2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
81 #include <linux/uaccess.h>
83 static DEFINE_IDR(loop_index_idr
);
84 static DEFINE_MUTEX(loop_index_mutex
);
87 static int part_shift
;
89 static int transfer_xor(struct loop_device
*lo
, int cmd
,
90 struct page
*raw_page
, unsigned raw_off
,
91 struct page
*loop_page
, unsigned loop_off
,
92 int size
, sector_t real_block
)
94 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
95 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
107 key
= lo
->lo_encrypt_key
;
108 keysize
= lo
->lo_encrypt_key_size
;
109 for (i
= 0; i
< size
; i
++)
110 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
112 kunmap_atomic(loop_buf
);
113 kunmap_atomic(raw_buf
);
118 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
120 if (unlikely(info
->lo_encrypt_key_size
<= 0))
125 static struct loop_func_table none_funcs
= {
126 .number
= LO_CRYPT_NONE
,
129 static struct loop_func_table xor_funcs
= {
130 .number
= LO_CRYPT_XOR
,
131 .transfer
= transfer_xor
,
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
141 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
145 /* Compute loopsize in bytes */
146 loopsize
= i_size_read(file
->f_mapping
->host
);
149 /* offset is beyond i_size, weird but possible */
153 if (sizelimit
> 0 && sizelimit
< loopsize
)
154 loopsize
= sizelimit
;
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
159 return loopsize
>> 9;
162 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
164 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
167 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
169 struct file
*file
= lo
->lo_backing_file
;
170 struct address_space
*mapping
= file
->f_mapping
;
171 struct inode
*inode
= mapping
->host
;
172 unsigned short sb_bsize
= 0;
173 unsigned dio_align
= 0;
176 if (inode
->i_sb
->s_bdev
) {
177 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
178 dio_align
= sb_bsize
- 1;
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane applications should be PAGE_SIZE aligned
192 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
193 !(lo
->lo_offset
& dio_align
) &&
194 mapping
->a_ops
->direct_IO
&&
203 if (lo
->use_dio
== use_dio
)
206 /* flush dirty pages before changing direct IO */
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
214 blk_mq_freeze_queue(lo
->lo_queue
);
215 lo
->use_dio
= use_dio
;
217 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
219 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
220 blk_mq_unfreeze_queue(lo
->lo_queue
);
224 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
226 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
227 sector_t x
= (sector_t
)size
;
228 struct block_device
*bdev
= lo
->lo_device
;
230 if (unlikely((loff_t
)x
!= size
))
232 if (lo
->lo_offset
!= offset
)
233 lo
->lo_offset
= offset
;
234 if (lo
->lo_sizelimit
!= sizelimit
)
235 lo
->lo_sizelimit
= sizelimit
;
236 set_capacity(lo
->lo_disk
, x
);
237 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
238 /* let user-space know about the new size */
239 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
244 lo_do_transfer(struct loop_device
*lo
, int cmd
,
245 struct page
*rpage
, unsigned roffs
,
246 struct page
*lpage
, unsigned loffs
,
247 int size
, sector_t rblock
)
251 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
255 printk_ratelimited(KERN_ERR
256 "loop: Transfer error at byte offset %llu, length %i.\n",
257 (unsigned long long)rblock
<< 9, size
);
261 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
266 iov_iter_bvec(&i
, ITER_BVEC
, bvec
, 1, bvec
->bv_len
);
268 file_start_write(file
);
269 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
270 file_end_write(file
);
272 if (likely(bw
== bvec
->bv_len
))
275 printk_ratelimited(KERN_ERR
276 "loop: Write error at byte offset %llu, length %i.\n",
277 (unsigned long long)*ppos
, bvec
->bv_len
);
283 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
287 struct req_iterator iter
;
290 rq_for_each_segment(bvec
, rq
, iter
) {
291 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
301 * This is the slow, transforming version that needs to double buffer the
302 * data as it cannot do the transformations in place without having direct
303 * access to the destination pages of the backing file.
305 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
308 struct bio_vec bvec
, b
;
309 struct req_iterator iter
;
313 page
= alloc_page(GFP_NOIO
);
317 rq_for_each_segment(bvec
, rq
, iter
) {
318 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
319 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
325 b
.bv_len
= bvec
.bv_len
;
326 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
335 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
339 struct req_iterator iter
;
343 rq_for_each_segment(bvec
, rq
, iter
) {
344 iov_iter_bvec(&i
, ITER_BVEC
, &bvec
, 1, bvec
.bv_len
);
345 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
349 flush_dcache_page(bvec
.bv_page
);
351 if (len
!= bvec
.bv_len
) {
354 __rq_for_each_bio(bio
, rq
)
364 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
367 struct bio_vec bvec
, b
;
368 struct req_iterator iter
;
374 page
= alloc_page(GFP_NOIO
);
378 rq_for_each_segment(bvec
, rq
, iter
) {
383 b
.bv_len
= bvec
.bv_len
;
385 iov_iter_bvec(&i
, ITER_BVEC
, &b
, 1, b
.bv_len
);
386 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
392 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
393 bvec
.bv_offset
, len
, offset
>> 9);
397 flush_dcache_page(bvec
.bv_page
);
399 if (len
!= bvec
.bv_len
) {
402 __rq_for_each_bio(bio
, rq
)
414 static int lo_discard(struct loop_device
*lo
, struct request
*rq
, loff_t pos
)
417 * We use punch hole to reclaim the free space used by the
418 * image a.k.a. discard. However we do not support discard if
419 * encryption is enabled, because it may give an attacker
420 * useful information.
422 struct file
*file
= lo
->lo_backing_file
;
423 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
426 if ((!file
->f_op
->fallocate
) || lo
->lo_encrypt_key_size
) {
431 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
432 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
438 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
440 struct file
*file
= lo
->lo_backing_file
;
441 int ret
= vfs_fsync(file
, 0);
442 if (unlikely(ret
&& ret
!= -EINVAL
))
448 static void lo_complete_rq(struct request
*rq
)
450 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
452 if (unlikely(req_op(cmd
->rq
) == REQ_OP_READ
&& cmd
->use_aio
&&
453 cmd
->ret
>= 0 && cmd
->ret
< blk_rq_bytes(cmd
->rq
))) {
454 struct bio
*bio
= cmd
->rq
->bio
;
456 bio_advance(bio
, cmd
->ret
);
460 blk_mq_end_request(rq
, cmd
->ret
< 0 ? BLK_STS_IOERR
: BLK_STS_OK
);
463 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
465 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
468 blk_mq_complete_request(cmd
->rq
);
471 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
474 struct iov_iter iter
;
475 struct bio_vec
*bvec
;
476 struct bio
*bio
= cmd
->rq
->bio
;
477 struct file
*file
= lo
->lo_backing_file
;
480 /* nomerge for loop request queue */
481 WARN_ON(cmd
->rq
->bio
!= cmd
->rq
->biotail
);
483 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
484 iov_iter_bvec(&iter
, ITER_BVEC
| rw
, bvec
,
485 bio_segments(bio
), blk_rq_bytes(cmd
->rq
));
487 * This bio may be started from the middle of the 'bvec'
488 * because of bio splitting, so offset from the bvec must
489 * be passed to iov iterator
491 iter
.iov_offset
= bio
->bi_iter
.bi_bvec_done
;
493 cmd
->iocb
.ki_pos
= pos
;
494 cmd
->iocb
.ki_filp
= file
;
495 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
496 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
499 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
501 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
503 if (ret
!= -EIOCBQUEUED
)
504 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
508 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
510 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
511 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
514 * lo_write_simple and lo_read_simple should have been covered
515 * by io submit style function like lo_rw_aio(), one blocker
516 * is that lo_read_simple() need to call flush_dcache_page after
517 * the page is written from kernel, and it isn't easy to handle
518 * this in io submit style function which submits all segments
519 * of the req at one time. And direct read IO doesn't need to
520 * run flush_dcache_page().
522 switch (req_op(rq
)) {
524 return lo_req_flush(lo
, rq
);
526 case REQ_OP_WRITE_ZEROES
:
527 return lo_discard(lo
, rq
, pos
);
530 return lo_write_transfer(lo
, rq
, pos
);
531 else if (cmd
->use_aio
)
532 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
534 return lo_write_simple(lo
, rq
, pos
);
537 return lo_read_transfer(lo
, rq
, pos
);
538 else if (cmd
->use_aio
)
539 return lo_rw_aio(lo
, cmd
, pos
, READ
);
541 return lo_read_simple(lo
, rq
, pos
);
549 struct switch_request
{
550 struct file
*file
, *virt_file
;
551 struct completion wait
;
554 static inline void loop_update_dio(struct loop_device
*lo
)
556 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
561 * Do the actual switch; called from the BIO completion routine
563 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
565 struct file
*file
= p
->file
;
566 struct file
*old_file
= lo
->lo_backing_file
;
567 struct address_space
*mapping
;
569 /* if no new file, only flush of queued bios requested */
573 mapping
= file
->f_mapping
;
574 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
575 lo
->lo_backing_file
= file
;
576 lo
->lo_backing_virt_file
= p
->virt_file
;
577 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
578 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
579 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
580 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
585 * loop_switch performs the hard work of switching a backing store.
586 * First it needs to flush existing IO, it does this by sending a magic
587 * BIO down the pipe. The completion of this BIO does the actual switch.
589 static int loop_switch(struct loop_device
*lo
, struct file
*file
,
590 struct file
*virt_file
)
592 struct switch_request w
;
595 w
.virt_file
= virt_file
;
597 /* freeze queue and wait for completion of scheduled requests */
598 blk_mq_freeze_queue(lo
->lo_queue
);
600 /* do the switch action */
601 do_loop_switch(lo
, &w
);
604 blk_mq_unfreeze_queue(lo
->lo_queue
);
610 * Helper to flush the IOs in loop, but keeping loop thread running
612 static int loop_flush(struct loop_device
*lo
)
614 /* loop not yet configured, no running thread, nothing to flush */
615 if (lo
->lo_state
!= Lo_bound
)
617 return loop_switch(lo
, NULL
, NULL
);
620 static struct file
*loop_real_file(struct file
*file
)
622 struct file
*f
= NULL
;
624 if (file
->f_path
.dentry
->d_sb
->s_op
->real_loop
)
625 f
= file
->f_path
.dentry
->d_sb
->s_op
->real_loop(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
);
652 * loop_change_fd switched the backing store of a loopback device to
653 * a new file. This is useful for operating system installers to free up
654 * the original file and in High Availability environments to switch to
655 * an alternative location for the content in case of server meltdown.
656 * This can only work if the loop device is used read-only, and if the
657 * new backing store is the same size and type as the old backing store.
659 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
662 struct file
*file
, *old_file
;
663 struct file
*f
, *virt_file
= NULL
, *old_virt_file
;
668 if (lo
->lo_state
!= Lo_bound
)
671 /* the loop device has to be read-only */
673 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
680 f
= loop_real_file(file
);
687 inode
= file
->f_mapping
->host
;
688 old_file
= lo
->lo_backing_file
;
689 old_virt_file
= lo
->lo_backing_virt_file
;
693 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
696 /* size of the new backing store needs to be the same */
697 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
701 error
= loop_switch(lo
, file
, virt_file
);
708 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
709 loop_reread_partitions(lo
, bdev
);
720 static inline int is_loop_device(struct file
*file
)
722 struct inode
*i
= file
->f_mapping
->host
;
724 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
729 * no get/put for file.
731 struct file
*loop_backing_file(struct super_block
*sb
)
734 struct loop_device
*l
;
737 if (MAJOR(sb
->s_dev
) == LOOP_MAJOR
) {
738 l
= sb
->s_bdev
->bd_disk
->private_data
;
739 ret
= l
->lo_backing_file
;
743 EXPORT_SYMBOL_GPL(loop_backing_file
);
745 /* loop sysfs attributes */
747 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
748 ssize_t (*callback
)(struct loop_device
*, char *))
750 struct gendisk
*disk
= dev_to_disk(dev
);
751 struct loop_device
*lo
= disk
->private_data
;
753 return callback(lo
, page
);
756 #define LOOP_ATTR_RO(_name) \
757 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
758 static ssize_t loop_attr_do_show_##_name(struct device *d, \
759 struct device_attribute *attr, char *b) \
761 return loop_attr_show(d, b, loop_attr_##_name##_show); \
763 static struct device_attribute loop_attr_##_name = \
764 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
766 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
771 spin_lock_irq(&lo
->lo_lock
);
772 if (lo
->lo_backing_file
)
773 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
774 spin_unlock_irq(&lo
->lo_lock
);
776 if (IS_ERR_OR_NULL(p
))
780 memmove(buf
, p
, ret
);
788 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
790 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
793 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
795 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
798 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
800 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
802 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
805 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
807 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
809 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
812 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
814 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
816 return sprintf(buf
, "%s\n", dio
? "1" : "0");
819 LOOP_ATTR_RO(backing_file
);
820 LOOP_ATTR_RO(offset
);
821 LOOP_ATTR_RO(sizelimit
);
822 LOOP_ATTR_RO(autoclear
);
823 LOOP_ATTR_RO(partscan
);
826 static struct attribute
*loop_attrs
[] = {
827 &loop_attr_backing_file
.attr
,
828 &loop_attr_offset
.attr
,
829 &loop_attr_sizelimit
.attr
,
830 &loop_attr_autoclear
.attr
,
831 &loop_attr_partscan
.attr
,
836 static struct attribute_group loop_attribute_group
= {
841 static int loop_sysfs_init(struct loop_device
*lo
)
843 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
844 &loop_attribute_group
);
847 static void loop_sysfs_exit(struct loop_device
*lo
)
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 queue_flag_clear_unlocked(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 queue_flag_set_unlocked(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
)
909 struct file
*file
, *f
, *virt_file
= NULL
;
911 struct address_space
*mapping
;
912 unsigned lo_blocksize
;
917 /* This is safe, since we have a reference from open(). */
918 __module_get(THIS_MODULE
);
924 f
= loop_real_file(file
);
932 if (lo
->lo_state
!= Lo_unbound
)
935 /* Avoid recursion */
937 while (is_loop_device(f
)) {
938 struct loop_device
*l
;
940 if (f
->f_mapping
->host
->i_bdev
== bdev
)
943 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
944 if (l
->lo_state
== Lo_unbound
) {
948 f
= l
->lo_backing_file
;
951 mapping
= file
->f_mapping
;
952 inode
= mapping
->host
;
955 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
958 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
959 !file
->f_op
->write_iter
)
960 lo_flags
|= LO_FLAGS_READ_ONLY
;
962 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
963 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
966 size
= get_loop_size(lo
, file
);
967 if ((loff_t
)(sector_t
)size
!= size
)
969 error
= loop_prepare_queue(lo
);
975 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
978 lo
->lo_blocksize
= lo_blocksize
;
979 lo
->lo_device
= bdev
;
980 lo
->lo_flags
= lo_flags
;
981 lo
->lo_backing_file
= file
;
982 lo
->lo_backing_virt_file
= virt_file
;
985 lo
->lo_sizelimit
= 0;
986 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
987 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
989 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
990 blk_queue_write_cache(lo
->lo_queue
, true, false);
993 set_capacity(lo
->lo_disk
, size
);
994 bd_set_size(bdev
, size
<< 9);
996 /* let user-space know about the new size */
997 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
999 set_blocksize(bdev
, lo_blocksize
);
1001 lo
->lo_state
= Lo_bound
;
1003 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1004 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
1005 loop_reread_partitions(lo
, bdev
);
1007 /* Grab the block_device to prevent its destruction after we
1008 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
1018 /* This is safe: open() is still holding a reference. */
1019 module_put(THIS_MODULE
);
1024 loop_release_xfer(struct loop_device
*lo
)
1027 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1031 err
= xfer
->release(lo
);
1032 lo
->transfer
= NULL
;
1033 lo
->lo_encryption
= NULL
;
1034 module_put(xfer
->owner
);
1040 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1041 const struct loop_info64
*i
)
1046 struct module
*owner
= xfer
->owner
;
1048 if (!try_module_get(owner
))
1051 err
= xfer
->init(lo
, i
);
1055 lo
->lo_encryption
= xfer
;
1060 static int loop_clr_fd(struct loop_device
*lo
)
1062 struct file
*filp
= lo
->lo_backing_file
;
1063 struct file
*virt_filp
= lo
->lo_backing_virt_file
;
1064 gfp_t gfp
= lo
->old_gfp_mask
;
1065 struct block_device
*bdev
= lo
->lo_device
;
1067 if (lo
->lo_state
!= Lo_bound
)
1071 * If we've explicitly asked to tear down the loop device,
1072 * and it has an elevated reference count, set it for auto-teardown when
1073 * the last reference goes away. This stops $!~#$@ udev from
1074 * preventing teardown because it decided that it needs to run blkid on
1075 * the loopback device whenever they appear. xfstests is notorious for
1076 * failing tests because blkid via udev races with a losetup
1077 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1078 * command to fail with EBUSY.
1080 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1081 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1082 mutex_unlock(&lo
->lo_ctl_mutex
);
1089 /* freeze request queue during the transition */
1090 blk_mq_freeze_queue(lo
->lo_queue
);
1092 spin_lock_irq(&lo
->lo_lock
);
1093 lo
->lo_state
= Lo_rundown
;
1094 lo
->lo_backing_file
= NULL
;
1095 lo
->lo_backing_virt_file
= NULL
;
1096 spin_unlock_irq(&lo
->lo_lock
);
1098 loop_release_xfer(lo
);
1099 lo
->transfer
= NULL
;
1101 lo
->lo_device
= NULL
;
1102 lo
->lo_encryption
= NULL
;
1104 lo
->lo_sizelimit
= 0;
1105 lo
->lo_encrypt_key_size
= 0;
1106 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1107 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1108 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1111 invalidate_bdev(bdev
);
1113 set_capacity(lo
->lo_disk
, 0);
1114 loop_sysfs_exit(lo
);
1116 bd_set_size(bdev
, 0);
1117 /* let user-space know about this change */
1118 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1120 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1121 lo
->lo_state
= Lo_unbound
;
1122 /* This is safe: open() is still holding a reference. */
1123 module_put(THIS_MODULE
);
1124 blk_mq_unfreeze_queue(lo
->lo_queue
);
1126 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1127 loop_reread_partitions(lo
, bdev
);
1130 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1131 loop_unprepare_queue(lo
);
1132 mutex_unlock(&lo
->lo_ctl_mutex
);
1134 * Need not hold lo_ctl_mutex to fput backing file.
1135 * Calling fput holding lo_ctl_mutex triggers a circular
1136 * lock dependency possibility warning as fput can take
1137 * bd_mutex which is usually taken before lo_ctl_mutex.
1146 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1149 struct loop_func_table
*xfer
;
1150 kuid_t uid
= current_uid();
1152 if (lo
->lo_encrypt_key_size
&&
1153 !uid_eq(lo
->lo_key_owner
, uid
) &&
1154 !capable(CAP_SYS_ADMIN
))
1156 if (lo
->lo_state
!= Lo_bound
)
1158 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1161 /* I/O need to be drained during transfer transition */
1162 blk_mq_freeze_queue(lo
->lo_queue
);
1164 err
= loop_release_xfer(lo
);
1168 if (info
->lo_encrypt_type
) {
1169 unsigned int type
= info
->lo_encrypt_type
;
1171 if (type
>= MAX_LO_CRYPT
)
1173 xfer
= xfer_funcs
[type
];
1179 err
= loop_init_xfer(lo
, xfer
, info
);
1183 if (lo
->lo_offset
!= info
->lo_offset
||
1184 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1185 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
)) {
1191 loop_config_discard(lo
);
1193 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1194 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1195 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1196 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1200 lo
->transfer
= xfer
->transfer
;
1201 lo
->ioctl
= xfer
->ioctl
;
1203 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1204 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1205 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1207 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1208 lo
->lo_init
[0] = info
->lo_init
[0];
1209 lo
->lo_init
[1] = info
->lo_init
[1];
1210 if (info
->lo_encrypt_key_size
) {
1211 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1212 info
->lo_encrypt_key_size
);
1213 lo
->lo_key_owner
= uid
;
1216 /* update dio if lo_offset or transfer is changed */
1217 __loop_update_dio(lo
, lo
->use_dio
);
1220 blk_mq_unfreeze_queue(lo
->lo_queue
);
1222 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1223 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1224 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1225 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1226 loop_reread_partitions(lo
, lo
->lo_device
);
1233 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1235 struct file
*file
= lo
->lo_backing_file
;
1239 if (lo
->lo_state
!= Lo_bound
)
1241 error
= vfs_getattr(&file
->f_path
, &stat
,
1242 STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1245 memset(info
, 0, sizeof(*info
));
1246 info
->lo_number
= lo
->lo_number
;
1247 info
->lo_device
= huge_encode_dev(stat
.dev
);
1248 info
->lo_inode
= stat
.ino
;
1249 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1250 info
->lo_offset
= lo
->lo_offset
;
1251 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1252 info
->lo_flags
= lo
->lo_flags
;
1253 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1254 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1255 info
->lo_encrypt_type
=
1256 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1257 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1258 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1259 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1260 lo
->lo_encrypt_key_size
);
1266 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1268 memset(info64
, 0, sizeof(*info64
));
1269 info64
->lo_number
= info
->lo_number
;
1270 info64
->lo_device
= info
->lo_device
;
1271 info64
->lo_inode
= info
->lo_inode
;
1272 info64
->lo_rdevice
= info
->lo_rdevice
;
1273 info64
->lo_offset
= info
->lo_offset
;
1274 info64
->lo_sizelimit
= 0;
1275 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1276 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1277 info64
->lo_flags
= info
->lo_flags
;
1278 info64
->lo_init
[0] = info
->lo_init
[0];
1279 info64
->lo_init
[1] = info
->lo_init
[1];
1280 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1281 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1283 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1284 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1288 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1290 memset(info
, 0, sizeof(*info
));
1291 info
->lo_number
= info64
->lo_number
;
1292 info
->lo_device
= info64
->lo_device
;
1293 info
->lo_inode
= info64
->lo_inode
;
1294 info
->lo_rdevice
= info64
->lo_rdevice
;
1295 info
->lo_offset
= info64
->lo_offset
;
1296 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1297 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1298 info
->lo_flags
= info64
->lo_flags
;
1299 info
->lo_init
[0] = info64
->lo_init
[0];
1300 info
->lo_init
[1] = info64
->lo_init
[1];
1301 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1302 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1304 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1305 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1307 /* error in case values were truncated */
1308 if (info
->lo_device
!= info64
->lo_device
||
1309 info
->lo_rdevice
!= info64
->lo_rdevice
||
1310 info
->lo_inode
!= info64
->lo_inode
||
1311 info
->lo_offset
!= info64
->lo_offset
)
1318 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1320 struct loop_info info
;
1321 struct loop_info64 info64
;
1323 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1325 loop_info64_from_old(&info
, &info64
);
1326 return loop_set_status(lo
, &info64
);
1330 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1332 struct loop_info64 info64
;
1334 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1336 return loop_set_status(lo
, &info64
);
1340 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1341 struct loop_info info
;
1342 struct loop_info64 info64
;
1348 err
= loop_get_status(lo
, &info64
);
1350 err
= loop_info64_to_old(&info64
, &info
);
1351 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1358 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1359 struct loop_info64 info64
;
1365 err
= loop_get_status(lo
, &info64
);
1366 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1372 static int loop_set_capacity(struct loop_device
*lo
)
1374 if (unlikely(lo
->lo_state
!= Lo_bound
))
1377 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1380 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1383 if (lo
->lo_state
!= Lo_bound
)
1386 __loop_update_dio(lo
, !!arg
);
1387 if (lo
->use_dio
== !!arg
)
1394 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1395 unsigned int cmd
, unsigned long arg
)
1397 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1400 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1403 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1405 case LOOP_CHANGE_FD
:
1406 err
= loop_change_fd(lo
, bdev
, arg
);
1409 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1410 err
= loop_clr_fd(lo
);
1414 case LOOP_SET_STATUS
:
1416 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1417 err
= loop_set_status_old(lo
,
1418 (struct loop_info __user
*)arg
);
1420 case LOOP_GET_STATUS
:
1421 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1423 case LOOP_SET_STATUS64
:
1425 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1426 err
= loop_set_status64(lo
,
1427 (struct loop_info64 __user
*) arg
);
1429 case LOOP_GET_STATUS64
:
1430 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1432 case LOOP_SET_CAPACITY
:
1434 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1435 err
= loop_set_capacity(lo
);
1437 case LOOP_SET_DIRECT_IO
:
1439 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1440 err
= loop_set_dio(lo
, arg
);
1443 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1445 mutex_unlock(&lo
->lo_ctl_mutex
);
1451 #ifdef CONFIG_COMPAT
1452 struct compat_loop_info
{
1453 compat_int_t lo_number
; /* ioctl r/o */
1454 compat_dev_t lo_device
; /* ioctl r/o */
1455 compat_ulong_t lo_inode
; /* ioctl r/o */
1456 compat_dev_t lo_rdevice
; /* ioctl r/o */
1457 compat_int_t lo_offset
;
1458 compat_int_t lo_encrypt_type
;
1459 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1460 compat_int_t lo_flags
; /* ioctl r/o */
1461 char lo_name
[LO_NAME_SIZE
];
1462 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1463 compat_ulong_t lo_init
[2];
1468 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1469 * - noinlined to reduce stack space usage in main part of driver
1472 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1473 struct loop_info64
*info64
)
1475 struct compat_loop_info info
;
1477 if (copy_from_user(&info
, arg
, sizeof(info
)))
1480 memset(info64
, 0, sizeof(*info64
));
1481 info64
->lo_number
= info
.lo_number
;
1482 info64
->lo_device
= info
.lo_device
;
1483 info64
->lo_inode
= info
.lo_inode
;
1484 info64
->lo_rdevice
= info
.lo_rdevice
;
1485 info64
->lo_offset
= info
.lo_offset
;
1486 info64
->lo_sizelimit
= 0;
1487 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1488 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1489 info64
->lo_flags
= info
.lo_flags
;
1490 info64
->lo_init
[0] = info
.lo_init
[0];
1491 info64
->lo_init
[1] = info
.lo_init
[1];
1492 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1493 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1495 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1496 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1501 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1502 * - noinlined to reduce stack space usage in main part of driver
1505 loop_info64_to_compat(const struct loop_info64
*info64
,
1506 struct compat_loop_info __user
*arg
)
1508 struct compat_loop_info info
;
1510 memset(&info
, 0, sizeof(info
));
1511 info
.lo_number
= info64
->lo_number
;
1512 info
.lo_device
= info64
->lo_device
;
1513 info
.lo_inode
= info64
->lo_inode
;
1514 info
.lo_rdevice
= info64
->lo_rdevice
;
1515 info
.lo_offset
= info64
->lo_offset
;
1516 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1517 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1518 info
.lo_flags
= info64
->lo_flags
;
1519 info
.lo_init
[0] = info64
->lo_init
[0];
1520 info
.lo_init
[1] = info64
->lo_init
[1];
1521 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1522 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1524 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1525 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1527 /* error in case values were truncated */
1528 if (info
.lo_device
!= info64
->lo_device
||
1529 info
.lo_rdevice
!= info64
->lo_rdevice
||
1530 info
.lo_inode
!= info64
->lo_inode
||
1531 info
.lo_offset
!= info64
->lo_offset
||
1532 info
.lo_init
[0] != info64
->lo_init
[0] ||
1533 info
.lo_init
[1] != info64
->lo_init
[1])
1536 if (copy_to_user(arg
, &info
, sizeof(info
)))
1542 loop_set_status_compat(struct loop_device
*lo
,
1543 const struct compat_loop_info __user
*arg
)
1545 struct loop_info64 info64
;
1548 ret
= loop_info64_from_compat(arg
, &info64
);
1551 return loop_set_status(lo
, &info64
);
1555 loop_get_status_compat(struct loop_device
*lo
,
1556 struct compat_loop_info __user
*arg
)
1558 struct loop_info64 info64
;
1564 err
= loop_get_status(lo
, &info64
);
1566 err
= loop_info64_to_compat(&info64
, arg
);
1570 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1571 unsigned int cmd
, unsigned long arg
)
1573 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1577 case LOOP_SET_STATUS
:
1578 mutex_lock(&lo
->lo_ctl_mutex
);
1579 err
= loop_set_status_compat(
1580 lo
, (const struct compat_loop_info __user
*) arg
);
1581 mutex_unlock(&lo
->lo_ctl_mutex
);
1583 case LOOP_GET_STATUS
:
1584 mutex_lock(&lo
->lo_ctl_mutex
);
1585 err
= loop_get_status_compat(
1586 lo
, (struct compat_loop_info __user
*) arg
);
1587 mutex_unlock(&lo
->lo_ctl_mutex
);
1589 case LOOP_SET_CAPACITY
:
1591 case LOOP_GET_STATUS64
:
1592 case LOOP_SET_STATUS64
:
1593 arg
= (unsigned long) compat_ptr(arg
);
1595 case LOOP_CHANGE_FD
:
1596 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1606 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1608 struct loop_device
*lo
;
1611 mutex_lock(&loop_index_mutex
);
1612 lo
= bdev
->bd_disk
->private_data
;
1618 atomic_inc(&lo
->lo_refcnt
);
1620 mutex_unlock(&loop_index_mutex
);
1624 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1626 struct loop_device
*lo
= disk
->private_data
;
1629 if (atomic_dec_return(&lo
->lo_refcnt
))
1632 mutex_lock(&lo
->lo_ctl_mutex
);
1633 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1635 * In autoclear mode, stop the loop thread
1636 * and remove configuration after last close.
1638 err
= loop_clr_fd(lo
);
1643 * Otherwise keep thread (if running) and config,
1644 * but flush possible ongoing bios in thread.
1649 mutex_unlock(&lo
->lo_ctl_mutex
);
1652 static const struct block_device_operations lo_fops
= {
1653 .owner
= THIS_MODULE
,
1655 .release
= lo_release
,
1657 #ifdef CONFIG_COMPAT
1658 .compat_ioctl
= lo_compat_ioctl
,
1663 * And now the modules code and kernel interface.
1665 static int max_loop
;
1666 module_param(max_loop
, int, S_IRUGO
);
1667 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1668 module_param(max_part
, int, S_IRUGO
);
1669 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1670 MODULE_LICENSE("GPL");
1671 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1673 int loop_register_transfer(struct loop_func_table
*funcs
)
1675 unsigned int n
= funcs
->number
;
1677 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1679 xfer_funcs
[n
] = funcs
;
1683 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1685 struct loop_device
*lo
= ptr
;
1686 struct loop_func_table
*xfer
= data
;
1688 mutex_lock(&lo
->lo_ctl_mutex
);
1689 if (lo
->lo_encryption
== xfer
)
1690 loop_release_xfer(lo
);
1691 mutex_unlock(&lo
->lo_ctl_mutex
);
1695 int loop_unregister_transfer(int number
)
1697 unsigned int n
= number
;
1698 struct loop_func_table
*xfer
;
1700 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1703 xfer_funcs
[n
] = NULL
;
1704 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1708 EXPORT_SYMBOL(loop_register_transfer
);
1709 EXPORT_SYMBOL(loop_unregister_transfer
);
1711 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1712 const struct blk_mq_queue_data
*bd
)
1714 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1715 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1717 blk_mq_start_request(bd
->rq
);
1719 if (lo
->lo_state
!= Lo_bound
)
1720 return BLK_STS_IOERR
;
1722 switch (req_op(cmd
->rq
)) {
1724 case REQ_OP_DISCARD
:
1725 case REQ_OP_WRITE_ZEROES
:
1726 cmd
->use_aio
= false;
1729 cmd
->use_aio
= lo
->use_dio
;
1733 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1738 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1740 const bool write
= op_is_write(req_op(cmd
->rq
));
1741 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1744 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1749 ret
= do_req_filebacked(lo
, cmd
->rq
);
1751 /* complete non-aio request */
1752 if (!cmd
->use_aio
|| ret
) {
1753 cmd
->ret
= ret
? -EIO
: 0;
1754 blk_mq_complete_request(cmd
->rq
);
1758 static void loop_queue_work(struct kthread_work
*work
)
1760 struct loop_cmd
*cmd
=
1761 container_of(work
, struct loop_cmd
, work
);
1763 loop_handle_cmd(cmd
);
1766 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1767 unsigned int hctx_idx
, unsigned int numa_node
)
1769 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1772 kthread_init_work(&cmd
->work
, loop_queue_work
);
1777 static const struct blk_mq_ops loop_mq_ops
= {
1778 .queue_rq
= loop_queue_rq
,
1779 .init_request
= loop_init_request
,
1780 .complete
= lo_complete_rq
,
1783 static int loop_add(struct loop_device
**l
, int i
)
1785 struct loop_device
*lo
;
1786 struct gendisk
*disk
;
1790 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1794 lo
->lo_state
= Lo_unbound
;
1796 /* allocate id, if @id >= 0, we're requesting that specific id */
1798 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1802 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1809 lo
->tag_set
.ops
= &loop_mq_ops
;
1810 lo
->tag_set
.nr_hw_queues
= 1;
1811 lo
->tag_set
.queue_depth
= 128;
1812 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1813 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1814 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1815 lo
->tag_set
.driver_data
= lo
;
1817 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1821 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1822 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1823 err
= PTR_ERR(lo
->lo_queue
);
1824 goto out_cleanup_tags
;
1826 lo
->lo_queue
->queuedata
= lo
;
1829 * It doesn't make sense to enable merge because the I/O
1830 * submitted to backing file is handled page by page.
1832 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
1835 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1837 goto out_free_queue
;
1840 * Disable partition scanning by default. The in-kernel partition
1841 * scanning can be requested individually per-device during its
1842 * setup. Userspace can always add and remove partitions from all
1843 * devices. The needed partition minors are allocated from the
1844 * extended minor space, the main loop device numbers will continue
1845 * to match the loop minors, regardless of the number of partitions
1848 * If max_part is given, partition scanning is globally enabled for
1849 * all loop devices. The minors for the main loop devices will be
1850 * multiples of max_part.
1852 * Note: Global-for-all-devices, set-only-at-init, read-only module
1853 * parameteters like 'max_loop' and 'max_part' make things needlessly
1854 * complicated, are too static, inflexible and may surprise
1855 * userspace tools. Parameters like this in general should be avoided.
1858 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1859 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1860 mutex_init(&lo
->lo_ctl_mutex
);
1861 atomic_set(&lo
->lo_refcnt
, 0);
1863 spin_lock_init(&lo
->lo_lock
);
1864 disk
->major
= LOOP_MAJOR
;
1865 disk
->first_minor
= i
<< part_shift
;
1866 disk
->fops
= &lo_fops
;
1867 disk
->private_data
= lo
;
1868 disk
->queue
= lo
->lo_queue
;
1869 sprintf(disk
->disk_name
, "loop%d", i
);
1872 return lo
->lo_number
;
1875 blk_cleanup_queue(lo
->lo_queue
);
1877 blk_mq_free_tag_set(&lo
->tag_set
);
1879 idr_remove(&loop_index_idr
, i
);
1886 static void loop_remove(struct loop_device
*lo
)
1888 blk_cleanup_queue(lo
->lo_queue
);
1889 del_gendisk(lo
->lo_disk
);
1890 blk_mq_free_tag_set(&lo
->tag_set
);
1891 put_disk(lo
->lo_disk
);
1895 static int find_free_cb(int id
, void *ptr
, void *data
)
1897 struct loop_device
*lo
= ptr
;
1898 struct loop_device
**l
= data
;
1900 if (lo
->lo_state
== Lo_unbound
) {
1907 static int loop_lookup(struct loop_device
**l
, int i
)
1909 struct loop_device
*lo
;
1915 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1918 ret
= lo
->lo_number
;
1923 /* lookup and return a specific i */
1924 lo
= idr_find(&loop_index_idr
, i
);
1927 ret
= lo
->lo_number
;
1933 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1935 struct loop_device
*lo
;
1936 struct kobject
*kobj
;
1939 mutex_lock(&loop_index_mutex
);
1940 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1942 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1946 kobj
= get_disk(lo
->lo_disk
);
1947 mutex_unlock(&loop_index_mutex
);
1953 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1956 struct loop_device
*lo
;
1959 mutex_lock(&loop_index_mutex
);
1962 ret
= loop_lookup(&lo
, parm
);
1967 ret
= loop_add(&lo
, parm
);
1969 case LOOP_CTL_REMOVE
:
1970 ret
= loop_lookup(&lo
, parm
);
1973 mutex_lock(&lo
->lo_ctl_mutex
);
1974 if (lo
->lo_state
!= Lo_unbound
) {
1976 mutex_unlock(&lo
->lo_ctl_mutex
);
1979 if (atomic_read(&lo
->lo_refcnt
) > 0) {
1981 mutex_unlock(&lo
->lo_ctl_mutex
);
1984 lo
->lo_disk
->private_data
= NULL
;
1985 mutex_unlock(&lo
->lo_ctl_mutex
);
1986 idr_remove(&loop_index_idr
, lo
->lo_number
);
1989 case LOOP_CTL_GET_FREE
:
1990 ret
= loop_lookup(&lo
, -1);
1993 ret
= loop_add(&lo
, -1);
1995 mutex_unlock(&loop_index_mutex
);
2000 static const struct file_operations loop_ctl_fops
= {
2001 .open
= nonseekable_open
,
2002 .unlocked_ioctl
= loop_control_ioctl
,
2003 .compat_ioctl
= loop_control_ioctl
,
2004 .owner
= THIS_MODULE
,
2005 .llseek
= noop_llseek
,
2008 static struct miscdevice loop_misc
= {
2009 .minor
= LOOP_CTRL_MINOR
,
2010 .name
= "loop-control",
2011 .fops
= &loop_ctl_fops
,
2014 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2015 MODULE_ALIAS("devname:loop-control");
2017 static int __init
loop_init(void)
2020 unsigned long range
;
2021 struct loop_device
*lo
;
2024 err
= misc_register(&loop_misc
);
2030 part_shift
= fls(max_part
);
2033 * Adjust max_part according to part_shift as it is exported
2034 * to user space so that user can decide correct minor number
2035 * if [s]he want to create more devices.
2037 * Note that -1 is required because partition 0 is reserved
2038 * for the whole disk.
2040 max_part
= (1UL << part_shift
) - 1;
2043 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2048 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2054 * If max_loop is specified, create that many devices upfront.
2055 * This also becomes a hard limit. If max_loop is not specified,
2056 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2057 * init time. Loop devices can be requested on-demand with the
2058 * /dev/loop-control interface, or be instantiated by accessing
2059 * a 'dead' device node.
2063 range
= max_loop
<< part_shift
;
2065 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2066 range
= 1UL << MINORBITS
;
2069 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2074 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2075 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2077 /* pre-create number of devices given by config or max_loop */
2078 mutex_lock(&loop_index_mutex
);
2079 for (i
= 0; i
< nr
; i
++)
2081 mutex_unlock(&loop_index_mutex
);
2083 printk(KERN_INFO
"loop: module loaded\n");
2087 misc_deregister(&loop_misc
);
2091 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2093 struct loop_device
*lo
= ptr
;
2099 static void __exit
loop_exit(void)
2101 unsigned long range
;
2103 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2105 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2106 idr_destroy(&loop_index_idr
);
2108 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2109 unregister_blkdev(LOOP_MAJOR
, "loop");
2111 misc_deregister(&loop_misc
);
2114 module_init(loop_init
);
2115 module_exit(loop_exit
);
2118 static int __init
max_loop_setup(char *str
)
2120 max_loop
= simple_strtol(str
, NULL
, 0);
2124 __setup("max_loop=", max_loop_setup
);