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
,
225 loff_t logical_blocksize
)
227 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
228 sector_t x
= (sector_t
)size
;
229 struct block_device
*bdev
= lo
->lo_device
;
231 if (unlikely((loff_t
)x
!= size
))
233 if (lo
->lo_offset
!= offset
)
234 lo
->lo_offset
= offset
;
235 if (lo
->lo_sizelimit
!= sizelimit
)
236 lo
->lo_sizelimit
= sizelimit
;
237 if (lo
->lo_flags
& LO_FLAGS_BLOCKSIZE
) {
238 lo
->lo_logical_blocksize
= logical_blocksize
;
239 blk_queue_physical_block_size(lo
->lo_queue
, lo
->lo_blocksize
);
240 blk_queue_logical_block_size(lo
->lo_queue
,
241 lo
->lo_logical_blocksize
);
243 set_capacity(lo
->lo_disk
, x
);
244 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
245 /* let user-space know about the new size */
246 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
251 lo_do_transfer(struct loop_device
*lo
, int cmd
,
252 struct page
*rpage
, unsigned roffs
,
253 struct page
*lpage
, unsigned loffs
,
254 int size
, sector_t rblock
)
258 ret
= lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
262 printk_ratelimited(KERN_ERR
263 "loop: Transfer error at byte offset %llu, length %i.\n",
264 (unsigned long long)rblock
<< 9, size
);
268 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
273 iov_iter_bvec(&i
, ITER_BVEC
, bvec
, 1, bvec
->bv_len
);
275 file_start_write(file
);
276 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
277 file_end_write(file
);
279 if (likely(bw
== bvec
->bv_len
))
282 printk_ratelimited(KERN_ERR
283 "loop: Write error at byte offset %llu, length %i.\n",
284 (unsigned long long)*ppos
, bvec
->bv_len
);
290 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
294 struct req_iterator iter
;
297 rq_for_each_segment(bvec
, rq
, iter
) {
298 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
308 * This is the slow, transforming version that needs to double buffer the
309 * data as it cannot do the transformations in place without having direct
310 * access to the destination pages of the backing file.
312 static int lo_write_transfer(struct loop_device
*lo
, struct request
*rq
,
315 struct bio_vec bvec
, b
;
316 struct req_iterator iter
;
320 page
= alloc_page(GFP_NOIO
);
324 rq_for_each_segment(bvec
, rq
, iter
) {
325 ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
.bv_page
,
326 bvec
.bv_offset
, bvec
.bv_len
, pos
>> 9);
332 b
.bv_len
= bvec
.bv_len
;
333 ret
= lo_write_bvec(lo
->lo_backing_file
, &b
, &pos
);
342 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
346 struct req_iterator iter
;
350 rq_for_each_segment(bvec
, rq
, iter
) {
351 iov_iter_bvec(&i
, ITER_BVEC
, &bvec
, 1, bvec
.bv_len
);
352 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
356 flush_dcache_page(bvec
.bv_page
);
358 if (len
!= bvec
.bv_len
) {
361 __rq_for_each_bio(bio
, rq
)
371 static int lo_read_transfer(struct loop_device
*lo
, struct request
*rq
,
374 struct bio_vec bvec
, b
;
375 struct req_iterator iter
;
381 page
= alloc_page(GFP_NOIO
);
385 rq_for_each_segment(bvec
, rq
, iter
) {
390 b
.bv_len
= bvec
.bv_len
;
392 iov_iter_bvec(&i
, ITER_BVEC
, &b
, 1, b
.bv_len
);
393 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
399 ret
= lo_do_transfer(lo
, READ
, page
, 0, bvec
.bv_page
,
400 bvec
.bv_offset
, len
, offset
>> 9);
404 flush_dcache_page(bvec
.bv_page
);
406 if (len
!= bvec
.bv_len
) {
409 __rq_for_each_bio(bio
, rq
)
421 static int lo_discard(struct loop_device
*lo
, struct request
*rq
, loff_t pos
)
424 * We use punch hole to reclaim the free space used by the
425 * image a.k.a. discard. However we do not support discard if
426 * encryption is enabled, because it may give an attacker
427 * useful information.
429 struct file
*file
= lo
->lo_backing_file
;
430 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
433 if ((!file
->f_op
->fallocate
) || lo
->lo_encrypt_key_size
) {
438 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
439 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
445 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
447 struct file
*file
= lo
->lo_backing_file
;
448 int ret
= vfs_fsync(file
, 0);
449 if (unlikely(ret
&& ret
!= -EINVAL
))
455 static void lo_complete_rq(struct request
*rq
)
457 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
459 if (unlikely(req_op(cmd
->rq
) == REQ_OP_READ
&& cmd
->use_aio
&&
460 cmd
->ret
>= 0 && cmd
->ret
< blk_rq_bytes(cmd
->rq
))) {
461 struct bio
*bio
= cmd
->rq
->bio
;
463 bio_advance(bio
, cmd
->ret
);
467 blk_mq_end_request(rq
, cmd
->ret
< 0 ? BLK_STS_IOERR
: BLK_STS_OK
);
470 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
472 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
475 blk_mq_complete_request(cmd
->rq
);
478 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
481 struct iov_iter iter
;
482 struct bio_vec
*bvec
;
483 struct bio
*bio
= cmd
->rq
->bio
;
484 struct file
*file
= lo
->lo_backing_file
;
487 /* nomerge for loop request queue */
488 WARN_ON(cmd
->rq
->bio
!= cmd
->rq
->biotail
);
490 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
491 iov_iter_bvec(&iter
, ITER_BVEC
| rw
, bvec
,
492 bio_segments(bio
), blk_rq_bytes(cmd
->rq
));
494 * This bio may be started from the middle of the 'bvec'
495 * because of bio splitting, so offset from the bvec must
496 * be passed to iov iterator
498 iter
.iov_offset
= bio
->bi_iter
.bi_bvec_done
;
500 cmd
->iocb
.ki_pos
= pos
;
501 cmd
->iocb
.ki_filp
= file
;
502 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
503 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
506 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
508 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
510 if (ret
!= -EIOCBQUEUED
)
511 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
515 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
517 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
518 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
521 * lo_write_simple and lo_read_simple should have been covered
522 * by io submit style function like lo_rw_aio(), one blocker
523 * is that lo_read_simple() need to call flush_dcache_page after
524 * the page is written from kernel, and it isn't easy to handle
525 * this in io submit style function which submits all segments
526 * of the req at one time. And direct read IO doesn't need to
527 * run flush_dcache_page().
529 switch (req_op(rq
)) {
531 return lo_req_flush(lo
, rq
);
533 case REQ_OP_WRITE_ZEROES
:
534 return lo_discard(lo
, rq
, pos
);
537 return lo_write_transfer(lo
, rq
, pos
);
538 else if (cmd
->use_aio
)
539 return lo_rw_aio(lo
, cmd
, pos
, WRITE
);
541 return lo_write_simple(lo
, rq
, pos
);
544 return lo_read_transfer(lo
, rq
, pos
);
545 else if (cmd
->use_aio
)
546 return lo_rw_aio(lo
, cmd
, pos
, READ
);
548 return lo_read_simple(lo
, rq
, pos
);
556 struct switch_request
{
558 struct completion wait
;
561 static inline void loop_update_dio(struct loop_device
*lo
)
563 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
568 * Do the actual switch; called from the BIO completion routine
570 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
572 struct file
*file
= p
->file
;
573 struct file
*old_file
= lo
->lo_backing_file
;
574 struct address_space
*mapping
;
576 /* if no new file, only flush of queued bios requested */
580 mapping
= file
->f_mapping
;
581 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
582 lo
->lo_backing_file
= file
;
583 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
584 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
585 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
586 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
591 * loop_switch performs the hard work of switching a backing store.
592 * First it needs to flush existing IO, it does this by sending a magic
593 * BIO down the pipe. The completion of this BIO does the actual switch.
595 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
597 struct switch_request w
;
601 /* freeze queue and wait for completion of scheduled requests */
602 blk_mq_freeze_queue(lo
->lo_queue
);
604 /* do the switch action */
605 do_loop_switch(lo
, &w
);
608 blk_mq_unfreeze_queue(lo
->lo_queue
);
614 * Helper to flush the IOs in loop, but keeping loop thread running
616 static int loop_flush(struct loop_device
*lo
)
618 /* loop not yet configured, no running thread, nothing to flush */
619 if (lo
->lo_state
!= Lo_bound
)
621 return loop_switch(lo
, NULL
);
624 static void loop_reread_partitions(struct loop_device
*lo
,
625 struct block_device
*bdev
)
630 * bd_mutex has been held already in release path, so don't
631 * acquire it if this function is called in such case.
633 * If the reread partition isn't from release path, lo_refcnt
634 * must be at least one and it can only become zero when the
635 * current holder is released.
637 if (!atomic_read(&lo
->lo_refcnt
))
638 rc
= __blkdev_reread_part(bdev
);
640 rc
= blkdev_reread_part(bdev
);
642 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
643 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
647 * loop_change_fd switched the backing store of a loopback device to
648 * a new file. This is useful for operating system installers to free up
649 * the original file and in High Availability environments to switch to
650 * an alternative location for the content in case of server meltdown.
651 * This can only work if the loop device is used read-only, and if the
652 * new backing store is the same size and type as the old backing store.
654 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
657 struct file
*file
, *old_file
;
662 if (lo
->lo_state
!= Lo_bound
)
665 /* the loop device has to be read-only */
667 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
675 inode
= file
->f_mapping
->host
;
676 old_file
= lo
->lo_backing_file
;
680 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
683 /* size of the new backing store needs to be the same */
684 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
688 error
= loop_switch(lo
, file
);
693 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
694 loop_reread_partitions(lo
, bdev
);
703 static inline int is_loop_device(struct file
*file
)
705 struct inode
*i
= file
->f_mapping
->host
;
707 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
710 /* loop sysfs attributes */
712 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
713 ssize_t (*callback
)(struct loop_device
*, char *))
715 struct gendisk
*disk
= dev_to_disk(dev
);
716 struct loop_device
*lo
= disk
->private_data
;
718 return callback(lo
, page
);
721 #define LOOP_ATTR_RO(_name) \
722 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
723 static ssize_t loop_attr_do_show_##_name(struct device *d, \
724 struct device_attribute *attr, char *b) \
726 return loop_attr_show(d, b, loop_attr_##_name##_show); \
728 static struct device_attribute loop_attr_##_name = \
729 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
731 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
736 spin_lock_irq(&lo
->lo_lock
);
737 if (lo
->lo_backing_file
)
738 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
739 spin_unlock_irq(&lo
->lo_lock
);
741 if (IS_ERR_OR_NULL(p
))
745 memmove(buf
, p
, ret
);
753 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
755 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
758 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
760 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
763 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
765 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
767 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
770 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
772 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
774 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
777 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
779 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
781 return sprintf(buf
, "%s\n", dio
? "1" : "0");
784 LOOP_ATTR_RO(backing_file
);
785 LOOP_ATTR_RO(offset
);
786 LOOP_ATTR_RO(sizelimit
);
787 LOOP_ATTR_RO(autoclear
);
788 LOOP_ATTR_RO(partscan
);
791 static struct attribute
*loop_attrs
[] = {
792 &loop_attr_backing_file
.attr
,
793 &loop_attr_offset
.attr
,
794 &loop_attr_sizelimit
.attr
,
795 &loop_attr_autoclear
.attr
,
796 &loop_attr_partscan
.attr
,
801 static struct attribute_group loop_attribute_group
= {
806 static int loop_sysfs_init(struct loop_device
*lo
)
808 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
809 &loop_attribute_group
);
812 static void loop_sysfs_exit(struct loop_device
*lo
)
814 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
815 &loop_attribute_group
);
818 static void loop_config_discard(struct loop_device
*lo
)
820 struct file
*file
= lo
->lo_backing_file
;
821 struct inode
*inode
= file
->f_mapping
->host
;
822 struct request_queue
*q
= lo
->lo_queue
;
826 * We use punch hole to reclaim the free space used by the
827 * image a.k.a. discard. However we do not support discard if
828 * encryption is enabled, because it may give an attacker
829 * useful information.
831 if ((!file
->f_op
->fallocate
) ||
832 lo
->lo_encrypt_key_size
) {
833 q
->limits
.discard_granularity
= 0;
834 q
->limits
.discard_alignment
= 0;
835 blk_queue_max_discard_sectors(q
, 0);
836 blk_queue_max_write_zeroes_sectors(q
, 0);
837 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
841 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
842 q
->limits
.discard_alignment
= 0;
843 if (lo
->lo_flags
& LO_FLAGS_BLOCKSIZE
)
844 lo_bits
= blksize_bits(lo
->lo_logical_blocksize
);
846 blk_queue_max_discard_sectors(q
, UINT_MAX
>> lo_bits
);
847 blk_queue_max_write_zeroes_sectors(q
, UINT_MAX
>> lo_bits
);
848 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
851 static void loop_unprepare_queue(struct loop_device
*lo
)
853 kthread_flush_worker(&lo
->worker
);
854 kthread_stop(lo
->worker_task
);
857 static int loop_kthread_worker_fn(void *worker_ptr
)
859 current
->flags
|= PF_LESS_THROTTLE
;
860 return kthread_worker_fn(worker_ptr
);
863 static int loop_prepare_queue(struct loop_device
*lo
)
865 kthread_init_worker(&lo
->worker
);
866 lo
->worker_task
= kthread_run(loop_kthread_worker_fn
,
867 &lo
->worker
, "loop%d", lo
->lo_number
);
868 if (IS_ERR(lo
->worker_task
))
870 set_user_nice(lo
->worker_task
, MIN_NICE
);
874 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
875 struct block_device
*bdev
, unsigned int arg
)
877 struct file
*file
, *f
;
879 struct address_space
*mapping
;
880 unsigned lo_blocksize
;
885 /* This is safe, since we have a reference from open(). */
886 __module_get(THIS_MODULE
);
894 if (lo
->lo_state
!= Lo_unbound
)
897 /* Avoid recursion */
899 while (is_loop_device(f
)) {
900 struct loop_device
*l
;
902 if (f
->f_mapping
->host
->i_bdev
== bdev
)
905 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
906 if (l
->lo_state
== Lo_unbound
) {
910 f
= l
->lo_backing_file
;
913 mapping
= file
->f_mapping
;
914 inode
= mapping
->host
;
917 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
920 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
921 !file
->f_op
->write_iter
)
922 lo_flags
|= LO_FLAGS_READ_ONLY
;
924 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
925 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
928 size
= get_loop_size(lo
, file
);
929 if ((loff_t
)(sector_t
)size
!= size
)
931 error
= loop_prepare_queue(lo
);
937 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
940 lo
->lo_blocksize
= lo_blocksize
;
941 lo
->lo_logical_blocksize
= 512;
942 lo
->lo_device
= bdev
;
943 lo
->lo_flags
= lo_flags
;
944 lo
->lo_backing_file
= file
;
947 lo
->lo_sizelimit
= 0;
948 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
949 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
951 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
952 blk_queue_write_cache(lo
->lo_queue
, true, false);
955 set_capacity(lo
->lo_disk
, size
);
956 bd_set_size(bdev
, size
<< 9);
958 /* let user-space know about the new size */
959 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
961 set_blocksize(bdev
, lo_blocksize
);
963 lo
->lo_state
= Lo_bound
;
965 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
966 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
967 loop_reread_partitions(lo
, bdev
);
969 /* Grab the block_device to prevent its destruction after we
970 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
978 /* This is safe: open() is still holding a reference. */
979 module_put(THIS_MODULE
);
984 loop_release_xfer(struct loop_device
*lo
)
987 struct loop_func_table
*xfer
= lo
->lo_encryption
;
991 err
= xfer
->release(lo
);
993 lo
->lo_encryption
= NULL
;
994 module_put(xfer
->owner
);
1000 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1001 const struct loop_info64
*i
)
1006 struct module
*owner
= xfer
->owner
;
1008 if (!try_module_get(owner
))
1011 err
= xfer
->init(lo
, i
);
1015 lo
->lo_encryption
= xfer
;
1020 static int loop_clr_fd(struct loop_device
*lo
)
1022 struct file
*filp
= lo
->lo_backing_file
;
1023 gfp_t gfp
= lo
->old_gfp_mask
;
1024 struct block_device
*bdev
= lo
->lo_device
;
1026 if (lo
->lo_state
!= Lo_bound
)
1030 * If we've explicitly asked to tear down the loop device,
1031 * and it has an elevated reference count, set it for auto-teardown when
1032 * the last reference goes away. This stops $!~#$@ udev from
1033 * preventing teardown because it decided that it needs to run blkid on
1034 * the loopback device whenever they appear. xfstests is notorious for
1035 * failing tests because blkid via udev races with a losetup
1036 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1037 * command to fail with EBUSY.
1039 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1040 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1041 mutex_unlock(&lo
->lo_ctl_mutex
);
1048 /* freeze request queue during the transition */
1049 blk_mq_freeze_queue(lo
->lo_queue
);
1051 spin_lock_irq(&lo
->lo_lock
);
1052 lo
->lo_state
= Lo_rundown
;
1053 lo
->lo_backing_file
= NULL
;
1054 spin_unlock_irq(&lo
->lo_lock
);
1056 loop_release_xfer(lo
);
1057 lo
->transfer
= NULL
;
1059 lo
->lo_device
= NULL
;
1060 lo
->lo_encryption
= NULL
;
1062 lo
->lo_sizelimit
= 0;
1063 lo
->lo_encrypt_key_size
= 0;
1064 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1065 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1066 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1069 invalidate_bdev(bdev
);
1071 set_capacity(lo
->lo_disk
, 0);
1072 loop_sysfs_exit(lo
);
1074 bd_set_size(bdev
, 0);
1075 /* let user-space know about this change */
1076 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1078 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1079 lo
->lo_state
= Lo_unbound
;
1080 /* This is safe: open() is still holding a reference. */
1081 module_put(THIS_MODULE
);
1082 blk_mq_unfreeze_queue(lo
->lo_queue
);
1084 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1085 loop_reread_partitions(lo
, bdev
);
1088 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1089 loop_unprepare_queue(lo
);
1090 mutex_unlock(&lo
->lo_ctl_mutex
);
1092 * Need not hold lo_ctl_mutex to fput backing file.
1093 * Calling fput holding lo_ctl_mutex triggers a circular
1094 * lock dependency possibility warning as fput can take
1095 * bd_mutex which is usually taken before lo_ctl_mutex.
1102 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1105 struct loop_func_table
*xfer
;
1106 kuid_t uid
= current_uid();
1107 int lo_flags
= lo
->lo_flags
;
1109 if (lo
->lo_encrypt_key_size
&&
1110 !uid_eq(lo
->lo_key_owner
, uid
) &&
1111 !capable(CAP_SYS_ADMIN
))
1113 if (lo
->lo_state
!= Lo_bound
)
1115 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1118 /* I/O need to be drained during transfer transition */
1119 blk_mq_freeze_queue(lo
->lo_queue
);
1121 err
= loop_release_xfer(lo
);
1125 if (info
->lo_encrypt_type
) {
1126 unsigned int type
= info
->lo_encrypt_type
;
1128 if (type
>= MAX_LO_CRYPT
)
1130 xfer
= xfer_funcs
[type
];
1136 err
= loop_init_xfer(lo
, xfer
, info
);
1140 if (info
->lo_flags
& LO_FLAGS_BLOCKSIZE
) {
1141 if (!(lo
->lo_flags
& LO_FLAGS_BLOCKSIZE
))
1142 lo
->lo_logical_blocksize
= 512;
1143 lo
->lo_flags
|= LO_FLAGS_BLOCKSIZE
;
1144 if (LO_INFO_BLOCKSIZE(info
) != 512 &&
1145 LO_INFO_BLOCKSIZE(info
) != 1024 &&
1146 LO_INFO_BLOCKSIZE(info
) != 2048 &&
1147 LO_INFO_BLOCKSIZE(info
) != 4096)
1149 if (LO_INFO_BLOCKSIZE(info
) > lo
->lo_blocksize
)
1153 if (lo
->lo_offset
!= info
->lo_offset
||
1154 lo
->lo_sizelimit
!= info
->lo_sizelimit
||
1155 lo
->lo_flags
!= lo_flags
||
1156 ((lo
->lo_flags
& LO_FLAGS_BLOCKSIZE
) &&
1157 lo
->lo_logical_blocksize
!= LO_INFO_BLOCKSIZE(info
))) {
1158 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
,
1159 LO_INFO_BLOCKSIZE(info
))) {
1165 loop_config_discard(lo
);
1167 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1168 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1169 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1170 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1174 lo
->transfer
= xfer
->transfer
;
1175 lo
->ioctl
= xfer
->ioctl
;
1177 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1178 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1179 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1181 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1182 lo
->lo_init
[0] = info
->lo_init
[0];
1183 lo
->lo_init
[1] = info
->lo_init
[1];
1184 if (info
->lo_encrypt_key_size
) {
1185 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1186 info
->lo_encrypt_key_size
);
1187 lo
->lo_key_owner
= uid
;
1190 /* update dio if lo_offset or transfer is changed */
1191 __loop_update_dio(lo
, lo
->use_dio
);
1194 blk_mq_unfreeze_queue(lo
->lo_queue
);
1196 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1197 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1198 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1199 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1200 loop_reread_partitions(lo
, lo
->lo_device
);
1207 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1209 struct file
*file
= lo
->lo_backing_file
;
1213 if (lo
->lo_state
!= Lo_bound
)
1215 error
= vfs_getattr(&file
->f_path
, &stat
,
1216 STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1219 memset(info
, 0, sizeof(*info
));
1220 info
->lo_number
= lo
->lo_number
;
1221 info
->lo_device
= huge_encode_dev(stat
.dev
);
1222 info
->lo_inode
= stat
.ino
;
1223 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1224 info
->lo_offset
= lo
->lo_offset
;
1225 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1226 info
->lo_flags
= lo
->lo_flags
;
1227 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1228 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1229 info
->lo_encrypt_type
=
1230 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1231 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1232 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1233 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1234 lo
->lo_encrypt_key_size
);
1240 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1242 memset(info64
, 0, sizeof(*info64
));
1243 info64
->lo_number
= info
->lo_number
;
1244 info64
->lo_device
= info
->lo_device
;
1245 info64
->lo_inode
= info
->lo_inode
;
1246 info64
->lo_rdevice
= info
->lo_rdevice
;
1247 info64
->lo_offset
= info
->lo_offset
;
1248 info64
->lo_sizelimit
= 0;
1249 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1250 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1251 info64
->lo_flags
= info
->lo_flags
;
1252 info64
->lo_init
[0] = info
->lo_init
[0];
1253 info64
->lo_init
[1] = info
->lo_init
[1];
1254 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1255 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1257 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1258 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1262 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1264 memset(info
, 0, sizeof(*info
));
1265 info
->lo_number
= info64
->lo_number
;
1266 info
->lo_device
= info64
->lo_device
;
1267 info
->lo_inode
= info64
->lo_inode
;
1268 info
->lo_rdevice
= info64
->lo_rdevice
;
1269 info
->lo_offset
= info64
->lo_offset
;
1270 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1271 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1272 info
->lo_flags
= info64
->lo_flags
;
1273 info
->lo_init
[0] = info64
->lo_init
[0];
1274 info
->lo_init
[1] = info64
->lo_init
[1];
1275 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1276 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1278 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1279 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1281 /* error in case values were truncated */
1282 if (info
->lo_device
!= info64
->lo_device
||
1283 info
->lo_rdevice
!= info64
->lo_rdevice
||
1284 info
->lo_inode
!= info64
->lo_inode
||
1285 info
->lo_offset
!= info64
->lo_offset
)
1292 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1294 struct loop_info info
;
1295 struct loop_info64 info64
;
1297 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1299 loop_info64_from_old(&info
, &info64
);
1300 return loop_set_status(lo
, &info64
);
1304 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1306 struct loop_info64 info64
;
1308 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1310 return loop_set_status(lo
, &info64
);
1314 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1315 struct loop_info info
;
1316 struct loop_info64 info64
;
1322 err
= loop_get_status(lo
, &info64
);
1324 err
= loop_info64_to_old(&info64
, &info
);
1325 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1332 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1333 struct loop_info64 info64
;
1339 err
= loop_get_status(lo
, &info64
);
1340 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1346 static int loop_set_capacity(struct loop_device
*lo
)
1348 if (unlikely(lo
->lo_state
!= Lo_bound
))
1351 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
,
1352 lo
->lo_logical_blocksize
);
1355 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1358 if (lo
->lo_state
!= Lo_bound
)
1361 __loop_update_dio(lo
, !!arg
);
1362 if (lo
->use_dio
== !!arg
)
1369 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1370 unsigned int cmd
, unsigned long arg
)
1372 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1375 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1378 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1380 case LOOP_CHANGE_FD
:
1381 err
= loop_change_fd(lo
, bdev
, arg
);
1384 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1385 err
= loop_clr_fd(lo
);
1389 case LOOP_SET_STATUS
:
1391 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1392 err
= loop_set_status_old(lo
,
1393 (struct loop_info __user
*)arg
);
1395 case LOOP_GET_STATUS
:
1396 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1398 case LOOP_SET_STATUS64
:
1400 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1401 err
= loop_set_status64(lo
,
1402 (struct loop_info64 __user
*) arg
);
1404 case LOOP_GET_STATUS64
:
1405 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1407 case LOOP_SET_CAPACITY
:
1409 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1410 err
= loop_set_capacity(lo
);
1412 case LOOP_SET_DIRECT_IO
:
1414 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1415 err
= loop_set_dio(lo
, arg
);
1418 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1420 mutex_unlock(&lo
->lo_ctl_mutex
);
1426 #ifdef CONFIG_COMPAT
1427 struct compat_loop_info
{
1428 compat_int_t lo_number
; /* ioctl r/o */
1429 compat_dev_t lo_device
; /* ioctl r/o */
1430 compat_ulong_t lo_inode
; /* ioctl r/o */
1431 compat_dev_t lo_rdevice
; /* ioctl r/o */
1432 compat_int_t lo_offset
;
1433 compat_int_t lo_encrypt_type
;
1434 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1435 compat_int_t lo_flags
; /* ioctl r/o */
1436 char lo_name
[LO_NAME_SIZE
];
1437 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1438 compat_ulong_t lo_init
[2];
1443 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1444 * - noinlined to reduce stack space usage in main part of driver
1447 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1448 struct loop_info64
*info64
)
1450 struct compat_loop_info info
;
1452 if (copy_from_user(&info
, arg
, sizeof(info
)))
1455 memset(info64
, 0, sizeof(*info64
));
1456 info64
->lo_number
= info
.lo_number
;
1457 info64
->lo_device
= info
.lo_device
;
1458 info64
->lo_inode
= info
.lo_inode
;
1459 info64
->lo_rdevice
= info
.lo_rdevice
;
1460 info64
->lo_offset
= info
.lo_offset
;
1461 info64
->lo_sizelimit
= 0;
1462 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1463 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1464 info64
->lo_flags
= info
.lo_flags
;
1465 info64
->lo_init
[0] = info
.lo_init
[0];
1466 info64
->lo_init
[1] = info
.lo_init
[1];
1467 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1468 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1470 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1471 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1476 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1477 * - noinlined to reduce stack space usage in main part of driver
1480 loop_info64_to_compat(const struct loop_info64
*info64
,
1481 struct compat_loop_info __user
*arg
)
1483 struct compat_loop_info info
;
1485 memset(&info
, 0, sizeof(info
));
1486 info
.lo_number
= info64
->lo_number
;
1487 info
.lo_device
= info64
->lo_device
;
1488 info
.lo_inode
= info64
->lo_inode
;
1489 info
.lo_rdevice
= info64
->lo_rdevice
;
1490 info
.lo_offset
= info64
->lo_offset
;
1491 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1492 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1493 info
.lo_flags
= info64
->lo_flags
;
1494 info
.lo_init
[0] = info64
->lo_init
[0];
1495 info
.lo_init
[1] = info64
->lo_init
[1];
1496 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1497 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1499 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1500 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1502 /* error in case values were truncated */
1503 if (info
.lo_device
!= info64
->lo_device
||
1504 info
.lo_rdevice
!= info64
->lo_rdevice
||
1505 info
.lo_inode
!= info64
->lo_inode
||
1506 info
.lo_offset
!= info64
->lo_offset
||
1507 info
.lo_init
[0] != info64
->lo_init
[0] ||
1508 info
.lo_init
[1] != info64
->lo_init
[1])
1511 if (copy_to_user(arg
, &info
, sizeof(info
)))
1517 loop_set_status_compat(struct loop_device
*lo
,
1518 const struct compat_loop_info __user
*arg
)
1520 struct loop_info64 info64
;
1523 ret
= loop_info64_from_compat(arg
, &info64
);
1526 return loop_set_status(lo
, &info64
);
1530 loop_get_status_compat(struct loop_device
*lo
,
1531 struct compat_loop_info __user
*arg
)
1533 struct loop_info64 info64
;
1539 err
= loop_get_status(lo
, &info64
);
1541 err
= loop_info64_to_compat(&info64
, arg
);
1545 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1546 unsigned int cmd
, unsigned long arg
)
1548 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1552 case LOOP_SET_STATUS
:
1553 mutex_lock(&lo
->lo_ctl_mutex
);
1554 err
= loop_set_status_compat(
1555 lo
, (const struct compat_loop_info __user
*) arg
);
1556 mutex_unlock(&lo
->lo_ctl_mutex
);
1558 case LOOP_GET_STATUS
:
1559 mutex_lock(&lo
->lo_ctl_mutex
);
1560 err
= loop_get_status_compat(
1561 lo
, (struct compat_loop_info __user
*) arg
);
1562 mutex_unlock(&lo
->lo_ctl_mutex
);
1564 case LOOP_SET_CAPACITY
:
1566 case LOOP_GET_STATUS64
:
1567 case LOOP_SET_STATUS64
:
1568 arg
= (unsigned long) compat_ptr(arg
);
1570 case LOOP_CHANGE_FD
:
1571 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1581 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1583 struct loop_device
*lo
;
1586 mutex_lock(&loop_index_mutex
);
1587 lo
= bdev
->bd_disk
->private_data
;
1593 atomic_inc(&lo
->lo_refcnt
);
1595 mutex_unlock(&loop_index_mutex
);
1599 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1601 struct loop_device
*lo
= disk
->private_data
;
1604 if (atomic_dec_return(&lo
->lo_refcnt
))
1607 mutex_lock(&lo
->lo_ctl_mutex
);
1608 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1610 * In autoclear mode, stop the loop thread
1611 * and remove configuration after last close.
1613 err
= loop_clr_fd(lo
);
1618 * Otherwise keep thread (if running) and config,
1619 * but flush possible ongoing bios in thread.
1624 mutex_unlock(&lo
->lo_ctl_mutex
);
1627 static const struct block_device_operations lo_fops
= {
1628 .owner
= THIS_MODULE
,
1630 .release
= lo_release
,
1632 #ifdef CONFIG_COMPAT
1633 .compat_ioctl
= lo_compat_ioctl
,
1638 * And now the modules code and kernel interface.
1640 static int max_loop
;
1641 module_param(max_loop
, int, S_IRUGO
);
1642 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1643 module_param(max_part
, int, S_IRUGO
);
1644 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1645 MODULE_LICENSE("GPL");
1646 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1648 int loop_register_transfer(struct loop_func_table
*funcs
)
1650 unsigned int n
= funcs
->number
;
1652 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1654 xfer_funcs
[n
] = funcs
;
1658 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1660 struct loop_device
*lo
= ptr
;
1661 struct loop_func_table
*xfer
= data
;
1663 mutex_lock(&lo
->lo_ctl_mutex
);
1664 if (lo
->lo_encryption
== xfer
)
1665 loop_release_xfer(lo
);
1666 mutex_unlock(&lo
->lo_ctl_mutex
);
1670 int loop_unregister_transfer(int number
)
1672 unsigned int n
= number
;
1673 struct loop_func_table
*xfer
;
1675 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1678 xfer_funcs
[n
] = NULL
;
1679 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1683 EXPORT_SYMBOL(loop_register_transfer
);
1684 EXPORT_SYMBOL(loop_unregister_transfer
);
1686 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1687 const struct blk_mq_queue_data
*bd
)
1689 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1690 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1692 blk_mq_start_request(bd
->rq
);
1694 if (lo
->lo_state
!= Lo_bound
)
1695 return BLK_STS_IOERR
;
1697 switch (req_op(cmd
->rq
)) {
1699 case REQ_OP_DISCARD
:
1700 case REQ_OP_WRITE_ZEROES
:
1701 cmd
->use_aio
= false;
1704 cmd
->use_aio
= lo
->use_dio
;
1708 kthread_queue_work(&lo
->worker
, &cmd
->work
);
1713 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1715 const bool write
= op_is_write(req_op(cmd
->rq
));
1716 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1719 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1724 ret
= do_req_filebacked(lo
, cmd
->rq
);
1726 /* complete non-aio request */
1727 if (!cmd
->use_aio
|| ret
) {
1728 cmd
->ret
= ret
? -EIO
: 0;
1729 blk_mq_complete_request(cmd
->rq
);
1733 static void loop_queue_work(struct kthread_work
*work
)
1735 struct loop_cmd
*cmd
=
1736 container_of(work
, struct loop_cmd
, work
);
1738 loop_handle_cmd(cmd
);
1741 static int loop_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1742 unsigned int hctx_idx
, unsigned int numa_node
)
1744 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1747 kthread_init_work(&cmd
->work
, loop_queue_work
);
1752 static const struct blk_mq_ops loop_mq_ops
= {
1753 .queue_rq
= loop_queue_rq
,
1754 .init_request
= loop_init_request
,
1755 .complete
= lo_complete_rq
,
1758 static int loop_add(struct loop_device
**l
, int i
)
1760 struct loop_device
*lo
;
1761 struct gendisk
*disk
;
1765 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1769 lo
->lo_state
= Lo_unbound
;
1771 /* allocate id, if @id >= 0, we're requesting that specific id */
1773 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1777 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1784 lo
->tag_set
.ops
= &loop_mq_ops
;
1785 lo
->tag_set
.nr_hw_queues
= 1;
1786 lo
->tag_set
.queue_depth
= 128;
1787 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1788 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1789 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1790 lo
->tag_set
.driver_data
= lo
;
1792 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1796 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1797 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1798 err
= PTR_ERR(lo
->lo_queue
);
1799 goto out_cleanup_tags
;
1801 lo
->lo_queue
->queuedata
= lo
;
1804 * It doesn't make sense to enable merge because the I/O
1805 * submitted to backing file is handled page by page.
1807 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
1810 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1812 goto out_free_queue
;
1815 * Disable partition scanning by default. The in-kernel partition
1816 * scanning can be requested individually per-device during its
1817 * setup. Userspace can always add and remove partitions from all
1818 * devices. The needed partition minors are allocated from the
1819 * extended minor space, the main loop device numbers will continue
1820 * to match the loop minors, regardless of the number of partitions
1823 * If max_part is given, partition scanning is globally enabled for
1824 * all loop devices. The minors for the main loop devices will be
1825 * multiples of max_part.
1827 * Note: Global-for-all-devices, set-only-at-init, read-only module
1828 * parameteters like 'max_loop' and 'max_part' make things needlessly
1829 * complicated, are too static, inflexible and may surprise
1830 * userspace tools. Parameters like this in general should be avoided.
1833 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1834 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1835 mutex_init(&lo
->lo_ctl_mutex
);
1836 atomic_set(&lo
->lo_refcnt
, 0);
1838 spin_lock_init(&lo
->lo_lock
);
1839 disk
->major
= LOOP_MAJOR
;
1840 disk
->first_minor
= i
<< part_shift
;
1841 disk
->fops
= &lo_fops
;
1842 disk
->private_data
= lo
;
1843 disk
->queue
= lo
->lo_queue
;
1844 sprintf(disk
->disk_name
, "loop%d", i
);
1847 return lo
->lo_number
;
1850 blk_cleanup_queue(lo
->lo_queue
);
1852 blk_mq_free_tag_set(&lo
->tag_set
);
1854 idr_remove(&loop_index_idr
, i
);
1861 static void loop_remove(struct loop_device
*lo
)
1863 blk_cleanup_queue(lo
->lo_queue
);
1864 del_gendisk(lo
->lo_disk
);
1865 blk_mq_free_tag_set(&lo
->tag_set
);
1866 put_disk(lo
->lo_disk
);
1870 static int find_free_cb(int id
, void *ptr
, void *data
)
1872 struct loop_device
*lo
= ptr
;
1873 struct loop_device
**l
= data
;
1875 if (lo
->lo_state
== Lo_unbound
) {
1882 static int loop_lookup(struct loop_device
**l
, int i
)
1884 struct loop_device
*lo
;
1890 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1893 ret
= lo
->lo_number
;
1898 /* lookup and return a specific i */
1899 lo
= idr_find(&loop_index_idr
, i
);
1902 ret
= lo
->lo_number
;
1908 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1910 struct loop_device
*lo
;
1911 struct kobject
*kobj
;
1914 mutex_lock(&loop_index_mutex
);
1915 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1917 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1921 kobj
= get_disk(lo
->lo_disk
);
1922 mutex_unlock(&loop_index_mutex
);
1928 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1931 struct loop_device
*lo
;
1934 mutex_lock(&loop_index_mutex
);
1937 ret
= loop_lookup(&lo
, parm
);
1942 ret
= loop_add(&lo
, parm
);
1944 case LOOP_CTL_REMOVE
:
1945 ret
= loop_lookup(&lo
, parm
);
1948 mutex_lock(&lo
->lo_ctl_mutex
);
1949 if (lo
->lo_state
!= Lo_unbound
) {
1951 mutex_unlock(&lo
->lo_ctl_mutex
);
1954 if (atomic_read(&lo
->lo_refcnt
) > 0) {
1956 mutex_unlock(&lo
->lo_ctl_mutex
);
1959 lo
->lo_disk
->private_data
= NULL
;
1960 mutex_unlock(&lo
->lo_ctl_mutex
);
1961 idr_remove(&loop_index_idr
, lo
->lo_number
);
1964 case LOOP_CTL_GET_FREE
:
1965 ret
= loop_lookup(&lo
, -1);
1968 ret
= loop_add(&lo
, -1);
1970 mutex_unlock(&loop_index_mutex
);
1975 static const struct file_operations loop_ctl_fops
= {
1976 .open
= nonseekable_open
,
1977 .unlocked_ioctl
= loop_control_ioctl
,
1978 .compat_ioctl
= loop_control_ioctl
,
1979 .owner
= THIS_MODULE
,
1980 .llseek
= noop_llseek
,
1983 static struct miscdevice loop_misc
= {
1984 .minor
= LOOP_CTRL_MINOR
,
1985 .name
= "loop-control",
1986 .fops
= &loop_ctl_fops
,
1989 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1990 MODULE_ALIAS("devname:loop-control");
1992 static int __init
loop_init(void)
1995 unsigned long range
;
1996 struct loop_device
*lo
;
1999 err
= misc_register(&loop_misc
);
2005 part_shift
= fls(max_part
);
2008 * Adjust max_part according to part_shift as it is exported
2009 * to user space so that user can decide correct minor number
2010 * if [s]he want to create more devices.
2012 * Note that -1 is required because partition 0 is reserved
2013 * for the whole disk.
2015 max_part
= (1UL << part_shift
) - 1;
2018 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2023 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2029 * If max_loop is specified, create that many devices upfront.
2030 * This also becomes a hard limit. If max_loop is not specified,
2031 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2032 * init time. Loop devices can be requested on-demand with the
2033 * /dev/loop-control interface, or be instantiated by accessing
2034 * a 'dead' device node.
2038 range
= max_loop
<< part_shift
;
2040 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2041 range
= 1UL << MINORBITS
;
2044 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2049 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2050 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2052 /* pre-create number of devices given by config or max_loop */
2053 mutex_lock(&loop_index_mutex
);
2054 for (i
= 0; i
< nr
; i
++)
2056 mutex_unlock(&loop_index_mutex
);
2058 printk(KERN_INFO
"loop: module loaded\n");
2062 misc_deregister(&loop_misc
);
2066 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2068 struct loop_device
*lo
= ptr
;
2074 static void __exit
loop_exit(void)
2076 unsigned long range
;
2078 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2080 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2081 idr_destroy(&loop_index_idr
);
2083 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2084 unregister_blkdev(LOOP_MAJOR
, "loop");
2086 misc_deregister(&loop_misc
);
2089 module_init(loop_init
);
2090 module_exit(loop_exit
);
2093 static int __init
max_loop_setup(char *str
)
2095 max_loop
= simple_strtol(str
, NULL
, 0);
2099 __setup("max_loop=", max_loop_setup
);