); \
822 static struct device_attribute loop_attr_##_name = \
823 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
825 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
830 spin_lock_irq(&lo
->lo_lock
);
831 if (lo
->lo_backing_file
)
832 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
833 spin_unlock_irq(&lo
->lo_lock
);
835 if (IS_ERR_OR_NULL(p
))
839 memmove(buf
, p
, ret
);
847 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
849 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
852 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
854 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
857 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
859 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
861 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
864 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
866 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
868 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
871 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
873 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
875 return sprintf(buf
, "%s\n", dio
? "1" : "0");
878 LOOP_ATTR_RO(backing_file
);
879 LOOP_ATTR_RO(offset
);
880 LOOP_ATTR_RO(sizelimit
);
881 LOOP_ATTR_RO(autoclear
);
882 LOOP_ATTR_RO(partscan
);
885 static struct attribute
*loop_attrs
[] = {
886 &loop_attr_backing_file
.attr
,
887 &loop_attr_offset
.attr
,
888 &loop_attr_sizelimit
.attr
,
889 &loop_attr_autoclear
.attr
,
890 &loop_attr_partscan
.attr
,
895 static struct attribute_group loop_attribute_group
= {
900 static void loop_sysfs_init(struct loop_device
*lo
)
902 lo
->sysfs_inited
= !sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
903 &loop_attribute_group
);
906 static void loop_sysfs_exit(struct loop_device
*lo
)
908 if (lo
->sysfs_inited
)
909 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
910 &loop_attribute_group
);
913 static void loop_config_discard(struct loop_device
*lo
)
915 struct file
*file
= lo
->lo_backing_file
;
916 struct inode
*inode
= file
->f_mapping
->host
;
917 struct request_queue
*q
= lo
->lo_queue
;
918 u32 granularity
, max_discard_sectors
;
921 * If the backing device is a block device, mirror its zeroing
922 * capability. Set the discard sectors to the block device's zeroing
923 * capabilities because loop discards result in blkdev_issue_zeroout(),
924 * not blkdev_issue_discard(). This maintains consistent behavior with
925 * file-backed loop devices: discarded regions read back as zero.
927 if (S_ISBLK(inode
->i_mode
) && !lo
->lo_encrypt_key_size
) {
928 struct request_queue
*backingq
= bdev_get_queue(I_BDEV(inode
));
930 max_discard_sectors
= backingq
->limits
.max_write_zeroes_sectors
;
931 granularity
= backingq
->limits
.discard_granularity
?:
932 queue_physical_block_size(backingq
);
935 * We use punch hole to reclaim the free space used by the
936 * image a.k.a. discard. However we do not support discard if
937 * encryption is enabled, because it may give an attacker
938 * useful information.
940 } else if (!file
->f_op
->fallocate
|| lo
->lo_encrypt_key_size
) {
941 max_discard_sectors
= 0;
945 max_discard_sectors
= UINT_MAX
>> 9;
946 granularity
= inode
->i_sb
->s_blocksize
;
949 if (max_discard_sectors
) {
950 q
->limits
.discard_granularity
= granularity
;
951 blk_queue_max_discard_sectors(q
, max_discard_sectors
);
952 blk_queue_max_write_zeroes_sectors(q
, max_discard_sectors
);
953 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, q
);
955 q
->limits
.discard_granularity
= 0;
956 blk_queue_max_discard_sectors(q
, 0);
957 blk_queue_max_write_zeroes_sectors(q
, 0);
958 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, q
);
960 q
->limits
.discard_alignment
= 0;
963 static void loop_unprepare_queue(struct loop_device
*lo
)
965 kthread_flush_worker(&lo
->worker
);
966 kthread_stop(lo
->worker_task
);
969 static int loop_kthread_worker_fn(void *worker_ptr
)
971 current
->flags
|= PF_LOCAL_THROTTLE
| PF_MEMALLOC_NOIO
;
972 return kthread_worker_fn(worker_ptr
);
975 static int loop_prepare_queue(struct loop_device
*lo
)
977 kthread_init_worker(&lo
->worker
);
978 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
979 &lo
->worker
, "loop%d", lo
->lo_number
);
980 if (IS_ERR(lo
->worker_task
))
982 set_user_nice(lo
->worker_task
, MIN_NICE
);
986 static void loop_update_rotational(struct loop_device
*lo
)
988 struct file
*file
= lo
->lo_backing_file
;
989 struct inode
*file_inode
= file
->f_mapping
->host
;
990 struct block_device
*file_bdev
= file_inode
->i_sb
->s_bdev
;
991 struct request_queue
*q
= lo
->lo_queue
;
994 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
996 nonrot
= blk_queue_nonrot(bdev_get_queue(file_bdev
));
999 blk_queue_flag_set(QUEUE_FLAG_NONROT
, q
);
1001 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, q
);
1005 loop_release_xfer(struct loop_device
*lo
)
1008 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1012 err
= xfer
->release(lo
);
1013 lo
->transfer
= NULL
;
1014 lo
->lo_encryption
= NULL
;
1015 module_put(xfer
->owner
);
1021 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1022 const struct loop_info64
*i
)
1027 struct module
*owner
= xfer
->owner
;
1029 if (!try_module_get(owner
))
1032 err
= xfer
->init(lo
, i
);
1036 lo
->lo_encryption
= xfer
;
1042 * loop_set_status_from_info - configure device from loop_info
1043 * @lo: struct loop_device to configure
1044 * @info: struct loop_info64 to configure the device with
1046 * Configures the loop device parameters according to the passed
1047 * in loop_info64 configuration.
1050 loop_set_status_from_info(struct loop_device
*lo
,
1051 const struct loop_info64
*info
)
1054 struct loop_func_table
*xfer
;
1055 kuid_t uid
= current_uid();
1057 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1060 err
= loop_release_xfer(lo
);
1064 if (info
->lo_encrypt_type
) {
1065 unsigned int type
= info
->lo_encrypt_type
;
1067 if (type
>= MAX_LO_CRYPT
)
1069 xfer
= xfer_funcs
[type
];
1075 err
= loop_init_xfer(lo
, xfer
, info
);
1079 lo
->lo_offset
= info
->lo_offset
;
1080 lo
->lo_sizelimit
= info
->lo_sizelimit
;
1081 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1082 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1083 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1084 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1088 lo
->transfer
= xfer
->transfer
;
1089 lo
->ioctl
= xfer
->ioctl
;
1091 lo
->lo_flags
= info
->lo_flags
;
1093 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1094 lo
->lo_init
[0] = info
->lo_init
[0];
1095 lo
->lo_init
[1] = info
->lo_init
[1];
1096 if (info
->lo_encrypt_key_size
) {
1097 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1098 info
->lo_encrypt_key_size
);
1099 lo
->lo_key_owner
= uid
;
1105 static int loop_configure(struct loop_device
*lo
, fmode_t mode
,
1106 struct block_device
*bdev
,
1107 const struct loop_config
*config
)
1109 struct file
*file
, *f
, *virt_file
= NULL
;
1110 struct inode
*inode
;
1111 struct address_space
*mapping
;
1115 unsigned short bsize
;
1117 /* This is safe, since we have a reference from open(). */
1118 __module_get(THIS_MODULE
);
1121 file
= fget(config
->fd
);
1124 f
= loop_real_file(file
);
1132 * If we don't hold exclusive handle for the device, upgrade to it
1133 * here to avoid changing device under exclusive owner.
1135 if (!(mode
& FMODE_EXCL
)) {
1136 error
= bd_prepare_to_claim(bdev
, loop_configure
);
1141 error
= mutex_lock_killable(&loop_ctl_mutex
);
1146 if (lo
->lo_state
!= Lo_unbound
)
1149 error
= loop_validate_file(file
, bdev
);
1153 mapping
= file
->f_mapping
;
1154 inode
= mapping
->host
;
1156 if ((config
->info
.lo_flags
& ~LOOP_CONFIGURE_SETTABLE_FLAGS
) != 0) {
1161 if (config
->block_size
) {
1162 error
= loop_validate_block_size(config
->block_size
);
1167 error
= loop_set_status_from_info(lo
, &config
->info
);
1171 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
1172 !file
->f_op
->write_iter
)
1173 lo
->lo_flags
|= LO_FLAGS_READ_ONLY
;
1175 error
= loop_prepare_queue(lo
);
1179 set_disk_ro(lo
->lo_disk
, (lo
->lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
1181 lo
->use_dio
= lo
->lo_flags
& LO_FLAGS_DIRECT_IO
;
1182 lo
->lo_device
= bdev
;
1183 lo
->lo_backing_file
= file
;
1184 lo
->lo_backing_virt_file
= virt_file
;
1185 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
1186 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
1188 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
1189 blk_queue_write_cache(lo
->lo_queue
, true, false);
1191 if (config
->block_size
)
1192 bsize
= config
->block_size
;
1193 else if ((lo
->lo_backing_file
->f_flags
& O_DIRECT
) && inode
->i_sb
->s_bdev
)
1194 /* In case of direct I/O, match underlying block size */
1195 bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
1199 blk_queue_logical_block_size(lo
->lo_queue
, bsize
);
1200 blk_queue_physical_block_size(lo
->lo_queue
, bsize
);
1201 blk_queue_io_min(lo
->lo_queue
, bsize
);
1203 loop_config_discard(lo
);
1204 loop_update_rotational(lo
);
1205 loop_update_dio(lo
);
1206 loop_sysfs_init(lo
);
1208 size
= get_loop_size(lo
, file
);
1209 loop_set_size(lo
, size
);
1211 lo
->lo_state
= Lo_bound
;
1213 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1214 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
1216 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1218 /* Grab the block_device to prevent its destruction after we
1219 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1222 mutex_unlock(&loop_ctl_mutex
);
1224 loop_reread_partitions(lo
, bdev
);
1225 if (!(mode
& FMODE_EXCL
))
1226 bd_abort_claiming(bdev
, loop_configure
);
1230 mutex_unlock(&loop_ctl_mutex
);
1232 if (!(mode
& FMODE_EXCL
))
1233 bd_abort_claiming(bdev
, loop_configure
);
1239 /* This is safe: open() is still holding a reference. */
1240 module_put(THIS_MODULE
);
1244 static int __loop_clr_fd(struct loop_device
*lo
, bool release
)
1246 struct file
*filp
= NULL
;
1247 struct file
*virt_filp
= lo
->lo_backing_virt_file
;
1248 gfp_t gfp
= lo
->old_gfp_mask
;
1249 struct block_device
*bdev
= lo
->lo_device
;
1251 bool partscan
= false;
1254 mutex_lock(&loop_ctl_mutex
);
1255 if (WARN_ON_ONCE(lo
->lo_state
!= Lo_rundown
)) {
1260 filp
= lo
->lo_backing_file
;
1266 if (test_bit(QUEUE_FLAG_WC
, &lo
->lo_queue
->queue_flags
))
1267 blk_queue_write_cache(lo
->lo_queue
, false, false);
1269 /* freeze request queue during the transition */
1270 blk_mq_freeze_queue(lo
->lo_queue
);
1272 spin_lock_irq(&lo
->lo_lock
);
1273 lo
->lo_backing_file
= NULL
;
1274 lo
->lo_backing_virt_file
= NULL
;
1275 spin_unlock_irq(&lo
->lo_lock
);
1277 loop_release_xfer(lo
);
1278 lo
->transfer
= NULL
;
1280 lo
->lo_device
= NULL
;
1281 lo
->lo_encryption
= NULL
;
1283 lo
->lo_sizelimit
= 0;
1284 lo
->lo_encrypt_key_size
= 0;
1285 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1286 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1287 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1288 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1289 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1290 blk_queue_io_min(lo
->lo_queue
, 512);
1293 invalidate_bdev(bdev
);
1294 bdev
->bd_inode
->i_mapping
->wb_err
= 0;
1296 set_capacity(lo
->lo_disk
, 0);
1297 loop_sysfs_exit(lo
);
1299 /* let user-space know about this change */
1300 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1302 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1303 /* This is safe: open() is still holding a reference. */
1304 module_put(THIS_MODULE
);
1305 blk_mq_unfreeze_queue(lo
->lo_queue
);
1307 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
;
1308 lo_number
= lo
->lo_number
;
1309 loop_unprepare_queue(lo
);
1311 mutex_unlock(&loop_ctl_mutex
);
1314 * bd_mutex has been held already in release path, so don't
1315 * acquire it if this function is called in such case.
1317 * If the reread partition isn't from release path, lo_refcnt
1318 * must be at least one and it can only become zero when the
1319 * current holder is released.
1322 mutex_lock(&bdev
->bd_mutex
);
1323 err
= bdev_disk_changed(bdev
, false);
1325 mutex_unlock(&bdev
->bd_mutex
);
1327 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1328 __func__
, lo_number
, err
);
1329 /* Device is gone, no point in returning error */
1334 * lo->lo_state is set to Lo_unbound here after above partscan has
1337 * There cannot be anybody else entering __loop_clr_fd() as
1338 * lo->lo_backing_file is already cleared and Lo_rundown state
1339 * protects us from all the other places trying to change the 'lo'
1342 mutex_lock(&loop_ctl_mutex
);
1345 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1346 lo
->lo_state
= Lo_unbound
;
1347 mutex_unlock(&loop_ctl_mutex
);
1350 * Need not hold loop_ctl_mutex to fput backing file.
1351 * Calling fput holding loop_ctl_mutex triggers a circular
1352 * lock dependency possibility warning as fput can take
1353 * bd_mutex which is usually taken before loop_ctl_mutex.
1362 static int loop_clr_fd(struct loop_device
*lo
)
1366 err
= mutex_lock_killable(&loop_ctl_mutex
);
1369 if (lo
->lo_state
!= Lo_bound
) {
1370 mutex_unlock(&loop_ctl_mutex
);
1374 * If we've explicitly asked to tear down the loop device,
1375 * and it has an elevated reference count, set it for auto-teardown when
1376 * the last reference goes away. This stops $!~#$@ udev from
1377 * preventing teardown because it decided that it needs to run blkid on
1378 * the loopback device whenever they appear. xfstests is notorious for
1379 * failing tests because blkid via udev races with a losetup
1380 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1381 * command to fail with EBUSY.
1383 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1384 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1385 mutex_unlock(&loop_ctl_mutex
);
1388 lo
->lo_state
= Lo_rundown
;
1389 mutex_unlock(&loop_ctl_mutex
);
1391 return __loop_clr_fd(lo
, false);
1395 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1398 struct block_device
*bdev
;
1399 kuid_t uid
= current_uid();
1401 bool partscan
= false;
1402 bool size_changed
= false;
1404 err
= mutex_lock_killable(&loop_ctl_mutex
);
1407 if (lo
->lo_encrypt_key_size
&&
1408 !uid_eq(lo
->lo_key_owner
, uid
) &&
1409 !capable(CAP_SYS_ADMIN
)) {
1413 if (lo
->lo_state
!= Lo_bound
) {
1418 if (lo
->lo_offset
!= info
->lo_offset
||
1419 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1420 size_changed
= true;
1421 sync_blockdev(lo
->lo_device
);
1422 invalidate_bdev(lo
->lo_device
);
1425 /* I/O need to be drained during transfer transition */
1426 blk_mq_freeze_queue(lo
->lo_queue
);
1428 if (size_changed
&& lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1429 /* If any pages were dirtied after invalidate_bdev(), try again */
1431 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1432 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1433 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1437 prev_lo_flags
= lo
->lo_flags
;
1439 err
= loop_set_status_from_info(lo
, info
);
1443 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1444 lo
->lo_flags
&= LOOP_SET_STATUS_SETTABLE_FLAGS
;
1445 /* For those flags, use the previous values instead */
1446 lo
->lo_flags
|= prev_lo_flags
& ~LOOP_SET_STATUS_SETTABLE_FLAGS
;
1447 /* For flags that can't be cleared, use previous values too */
1448 lo
->lo_flags
|= prev_lo_flags
& ~LOOP_SET_STATUS_CLEARABLE_FLAGS
;
1451 loff_t new_size
= get_size(lo
->lo_offset
, lo
->lo_sizelimit
,
1452 lo
->lo_backing_file
);
1453 loop_set_size(lo
, new_size
);
1456 loop_config_discard(lo
);
1458 /* update dio if lo_offset or transfer is changed */
1459 __loop_update_dio(lo
, lo
->use_dio
);
1462 blk_mq_unfreeze_queue(lo
->lo_queue
);
1464 if (!err
&& (lo
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1465 !(prev_lo_flags
& LO_FLAGS_PARTSCAN
)) {
1466 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1467 bdev
= lo
->lo_device
;
1471 mutex_unlock(&loop_ctl_mutex
);
1473 loop_reread_partitions(lo
, bdev
);
1479 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1485 ret
= mutex_lock_killable(&loop_ctl_mutex
);
1488 if (lo
->lo_state
!= Lo_bound
) {
1489 mutex_unlock(&loop_ctl_mutex
);
1493 memset(info
, 0, sizeof(*info
));
1494 info
->lo_number
= lo
->lo_number
;
1495 info
->lo_offset
= lo
->lo_offset
;
1496 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1497 info
->lo_flags
= lo
->lo_flags
;
1498 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1499 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1500 info
->lo_encrypt_type
=
1501 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1502 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1503 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1504 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1505 lo
->lo_encrypt_key_size
);
1508 /* Drop loop_ctl_mutex while we call into the filesystem. */
1509 path
= lo
->lo_backing_file
->f_path
;
1511 mutex_unlock(&loop_ctl_mutex
);
1512 ret
= vfs_getattr(&path
, &stat
, STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1514 info
->lo_device
= huge_encode_dev(stat
.dev
);
1515 info
->lo_inode
= stat
.ino
;
1516 info
->lo_rdevice
= huge_encode_dev(stat
.rdev
);
1523 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1525 memset(info64
, 0, sizeof(*info64
));
1526 info64
->lo_number
= info
->lo_number
;
1527 info64
->lo_device
= info
->lo_device
;
1528 info64
->lo_inode
= info
->lo_inode
;
1529 info64
->lo_rdevice
= info
->lo_rdevice
;
1530 info64
->lo_offset
= info
->lo_offset
;
1531 info64
->lo_sizelimit
= 0;
1532 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1533 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1534 info64
->lo_flags
= info
->lo_flags
;
1535 info64
->lo_init
[0] = info
->lo_init
[0];
1536 info64
->lo_init
[1] = info
->lo_init
[1];
1537 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1538 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1540 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1541 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1545 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1547 memset(info
, 0, sizeof(*info
));
1548 info
->lo_number
= info64
->lo_number
;
1549 info
->lo_device
= info64
->lo_device
;
1550 info
->lo_inode
= info64
->lo_inode
;
1551 info
->lo_rdevice
= info64
->lo_rdevice
;
1552 info
->lo_offset
= info64
->lo_offset
;
1553 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1554 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1555 info
->lo_flags
= info64
->lo_flags
;
1556 info
->lo_init
[0] = info64
->lo_init
[0];
1557 info
->lo_init
[1] = info64
->lo_init
[1];
1558 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1559 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1561 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1562 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1564 /* error in case values were truncated */
1565 if (info
->lo_device
!= info64
->lo_device
||
1566 info
->lo_rdevice
!= info64
->lo_rdevice
||
1567 info
->lo_inode
!= info64
->lo_inode
||
1568 info
->lo_offset
!= info64
->lo_offset
)
1575 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1577 struct loop_info info
;
1578 struct loop_info64 info64
;
1580 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1582 loop_info64_from_old(&info
, &info64
);
1583 return loop_set_status(lo
, &info64
);
1587 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1589 struct loop_info64 info64
;
1591 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1593 return loop_set_status(lo
, &info64
);
1597 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1598 struct loop_info info
;
1599 struct loop_info64 info64
;
1604 err
= loop_get_status(lo
, &info64
);
1606 err
= loop_info64_to_old(&info64
, &info
);
1607 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1614 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1615 struct loop_info64 info64
;
1620 err
= loop_get_status(lo
, &info64
);
1621 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1627 static int loop_set_capacity(struct loop_device
*lo
)
1631 if (unlikely(lo
->lo_state
!= Lo_bound
))
1634 size
= get_loop_size(lo
, lo
->lo_backing_file
);
1635 loop_set_size(lo
, size
);
1640 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1643 if (lo
->lo_state
!= Lo_bound
)
1646 __loop_update_dio(lo
, !!arg
);
1647 if (lo
->use_dio
== !!arg
)
1654 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1658 if (lo
->lo_state
!= Lo_bound
)
1661 err
= loop_validate_block_size(arg
);
1665 if (lo
->lo_queue
->limits
.logical_block_size
== arg
)
1668 sync_blockdev(lo
->lo_device
);
1669 invalidate_bdev(lo
->lo_device
);
1671 blk_mq_freeze_queue(lo
->lo_queue
);
1673 /* invalidate_bdev should have truncated all the pages */
1674 if (lo
->lo_device
->bd_inode
->i_mapping
->nrpages
) {
1676 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1677 __func__
, lo
->lo_number
, lo
->lo_file_name
,
1678 lo
->lo_device
->bd_inode
->i_mapping
->nrpages
);
1682 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1683 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1684 blk_queue_io_min(lo
->lo_queue
, arg
);
1685 loop_update_dio(lo
);
1687 blk_mq_unfreeze_queue(lo
->lo_queue
);
1692 static int lo_simple_ioctl(struct loop_device
*lo
, unsigned int cmd
,
1697 err
= mutex_lock_killable(&loop_ctl_mutex
);
1701 case LOOP_SET_CAPACITY
:
1702 err
= loop_set_capacity(lo
);
1704 case LOOP_SET_DIRECT_IO
:
1705 err
= loop_set_dio(lo
, arg
);
1707 case LOOP_SET_BLOCK_SIZE
:
1708 err
= loop_set_block_size(lo
, arg
);
1711 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1713 mutex_unlock(&loop_ctl_mutex
);
1717 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1718 unsigned int cmd
, unsigned long arg
)
1720 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1721 void __user
*argp
= (void __user
*) arg
;
1727 * Legacy case - pass in a zeroed out struct loop_config with
1728 * only the file descriptor set , which corresponds with the
1729 * default parameters we'd have used otherwise.
1731 struct loop_config config
;
1733 memset(&config
, 0, sizeof(config
));
1736 return loop_configure(lo
, mode
, bdev
, &config
);
1738 case LOOP_CONFIGURE
: {
1739 struct loop_config config
;
1741 if (copy_from_user(&config
, argp
, sizeof(config
)))
1744 return loop_configure(lo
, mode
, bdev
, &config
);
1746 case LOOP_CHANGE_FD
:
1747 return loop_change_fd(lo
, bdev
, arg
);
1749 return loop_clr_fd(lo
);
1750 case LOOP_SET_STATUS
:
1752 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1753 err
= loop_set_status_old(lo
, argp
);
1756 case LOOP_GET_STATUS
:
1757 return loop_get_status_old(lo
, argp
);
1758 case LOOP_SET_STATUS64
:
1760 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
)) {
1761 err
= loop_set_status64(lo
, argp
);
1764 case LOOP_GET_STATUS64
:
1765 return loop_get_status64(lo
, argp
);
1766 case LOOP_SET_CAPACITY
:
1767 case LOOP_SET_DIRECT_IO
:
1768 case LOOP_SET_BLOCK_SIZE
:
1769 if (!(mode
& FMODE_WRITE
) && !capable(CAP_SYS_ADMIN
))
1773 err
= lo_simple_ioctl(lo
, cmd
, arg
);
1780 #ifdef CONFIG_COMPAT
1781 struct compat_loop_info
{
1782 compat_int_t lo_number
; /* ioctl r/o */
1783 compat_dev_t lo_device
; /* ioctl r/o */
1784 compat_ulong_t lo_inode
; /* ioctl r/o */
1785 compat_dev_t lo_rdevice
; /* ioctl r/o */
1786 compat_int_t lo_offset
;
1787 compat_int_t lo_encrypt_type
;
1788 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1789 compat_int_t lo_flags
; /* ioctl r/o */
1790 char lo_name
[LO_NAME_SIZE
];
1791 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1792 compat_ulong_t lo_init
[2];
1797 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1798 * - noinlined to reduce stack space usage in main part of driver
1801 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1802 struct loop_info64
*info64
)
1804 struct compat_loop_info info
;
1806 if (copy_from_user(&info
, arg
, sizeof(info
)))
1809 memset(info64
, 0, sizeof(*info64
));
1810 info64
->lo_number
= info
.lo_number
;
1811 info64
->lo_device
= info
.lo_device
;
1812 info64
->lo_inode
= info
.lo_inode
;
1813 info64
->lo_rdevice
= info
.lo_rdevice
;
1814 info64
->lo_offset
= info
.lo_offset
;
1815 info64
->lo_sizelimit
= 0;
1816 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1817 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1818 info64
->lo_flags
= info
.lo_flags
;
1819 info64
->lo_init
[0] = info
.lo_init
[0];
1820 info64
->lo_init
[1] = info
.lo_init
[1];
1821 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1822 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1824 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1825 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1830 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1831 * - noinlined to reduce stack space usage in main part of driver
1834 loop_info64_to_compat(const struct loop_info64
*info64
,
1835 struct compat_loop_info __user
*arg
)
1837 struct compat_loop_info info
;
1839 memset(&info
, 0, sizeof(info
));
1840 info
.lo_number
= info64
->lo_number
;
1841 info
.lo_device
= info64
->lo_device
;
1842 info
.lo_inode
= info64
->lo_inode
;
1843 info
.lo_rdevice
= info64
->lo_rdevice
;
1844 info
.lo_offset
= info64
->lo_offset
;
1845 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1846 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1847 info
.lo_flags
= info64
->lo_flags
;
1848 info
.lo_init
[0] = info64
->lo_init
[0];
1849 info
.lo_init
[1] = info64
->lo_init
[1];
1850 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1851 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1853 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1854 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1856 /* error in case values were truncated */
1857 if (info
.lo_device
!= info64
->lo_device
||
1858 info
.lo_rdevice
!= info64
->lo_rdevice
||
1859 info
.lo_inode
!= info64
->lo_inode
||
1860 info
.lo_offset
!= info64
->lo_offset
||
1861 info
.lo_init
[0] != info64
->lo_init
[0] ||
1862 info
.lo_init
[1] != info64
->lo_init
[1])
1865 if (copy_to_user(arg
, &info
, sizeof(info
)))
1871 loop_set_status_compat(struct loop_device
*lo
,
1872 const struct compat_loop_info __user
*arg
)
1874 struct loop_info64 info64
;
1877 ret
= loop_info64_from_compat(arg
, &info64
);
1880 return loop_set_status(lo
, &info64
);
1884 loop_get_status_compat(struct loop_device
*lo
,
1885 struct compat_loop_info __user
*arg
)
1887 struct loop_info64 info64
;
1892 err
= loop_get_status(lo
, &info64
);
1894 err
= loop_info64_to_compat(&info64
, arg
);
1898 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1899 unsigned int cmd
, unsigned long arg
)
1901 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1905 case LOOP_SET_STATUS
:
1906 err
= loop_set_status_compat(lo
,
1907 (const struct compat_loop_info __user
*)arg
);
1909 case LOOP_GET_STATUS
:
1910 err
= loop_get_status_compat(lo
,
1911 (struct compat_loop_info __user
*)arg
);
1913 case LOOP_SET_CAPACITY
:
1915 case LOOP_GET_STATUS64
:
1916 case LOOP_SET_STATUS64
:
1917 case LOOP_CONFIGURE
:
1918 arg
= (unsigned long) compat_ptr(arg
);
1921 case LOOP_CHANGE_FD
:
1922 case LOOP_SET_BLOCK_SIZE
:
1923 case LOOP_SET_DIRECT_IO
:
1924 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1934 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1936 struct loop_device
*lo
;
1939 err
= mutex_lock_killable(&loop_ctl_mutex
);
1942 lo
= bdev
->bd_disk
->private_data
;
1948 atomic_inc(&lo
->lo_refcnt
);
1950 mutex_unlock(&loop_ctl_mutex
);
1954 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1956 struct loop_device
*lo
;
1958 mutex_lock(&loop_ctl_mutex
);
1959 lo
= disk
->private_data
;
1960 if (atomic_dec_return(&lo
->lo_refcnt
))
1963 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1964 if (lo
->lo_state
!= Lo_bound
)
1966 lo
->lo_state
= Lo_rundown
;
1967 mutex_unlock(&loop_ctl_mutex
);
1969 * In autoclear mode, stop the loop thread
1970 * and remove configuration after last close.
1972 __loop_clr_fd(lo
, true);
1974 } else if (lo
->lo_state
== Lo_bound
) {
1976 * Otherwise keep thread (if running) and config,
1977 * but flush possible ongoing bios in thread.
1979 blk_mq_freeze_queue(lo
->lo_queue
);
1980 blk_mq_unfreeze_queue(lo
->lo_queue
);
1984 mutex_unlock(&loop_ctl_mutex
);
1987 static const struct block_device_operations lo_fops
= {
1988 .owner
= THIS_MODULE
,
1990 .release
= lo_release
,
1992 #ifdef CONFIG_COMPAT
1993 .compat_ioctl
= lo_compat_ioctl
,
1998 * And now the modules code and kernel interface.
2000 static int max_loop
;
2001 module_param(max_loop
, int, 0444);
2002 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
2003 module_param(max_part
, int, 0444);
2004 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
2005 MODULE_LICENSE("GPL");
2006 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
2008 int loop_register_transfer(struct loop_func_table
*funcs
)
2010 unsigned int n
= funcs
->number
;
2012 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
2014 xfer_funcs
[n
] = funcs
;
2018 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
2020 struct loop_device
*lo
= ptr
;
2021 struct loop_func_table
*xfer
= data
;
2023 mutex_lock(&loop_ctl_mutex
);
2024 if (lo
->lo_encryption
== xfer
)
2025 loop_release_xfer(lo
);
2026 mutex_unlock(&loop_ctl_mutex
);
2030 int loop_unregister_transfer(int number
)
2032 unsigned int n
= number
;
2033 struct loop_func_table
*xfer
;
2035 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
2038 xfer_funcs
[n
] = NULL
;
2039 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
2043 EXPORT_SYMBOL(loop_register_transfer
);
2044 EXPORT_SYMBOL(loop_unregister_transfer
);
2046 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2047 const struct blk_mq_queue_data
*bd
)
2049 struct request
*rq
= bd
->rq
;
2050 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
2051 struct loop_device
*lo
= rq
->q
->queuedata
;
2053 blk_mq_start_request(rq
);
2055 if (lo
->lo_state
!= Lo_bound
)
2056 return BLK_STS_IOERR
;
2058 switch (req_op(rq
)) {
2060 case REQ_OP_DISCARD
:
2061 case REQ_OP_WRITE_ZEROES
:
2062 cmd
->use_aio
= false;
2065 cmd
->use_aio
= lo
->use_dio
;
2069 /* always use the first bio's css */
2070 #ifdef CONFIG_BLK_CGROUP
2071 if (cmd
->use_aio
&& rq
->bio
&& rq
->bio
->bi_blkg
) {
2072 cmd
->css
= &bio_blkcg(rq
->bio
)->css
;
2077 kthread_queue_work(&lo
->worker
, &cmd
->work
);
2082 static void loop_handle_cmd(struct loop_cmd
*cmd
)
2084 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
2085 const bool write
= op_is_write(req_op(rq
));
2086 struct loop_device
*lo
= rq
->q
->queuedata
;
2089 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
2094 ret
= do_req_filebacked(lo
, rq
);
2096 /* complete non-aio request */
2097 if (!cmd
->use_aio
|| ret
) {
2098 if (ret
== -EOPNOTSUPP
)
2101 cmd
->ret
= ret
? -EIO
: 0;
2102 if (likely(!blk_should_fake_timeout(rq
->q
)))
2103 blk_mq_complete_request(rq
);
2107 static void loop_queue_work(struct kthread_work
*work
)
2109 struct loop_cmd
*cmd
=
2110 container_of(work
, struct loop_cmd
, work
);
2112 loop_handle_cmd(cmd
);
2115 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
2116 unsigned int hctx_idx
, unsigned int numa_node
)
2118 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
2120 kthread_init_work(&cmd
->work
, loop_queue_work
);
2124 static const struct blk_mq_ops loop_mq_ops
= {
2125 .queue_rq
= loop_queue_rq
,
2126 .init_request
= loop_init_request
,
2127 .complete
= lo_complete_rq
,
2130 static int loop_add(struct loop_device
**l
, int i
)
2132 struct loop_device
*lo
;
2133 struct gendisk
*disk
;
2137 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
2141 lo
->lo_state
= Lo_unbound
;
2143 /* allocate id, if @id >= 0, we're requesting that specific id */
2145 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
2149 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
2156 lo
->tag_set
.ops
= &loop_mq_ops
;
2157 lo
->tag_set
.nr_hw_queues
= 1;
2158 lo
->tag_set
.queue_depth
= 128;
2159 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
2160 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
2161 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_STACKING
;
2162 lo
->tag_set
.driver_data
= lo
;
2164 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
2168 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
2169 if (IS_ERR(lo
->lo_queue
)) {
2170 err
= PTR_ERR(lo
->lo_queue
);
2171 goto out_cleanup_tags
;
2173 lo
->lo_queue
->queuedata
= lo
;
2175 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
2178 * By default, we do buffer IO, so it doesn't make sense to enable
2179 * merge because the I/O submitted to backing file is handled page by
2180 * page. For directio mode, merge does help to dispatch bigger request
2181 * to underlayer disk. We will enable merge once directio is enabled.
2183 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
2186 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
2188 goto out_free_queue
;
2191 * Disable partition scanning by default. The in-kernel partition
2192 * scanning can be requested individually per-device during its
2193 * setup. Userspace can always add and remove partitions from all
2194 * devices. The needed partition minors are allocated from the
2195 * extended minor space, the main loop device numbers will continue
2196 * to match the loop minors, regardless of the number of partitions
2199 * If max_part is given, partition scanning is globally enabled for
2200 * all loop devices. The minors for the main loop devices will be
2201 * multiples of max_part.
2203 * Note: Global-for-all-devices, set-only-at-init, read-only module
2204 * parameteters like 'max_loop' and 'max_part' make things needlessly
2205 * complicated, are too static, inflexible and may surprise
2206 * userspace tools. Parameters like this in general should be avoided.
2209 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
2210 disk
->flags
|= GENHD_FL_EXT_DEVT
;
2211 atomic_set(&lo
->lo_refcnt
, 0);
2213 spin_lock_init(&lo
->lo_lock
);
2214 disk
->major
= LOOP_MAJOR
;
2215 disk
->first_minor
= i
<< part_shift
;
2216 disk
->fops
= &lo_fops
;
2217 disk
->private_data
= lo
;
2218 disk
->queue
= lo
->lo_queue
;
2219 sprintf(disk
->disk_name
, "loop%d", i
);
2222 return lo
->lo_number
;
2225 blk_cleanup_queue(lo
->lo_queue
);
2227 blk_mq_free_tag_set(&lo
->tag_set
);
2229 idr_remove(&loop_index_idr
, i
);
2236 static void loop_remove(struct loop_device
*lo
)
2238 del_gendisk(lo
->lo_disk
);
2239 blk_cleanup_queue(lo
->lo_queue
);
2240 blk_mq_free_tag_set(&lo
->tag_set
);
2241 put_disk(lo
->lo_disk
);
2245 static int find_free_cb(int id
, void *ptr
, void *data
)
2247 struct loop_device
*lo
= ptr
;
2248 struct loop_device
**l
= data
;
2250 if (lo
->lo_state
== Lo_unbound
) {
2257 static int loop_lookup(struct loop_device
**l
, int i
)
2259 struct loop_device
*lo
;
2265 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
2268 ret
= lo
->lo_number
;
2273 /* lookup and return a specific i */
2274 lo
= idr_find(&loop_index_idr
, i
);
2277 ret
= lo
->lo_number
;
2283 static void loop_probe(dev_t dev
)
2285 int idx
= MINOR(dev
) >> part_shift
;
2286 struct loop_device
*lo
;
2288 if (max_loop
&& idx
>= max_loop
)
2291 mutex_lock(&loop_ctl_mutex
);
2292 if (loop_lookup(&lo
, idx
) < 0)
2294 mutex_unlock(&loop_ctl_mutex
);
2297 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
2300 struct loop_device
*lo
;
2303 ret
= mutex_lock_killable(&loop_ctl_mutex
);
2310 ret
= loop_lookup(&lo
, parm
);
2315 ret
= loop_add(&lo
, parm
);
2317 case LOOP_CTL_REMOVE
:
2318 ret
= loop_lookup(&lo
, parm
);
2321 if (lo
->lo_state
!= Lo_unbound
) {
2325 if (atomic_read(&lo
->lo_refcnt
) > 0) {
2329 lo
->lo_disk
->private_data
= NULL
;
2330 idr_remove(&loop_index_idr
, lo
->lo_number
);
2333 case LOOP_CTL_GET_FREE
:
2334 ret
= loop_lookup(&lo
, -1);
2337 ret
= loop_add(&lo
, -1);
2339 mutex_unlock(&loop_ctl_mutex
);
2344 static const struct file_operations loop_ctl_fops
= {
2345 .open
= nonseekable_open
,
2346 .unlocked_ioctl
= loop_control_ioctl
,
2347 .compat_ioctl
= loop_control_ioctl
,
2348 .owner
= THIS_MODULE
,
2349 .llseek
= noop_llseek
,
2352 static struct miscdevice loop_misc
= {
2353 .minor
= LOOP_CTRL_MINOR
,
2354 .name
= "loop-control",
2355 .fops
= &loop_ctl_fops
,
2358 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2359 MODULE_ALIAS("devname:loop-control");
2361 static int __init
loop_init(void)
2364 struct loop_device
*lo
;
2369 part_shift
= fls(max_part
);
2372 * Adjust max_part according to part_shift as it is exported
2373 * to user space so that user can decide correct minor number
2374 * if [s]he want to create more devices.
2376 * Note that -1 is required because partition 0 is reserved
2377 * for the whole disk.
2379 max_part
= (1UL << part_shift
) - 1;
2382 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2387 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2393 * If max_loop is specified, create that many devices upfront.
2394 * This also becomes a hard limit. If max_loop is not specified,
2395 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2396 * init time. Loop devices can be requested on-demand with the
2397 * /dev/loop-control interface, or be instantiated by accessing
2398 * a 'dead' device node.
2403 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2405 err
= misc_register(&loop_misc
);
2410 if (__register_blkdev(LOOP_MAJOR
, "loop", loop_probe
)) {
2415 /* pre-create number of devices given by config or max_loop */
2416 mutex_lock(&loop_ctl_mutex
);
2417 for (i
= 0; i
< nr
; i
++)
2419 mutex_unlock(&loop_ctl_mutex
);
2421 printk(KERN_INFO
"loop: module loaded\n");
2425 misc_deregister(&loop_misc
);
2430 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2432 struct loop_device
*lo
= ptr
;
2438 static void __exit
loop_exit(void)
2440 mutex_lock(&loop_ctl_mutex
);
2442 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2443 idr_destroy(&loop_index_idr
);
2445 unregister_blkdev(LOOP_MAJOR
, "loop");
2447 misc_deregister(&loop_misc
);
2449 mutex_unlock(&loop_ctl_mutex
);
2452 module_init(loop_init
);
2453 module_exit(loop_exit
);
2456 static int __init
max_loop_setup(char *str
)
2458 max_loop
= simple_strtol(str
, NULL
, 0);
2462 __setup("max_loop=", max_loop_setup
);