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 <asm/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 appplications should be PAGE_SIZE algined
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
);
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
);
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
);
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 inline void handle_partial_read(struct loop_cmd
*cmd
, long bytes
)
450 if (bytes
< 0 || (cmd
->rq
->cmd_flags
& REQ_WRITE
))
453 if (unlikely(bytes
< blk_rq_bytes(cmd
->rq
))) {
454 struct bio
*bio
= cmd
->rq
->bio
;
456 bio_advance(bio
, bytes
);
461 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
, long ret2
)
463 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
464 struct request
*rq
= cmd
->rq
;
466 handle_partial_read(cmd
, ret
);
473 blk_mq_complete_request(rq
, ret
);
476 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
479 struct iov_iter iter
;
480 struct bio_vec
*bvec
;
481 struct bio
*bio
= cmd
->rq
->bio
;
482 struct file
*file
= lo
->lo_backing_file
;
485 /* nomerge for loop request queue */
486 WARN_ON(cmd
->rq
->bio
!= cmd
->rq
->biotail
);
488 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
489 iov_iter_bvec(&iter
, ITER_BVEC
| rw
, bvec
,
490 bio_segments(bio
), blk_rq_bytes(cmd
->rq
));
492 * This bio may be started from the middle of the 'bvec'
493 * because of bio splitting, so offset from the bvec must
494 * be passed to iov iterator
496 iter
.iov_offset
= bio
->bi_iter
.bi_bvec_done
;
498 cmd
->iocb
.ki_pos
= pos
;
499 cmd
->iocb
.ki_filp
= file
;
500 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
501 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
504 ret
= file
->f_op
->write_iter(&cmd
->iocb
, &iter
);
506 ret
= file
->f_op
->read_iter(&cmd
->iocb
, &iter
);
508 if (ret
!= -EIOCBQUEUED
)
509 cmd
->iocb
.ki_complete(&cmd
->iocb
, ret
, 0);
514 static inline int lo_rw_simple(struct loop_device
*lo
,
515 struct request
*rq
, loff_t pos
, bool rw
)
517 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
520 return lo_rw_aio(lo
, cmd
, pos
, rw
);
523 * lo_write_simple and lo_read_simple should have been covered
524 * by io submit style function like lo_rw_aio(), one blocker
525 * is that lo_read_simple() need to call flush_dcache_page after
526 * the page is written from kernel, and it isn't easy to handle
527 * this in io submit style function which submits all segments
528 * of the req at one time. And direct read IO doesn't need to
529 * run flush_dcache_page().
532 return lo_write_simple(lo
, rq
, pos
);
534 return lo_read_simple(lo
, rq
, pos
);
537 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
542 pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
544 if (rq
->cmd_flags
& REQ_WRITE
) {
545 if (rq
->cmd_flags
& REQ_FLUSH
)
546 ret
= lo_req_flush(lo
, rq
);
547 else if (rq
->cmd_flags
& REQ_DISCARD
)
548 ret
= lo_discard(lo
, rq
, pos
);
549 else if (lo
->transfer
)
550 ret
= lo_write_transfer(lo
, rq
, pos
);
552 ret
= lo_rw_simple(lo
, rq
, pos
, WRITE
);
556 ret
= lo_read_transfer(lo
, rq
, pos
);
558 ret
= lo_rw_simple(lo
, rq
, pos
, READ
);
564 struct switch_request
{
566 struct completion wait
;
569 static inline void loop_update_dio(struct loop_device
*lo
)
571 __loop_update_dio(lo
, io_is_direct(lo
->lo_backing_file
) |
576 * Do the actual switch; called from the BIO completion routine
578 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
580 struct file
*file
= p
->file
;
581 struct file
*old_file
= lo
->lo_backing_file
;
582 struct address_space
*mapping
;
584 /* if no new file, only flush of queued bios requested */
588 mapping
= file
->f_mapping
;
589 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
590 lo
->lo_backing_file
= file
;
591 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
592 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
593 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
594 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
599 * loop_switch performs the hard work of switching a backing store.
600 * First it needs to flush existing IO, it does this by sending a magic
601 * BIO down the pipe. The completion of this BIO does the actual switch.
603 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
605 struct switch_request w
;
609 /* freeze queue and wait for completion of scheduled requests */
610 blk_mq_freeze_queue(lo
->lo_queue
);
612 /* do the switch action */
613 do_loop_switch(lo
, &w
);
616 blk_mq_unfreeze_queue(lo
->lo_queue
);
622 * Helper to flush the IOs in loop, but keeping loop thread running
624 static int loop_flush(struct loop_device
*lo
)
626 /* loop not yet configured, no running thread, nothing to flush */
627 if (lo
->lo_state
!= Lo_bound
)
629 return loop_switch(lo
, NULL
);
632 static void loop_reread_partitions(struct loop_device
*lo
,
633 struct block_device
*bdev
)
638 * bd_mutex has been held already in release path, so don't
639 * acquire it if this function is called in such case.
641 * If the reread partition isn't from release path, lo_refcnt
642 * must be at least one and it can only become zero when the
643 * current holder is released.
645 if (!atomic_read(&lo
->lo_refcnt
))
646 rc
= __blkdev_reread_part(bdev
);
648 rc
= blkdev_reread_part(bdev
);
650 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
651 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
655 * loop_change_fd switched the backing store of a loopback device to
656 * a new file. This is useful for operating system installers to free up
657 * the original file and in High Availability environments to switch to
658 * an alternative location for the content in case of server meltdown.
659 * This can only work if the loop device is used read-only, and if the
660 * new backing store is the same size and type as the old backing store.
662 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
665 struct file
*file
, *old_file
;
670 if (lo
->lo_state
!= Lo_bound
)
673 /* the loop device has to be read-only */
675 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
683 inode
= file
->f_mapping
->host
;
684 old_file
= lo
->lo_backing_file
;
688 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
691 /* size of the new backing store needs to be the same */
692 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
696 error
= loop_switch(lo
, file
);
701 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
702 loop_reread_partitions(lo
, bdev
);
711 static inline int is_loop_device(struct file
*file
)
713 struct inode
*i
= file
->f_mapping
->host
;
715 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
718 /* loop sysfs attributes */
720 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
721 ssize_t (*callback
)(struct loop_device
*, char *))
723 struct gendisk
*disk
= dev_to_disk(dev
);
724 struct loop_device
*lo
= disk
->private_data
;
726 return callback(lo
, page
);
729 #define LOOP_ATTR_RO(_name) \
730 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
731 static ssize_t loop_attr_do_show_##_name(struct device *d, \
732 struct device_attribute *attr, char *b) \
734 return loop_attr_show(d, b, loop_attr_##_name##_show); \
736 static struct device_attribute loop_attr_##_name = \
737 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
739 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
744 spin_lock_irq(&lo
->lo_lock
);
745 if (lo
->lo_backing_file
)
746 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
747 spin_unlock_irq(&lo
->lo_lock
);
749 if (IS_ERR_OR_NULL(p
))
753 memmove(buf
, p
, ret
);
761 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
763 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
766 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
768 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
771 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
773 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
775 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
778 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
780 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
782 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
785 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
787 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
789 return sprintf(buf
, "%s\n", dio
? "1" : "0");
792 LOOP_ATTR_RO(backing_file
);
793 LOOP_ATTR_RO(offset
);
794 LOOP_ATTR_RO(sizelimit
);
795 LOOP_ATTR_RO(autoclear
);
796 LOOP_ATTR_RO(partscan
);
799 static struct attribute
*loop_attrs
[] = {
800 &loop_attr_backing_file
.attr
,
801 &loop_attr_offset
.attr
,
802 &loop_attr_sizelimit
.attr
,
803 &loop_attr_autoclear
.attr
,
804 &loop_attr_partscan
.attr
,
809 static struct attribute_group loop_attribute_group
= {
814 static int loop_sysfs_init(struct loop_device
*lo
)
816 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
817 &loop_attribute_group
);
820 static void loop_sysfs_exit(struct loop_device
*lo
)
822 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
823 &loop_attribute_group
);
826 static void loop_config_discard(struct loop_device
*lo
)
828 struct file
*file
= lo
->lo_backing_file
;
829 struct inode
*inode
= file
->f_mapping
->host
;
830 struct request_queue
*q
= lo
->lo_queue
;
833 * We use punch hole to reclaim the free space used by the
834 * image a.k.a. discard. However we do not support discard if
835 * encryption is enabled, because it may give an attacker
836 * useful information.
838 if ((!file
->f_op
->fallocate
) ||
839 lo
->lo_encrypt_key_size
) {
840 q
->limits
.discard_granularity
= 0;
841 q
->limits
.discard_alignment
= 0;
842 blk_queue_max_discard_sectors(q
, 0);
843 q
->limits
.discard_zeroes_data
= 0;
844 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
848 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
849 q
->limits
.discard_alignment
= 0;
850 blk_queue_max_discard_sectors(q
, UINT_MAX
>> 9);
851 q
->limits
.discard_zeroes_data
= 1;
852 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
855 static void loop_unprepare_queue(struct loop_device
*lo
)
857 flush_kthread_worker(&lo
->worker
);
858 kthread_stop(lo
->worker_task
);
861 static int loop_prepare_queue(struct loop_device
*lo
)
863 init_kthread_worker(&lo
->worker
);
864 lo
->worker_task
= kthread_run(kthread_worker_fn
,
865 &lo
->worker
, "loop%d", lo
->lo_number
);
866 if (IS_ERR(lo
->worker_task
))
868 set_user_nice(lo
->worker_task
, MIN_NICE
);
872 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
873 struct block_device
*bdev
, unsigned int arg
)
875 struct file
*file
, *f
;
877 struct address_space
*mapping
;
878 unsigned lo_blocksize
;
883 /* This is safe, since we have a reference from open(). */
884 __module_get(THIS_MODULE
);
892 if (lo
->lo_state
!= Lo_unbound
)
895 /* Avoid recursion */
897 while (is_loop_device(f
)) {
898 struct loop_device
*l
;
900 if (f
->f_mapping
->host
->i_bdev
== bdev
)
903 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
904 if (l
->lo_state
== Lo_unbound
) {
908 f
= l
->lo_backing_file
;
911 mapping
= file
->f_mapping
;
912 inode
= mapping
->host
;
915 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
918 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
919 !file
->f_op
->write_iter
)
920 lo_flags
|= LO_FLAGS_READ_ONLY
;
922 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
923 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
926 size
= get_loop_size(lo
, file
);
927 if ((loff_t
)(sector_t
)size
!= size
)
929 error
= loop_prepare_queue(lo
);
935 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
938 lo
->lo_blocksize
= lo_blocksize
;
939 lo
->lo_device
= bdev
;
940 lo
->lo_flags
= lo_flags
;
941 lo
->lo_backing_file
= file
;
944 lo
->lo_sizelimit
= 0;
945 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
946 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
948 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
949 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
952 set_capacity(lo
->lo_disk
, size
);
953 bd_set_size(bdev
, size
<< 9);
955 /* let user-space know about the new size */
956 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
958 set_blocksize(bdev
, lo_blocksize
);
960 lo
->lo_state
= Lo_bound
;
962 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
963 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
964 loop_reread_partitions(lo
, bdev
);
966 /* Grab the block_device to prevent its destruction after we
967 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
975 /* This is safe: open() is still holding a reference. */
976 module_put(THIS_MODULE
);
981 loop_release_xfer(struct loop_device
*lo
)
984 struct loop_func_table
*xfer
= lo
->lo_encryption
;
988 err
= xfer
->release(lo
);
990 lo
->lo_encryption
= NULL
;
991 module_put(xfer
->owner
);
997 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
998 const struct loop_info64
*i
)
1003 struct module
*owner
= xfer
->owner
;
1005 if (!try_module_get(owner
))
1008 err
= xfer
->init(lo
, i
);
1012 lo
->lo_encryption
= xfer
;
1017 static int loop_clr_fd(struct loop_device
*lo
)
1019 struct file
*filp
= lo
->lo_backing_file
;
1020 gfp_t gfp
= lo
->old_gfp_mask
;
1021 struct block_device
*bdev
= lo
->lo_device
;
1023 if (lo
->lo_state
!= Lo_bound
)
1027 * If we've explicitly asked to tear down the loop device,
1028 * and it has an elevated reference count, set it for auto-teardown when
1029 * the last reference goes away. This stops $!~#$@ udev from
1030 * preventing teardown because it decided that it needs to run blkid on
1031 * the loopback device whenever they appear. xfstests is notorious for
1032 * failing tests because blkid via udev races with a losetup
1033 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1034 * command to fail with EBUSY.
1036 if (atomic_read(&lo
->lo_refcnt
) > 1) {
1037 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1038 mutex_unlock(&lo
->lo_ctl_mutex
);
1045 /* freeze request queue during the transition */
1046 blk_mq_freeze_queue(lo
->lo_queue
);
1048 spin_lock_irq(&lo
->lo_lock
);
1049 lo
->lo_state
= Lo_rundown
;
1050 lo
->lo_backing_file
= NULL
;
1051 spin_unlock_irq(&lo
->lo_lock
);
1053 loop_release_xfer(lo
);
1054 lo
->transfer
= NULL
;
1056 lo
->lo_device
= NULL
;
1057 lo
->lo_encryption
= NULL
;
1059 lo
->lo_sizelimit
= 0;
1060 lo
->lo_encrypt_key_size
= 0;
1061 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1062 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1063 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1064 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1067 invalidate_bdev(bdev
);
1069 set_capacity(lo
->lo_disk
, 0);
1070 loop_sysfs_exit(lo
);
1072 bd_set_size(bdev
, 0);
1073 /* let user-space know about this change */
1074 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1076 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1077 lo
->lo_state
= Lo_unbound
;
1078 /* This is safe: open() is still holding a reference. */
1079 module_put(THIS_MODULE
);
1080 blk_mq_unfreeze_queue(lo
->lo_queue
);
1082 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1083 loop_reread_partitions(lo
, bdev
);
1086 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1087 loop_unprepare_queue(lo
);
1088 mutex_unlock(&lo
->lo_ctl_mutex
);
1090 * Need not hold lo_ctl_mutex to fput backing file.
1091 * Calling fput holding lo_ctl_mutex triggers a circular
1092 * lock dependency possibility warning as fput can take
1093 * bd_mutex which is usually taken before lo_ctl_mutex.
1100 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1103 struct loop_func_table
*xfer
;
1104 kuid_t uid
= current_uid();
1106 if (lo
->lo_encrypt_key_size
&&
1107 !uid_eq(lo
->lo_key_owner
, uid
) &&
1108 !capable(CAP_SYS_ADMIN
))
1110 if (lo
->lo_state
!= Lo_bound
)
1112 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1115 /* I/O need to be drained during transfer transition */
1116 blk_mq_freeze_queue(lo
->lo_queue
);
1118 err
= loop_release_xfer(lo
);
1122 if (info
->lo_encrypt_type
) {
1123 unsigned int type
= info
->lo_encrypt_type
;
1125 if (type
>= MAX_LO_CRYPT
)
1127 xfer
= xfer_funcs
[type
];
1133 err
= loop_init_xfer(lo
, xfer
, info
);
1137 if (lo
->lo_offset
!= info
->lo_offset
||
1138 lo
->lo_sizelimit
!= info
->lo_sizelimit
)
1139 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
)) {
1144 loop_config_discard(lo
);
1146 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1147 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1148 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1149 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1153 lo
->transfer
= xfer
->transfer
;
1154 lo
->ioctl
= xfer
->ioctl
;
1156 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1157 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1158 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1160 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1161 lo
->lo_init
[0] = info
->lo_init
[0];
1162 lo
->lo_init
[1] = info
->lo_init
[1];
1163 if (info
->lo_encrypt_key_size
) {
1164 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1165 info
->lo_encrypt_key_size
);
1166 lo
->lo_key_owner
= uid
;
1169 /* update dio if lo_offset or transfer is changed */
1170 __loop_update_dio(lo
, lo
->use_dio
);
1173 blk_mq_unfreeze_queue(lo
->lo_queue
);
1175 if (!err
&& (info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1176 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1177 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1178 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1179 loop_reread_partitions(lo
, lo
->lo_device
);
1186 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1188 struct file
*file
= lo
->lo_backing_file
;
1192 if (lo
->lo_state
!= Lo_bound
)
1194 error
= vfs_getattr(&file
->f_path
, &stat
);
1197 memset(info
, 0, sizeof(*info
));
1198 info
->lo_number
= lo
->lo_number
;
1199 info
->lo_device
= huge_encode_dev(stat
.dev
);
1200 info
->lo_inode
= stat
.ino
;
1201 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1202 info
->lo_offset
= lo
->lo_offset
;
1203 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1204 info
->lo_flags
= lo
->lo_flags
;
1205 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1206 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1207 info
->lo_encrypt_type
=
1208 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1209 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1210 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1211 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1212 lo
->lo_encrypt_key_size
);
1218 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1220 memset(info64
, 0, sizeof(*info64
));
1221 info64
->lo_number
= info
->lo_number
;
1222 info64
->lo_device
= info
->lo_device
;
1223 info64
->lo_inode
= info
->lo_inode
;
1224 info64
->lo_rdevice
= info
->lo_rdevice
;
1225 info64
->lo_offset
= info
->lo_offset
;
1226 info64
->lo_sizelimit
= 0;
1227 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1228 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1229 info64
->lo_flags
= info
->lo_flags
;
1230 info64
->lo_init
[0] = info
->lo_init
[0];
1231 info64
->lo_init
[1] = info
->lo_init
[1];
1232 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1233 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1235 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1236 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1240 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1242 memset(info
, 0, sizeof(*info
));
1243 info
->lo_number
= info64
->lo_number
;
1244 info
->lo_device
= info64
->lo_device
;
1245 info
->lo_inode
= info64
->lo_inode
;
1246 info
->lo_rdevice
= info64
->lo_rdevice
;
1247 info
->lo_offset
= info64
->lo_offset
;
1248 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1249 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1250 info
->lo_flags
= info64
->lo_flags
;
1251 info
->lo_init
[0] = info64
->lo_init
[0];
1252 info
->lo_init
[1] = info64
->lo_init
[1];
1253 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1254 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1256 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1257 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1259 /* error in case values were truncated */
1260 if (info
->lo_device
!= info64
->lo_device
||
1261 info
->lo_rdevice
!= info64
->lo_rdevice
||
1262 info
->lo_inode
!= info64
->lo_inode
||
1263 info
->lo_offset
!= info64
->lo_offset
)
1270 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1272 struct loop_info info
;
1273 struct loop_info64 info64
;
1275 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1277 loop_info64_from_old(&info
, &info64
);
1278 return loop_set_status(lo
, &info64
);
1282 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1284 struct loop_info64 info64
;
1286 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1288 return loop_set_status(lo
, &info64
);
1292 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1293 struct loop_info info
;
1294 struct loop_info64 info64
;
1300 err
= loop_get_status(lo
, &info64
);
1302 err
= loop_info64_to_old(&info64
, &info
);
1303 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1310 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1311 struct loop_info64 info64
;
1317 err
= loop_get_status(lo
, &info64
);
1318 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1324 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1326 if (unlikely(lo
->lo_state
!= Lo_bound
))
1329 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1332 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1335 if (lo
->lo_state
!= Lo_bound
)
1338 __loop_update_dio(lo
, !!arg
);
1339 if (lo
->use_dio
== !!arg
)
1346 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1348 if (lo
->lo_state
!= Lo_bound
)
1351 if (arg
< 512 || arg
> PAGE_SIZE
|| !is_power_of_2(arg
))
1354 blk_mq_freeze_queue(lo
->lo_queue
);
1356 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1357 loop_update_dio(lo
);
1359 blk_mq_unfreeze_queue(lo
->lo_queue
);
1364 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1365 unsigned int cmd
, unsigned long arg
)
1367 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1370 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1373 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1375 case LOOP_CHANGE_FD
:
1376 err
= loop_change_fd(lo
, bdev
, arg
);
1379 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1380 err
= loop_clr_fd(lo
);
1384 case LOOP_SET_STATUS
:
1386 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1387 err
= loop_set_status_old(lo
,
1388 (struct loop_info __user
*)arg
);
1390 case LOOP_GET_STATUS
:
1391 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1393 case LOOP_SET_STATUS64
:
1395 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1396 err
= loop_set_status64(lo
,
1397 (struct loop_info64 __user
*) arg
);
1399 case LOOP_GET_STATUS64
:
1400 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1402 case LOOP_SET_CAPACITY
:
1404 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1405 err
= loop_set_capacity(lo
, bdev
);
1407 case LOOP_SET_DIRECT_IO
:
1409 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1410 err
= loop_set_dio(lo
, arg
);
1412 case LOOP_SET_BLOCK_SIZE
:
1414 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1415 err
= loop_set_block_size(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 case LOOP_SET_BLOCK_SIZE
:
1572 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1582 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1584 struct loop_device
*lo
;
1587 mutex_lock(&loop_index_mutex
);
1588 lo
= bdev
->bd_disk
->private_data
;
1594 atomic_inc(&lo
->lo_refcnt
);
1596 mutex_unlock(&loop_index_mutex
);
1600 static void __lo_release(struct loop_device
*lo
)
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 void lo_release(struct gendisk
*disk
, fmode_t mode
)
1629 mutex_lock(&loop_index_mutex
);
1630 __lo_release(disk
->private_data
);
1631 mutex_unlock(&loop_index_mutex
);
1634 static const struct block_device_operations lo_fops
= {
1635 .owner
= THIS_MODULE
,
1637 .release
= lo_release
,
1639 #ifdef CONFIG_COMPAT
1640 .compat_ioctl
= lo_compat_ioctl
,
1645 * And now the modules code and kernel interface.
1647 static int max_loop
;
1648 module_param(max_loop
, int, S_IRUGO
);
1649 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1650 module_param(max_part
, int, S_IRUGO
);
1651 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1652 MODULE_LICENSE("GPL");
1653 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1655 int loop_register_transfer(struct loop_func_table
*funcs
)
1657 unsigned int n
= funcs
->number
;
1659 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1661 xfer_funcs
[n
] = funcs
;
1665 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1667 struct loop_device
*lo
= ptr
;
1668 struct loop_func_table
*xfer
= data
;
1670 mutex_lock(&lo
->lo_ctl_mutex
);
1671 if (lo
->lo_encryption
== xfer
)
1672 loop_release_xfer(lo
);
1673 mutex_unlock(&lo
->lo_ctl_mutex
);
1677 int loop_unregister_transfer(int number
)
1679 unsigned int n
= number
;
1680 struct loop_func_table
*xfer
;
1682 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1685 xfer_funcs
[n
] = NULL
;
1686 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1690 EXPORT_SYMBOL(loop_register_transfer
);
1691 EXPORT_SYMBOL(loop_unregister_transfer
);
1693 static int loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1694 const struct blk_mq_queue_data
*bd
)
1696 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1697 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1699 blk_mq_start_request(bd
->rq
);
1701 if (lo
->lo_state
!= Lo_bound
)
1702 return BLK_MQ_RQ_QUEUE_ERROR
;
1704 if (lo
->use_dio
&& !(cmd
->rq
->cmd_flags
& (REQ_FLUSH
|
1706 cmd
->use_aio
= true;
1708 cmd
->use_aio
= false;
1710 queue_kthread_work(&lo
->worker
, &cmd
->work
);
1712 return BLK_MQ_RQ_QUEUE_OK
;
1715 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1717 const bool write
= cmd
->rq
->cmd_flags
& REQ_WRITE
;
1718 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1721 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1726 ret
= do_req_filebacked(lo
, cmd
->rq
);
1728 /* complete non-aio request */
1729 if (!cmd
->use_aio
|| ret
)
1730 blk_mq_complete_request(cmd
->rq
, ret
? -EIO
: 0);
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(void *data
, struct request
*rq
,
1742 unsigned int hctx_idx
, unsigned int request_idx
,
1743 unsigned int numa_node
)
1745 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1748 init_kthread_work(&cmd
->work
, loop_queue_work
);
1753 static struct blk_mq_ops loop_mq_ops
= {
1754 .queue_rq
= loop_queue_rq
,
1755 .map_queue
= blk_mq_map_queue
,
1756 .init_request
= loop_init_request
,
1759 static int loop_add(struct loop_device
**l
, int i
)
1761 struct loop_device
*lo
;
1762 struct gendisk
*disk
;
1766 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1770 lo
->lo_state
= Lo_unbound
;
1772 /* allocate id, if @id >= 0, we're requesting that specific id */
1774 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1778 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1785 lo
->tag_set
.ops
= &loop_mq_ops
;
1786 lo
->tag_set
.nr_hw_queues
= 1;
1787 lo
->tag_set
.queue_depth
= 128;
1788 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1789 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1790 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1791 lo
->tag_set
.driver_data
= lo
;
1793 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1797 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1798 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1799 err
= PTR_ERR(lo
->lo_queue
);
1800 goto out_cleanup_tags
;
1802 lo
->lo_queue
->queuedata
= lo
;
1804 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
1806 * It doesn't make sense to enable merge because the I/O
1807 * submitted to backing file is handled page by page.
1809 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
1811 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1813 goto out_free_queue
;
1816 * Disable partition scanning by default. The in-kernel partition
1817 * scanning can be requested individually per-device during its
1818 * setup. Userspace can always add and remove partitions from all
1819 * devices. The needed partition minors are allocated from the
1820 * extended minor space, the main loop device numbers will continue
1821 * to match the loop minors, regardless of the number of partitions
1824 * If max_part is given, partition scanning is globally enabled for
1825 * all loop devices. The minors for the main loop devices will be
1826 * multiples of max_part.
1828 * Note: Global-for-all-devices, set-only-at-init, read-only module
1829 * parameteters like 'max_loop' and 'max_part' make things needlessly
1830 * complicated, are too static, inflexible and may surprise
1831 * userspace tools. Parameters like this in general should be avoided.
1834 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1835 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1836 mutex_init(&lo
->lo_ctl_mutex
);
1837 atomic_set(&lo
->lo_refcnt
, 0);
1839 spin_lock_init(&lo
->lo_lock
);
1840 disk
->major
= LOOP_MAJOR
;
1841 disk
->first_minor
= i
<< part_shift
;
1842 disk
->fops
= &lo_fops
;
1843 disk
->private_data
= lo
;
1844 disk
->queue
= lo
->lo_queue
;
1845 sprintf(disk
->disk_name
, "loop%d", i
);
1848 return lo
->lo_number
;
1851 blk_cleanup_queue(lo
->lo_queue
);
1853 blk_mq_free_tag_set(&lo
->tag_set
);
1855 idr_remove(&loop_index_idr
, i
);
1862 static void loop_remove(struct loop_device
*lo
)
1864 blk_cleanup_queue(lo
->lo_queue
);
1865 del_gendisk(lo
->lo_disk
);
1866 blk_mq_free_tag_set(&lo
->tag_set
);
1867 put_disk(lo
->lo_disk
);
1871 static int find_free_cb(int id
, void *ptr
, void *data
)
1873 struct loop_device
*lo
= ptr
;
1874 struct loop_device
**l
= data
;
1876 if (lo
->lo_state
== Lo_unbound
) {
1883 static int loop_lookup(struct loop_device
**l
, int i
)
1885 struct loop_device
*lo
;
1891 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1894 ret
= lo
->lo_number
;
1899 /* lookup and return a specific i */
1900 lo
= idr_find(&loop_index_idr
, i
);
1903 ret
= lo
->lo_number
;
1909 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1911 struct loop_device
*lo
;
1912 struct kobject
*kobj
;
1915 mutex_lock(&loop_index_mutex
);
1916 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1918 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1922 kobj
= get_disk(lo
->lo_disk
);
1923 mutex_unlock(&loop_index_mutex
);
1929 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1932 struct loop_device
*lo
;
1935 mutex_lock(&loop_index_mutex
);
1938 ret
= loop_lookup(&lo
, parm
);
1943 ret
= loop_add(&lo
, parm
);
1945 case LOOP_CTL_REMOVE
:
1946 ret
= loop_lookup(&lo
, parm
);
1949 mutex_lock(&lo
->lo_ctl_mutex
);
1950 if (lo
->lo_state
!= Lo_unbound
) {
1952 mutex_unlock(&lo
->lo_ctl_mutex
);
1955 if (atomic_read(&lo
->lo_refcnt
) > 0) {
1957 mutex_unlock(&lo
->lo_ctl_mutex
);
1960 lo
->lo_disk
->private_data
= NULL
;
1961 mutex_unlock(&lo
->lo_ctl_mutex
);
1962 idr_remove(&loop_index_idr
, lo
->lo_number
);
1965 case LOOP_CTL_GET_FREE
:
1966 ret
= loop_lookup(&lo
, -1);
1969 ret
= loop_add(&lo
, -1);
1971 mutex_unlock(&loop_index_mutex
);
1976 static const struct file_operations loop_ctl_fops
= {
1977 .open
= nonseekable_open
,
1978 .unlocked_ioctl
= loop_control_ioctl
,
1979 .compat_ioctl
= loop_control_ioctl
,
1980 .owner
= THIS_MODULE
,
1981 .llseek
= noop_llseek
,
1984 static struct miscdevice loop_misc
= {
1985 .minor
= LOOP_CTRL_MINOR
,
1986 .name
= "loop-control",
1987 .fops
= &loop_ctl_fops
,
1990 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1991 MODULE_ALIAS("devname:loop-control");
1993 static int __init
loop_init(void)
1996 unsigned long range
;
1997 struct loop_device
*lo
;
2000 err
= misc_register(&loop_misc
);
2006 part_shift
= fls(max_part
);
2009 * Adjust max_part according to part_shift as it is exported
2010 * to user space so that user can decide correct minor number
2011 * if [s]he want to create more devices.
2013 * Note that -1 is required because partition 0 is reserved
2014 * for the whole disk.
2016 max_part
= (1UL << part_shift
) - 1;
2019 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2024 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2030 * If max_loop is specified, create that many devices upfront.
2031 * This also becomes a hard limit. If max_loop is not specified,
2032 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2033 * init time. Loop devices can be requested on-demand with the
2034 * /dev/loop-control interface, or be instantiated by accessing
2035 * a 'dead' device node.
2039 range
= max_loop
<< part_shift
;
2041 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
2042 range
= 1UL << MINORBITS
;
2045 if (register_blkdev(LOOP_MAJOR
, "loop")) {
2050 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
2051 THIS_MODULE
, loop_probe
, NULL
, NULL
);
2053 /* pre-create number of devices given by config or max_loop */
2054 mutex_lock(&loop_index_mutex
);
2055 for (i
= 0; i
< nr
; i
++)
2057 mutex_unlock(&loop_index_mutex
);
2059 printk(KERN_INFO
"loop: module loaded\n");
2063 misc_deregister(&loop_misc
);
2067 static int loop_exit_cb(int id
, void *ptr
, void *data
)
2069 struct loop_device
*lo
= ptr
;
2075 static void __exit
loop_exit(void)
2077 unsigned long range
;
2079 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
2081 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
2082 idr_destroy(&loop_index_idr
);
2084 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
2085 unregister_blkdev(LOOP_MAJOR
, "loop");
2087 misc_deregister(&loop_misc
);
2090 module_init(loop_init
);
2091 module_exit(loop_exit
);
2094 static int __init
max_loop_setup(char *str
)
2096 max_loop
= simple_strtol(str
, NULL
, 0);
2100 __setup("max_loop=", max_loop_setup
);