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/blk-mq.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 struct workqueue_struct
*loop_wq
;
94 static int transfer_none(struct loop_device
*lo
, int cmd
,
95 struct page
*raw_page
, unsigned raw_off
,
96 struct page
*loop_page
, unsigned loop_off
,
97 int size
, sector_t real_block
)
99 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
100 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
103 memcpy(loop_buf
, raw_buf
, size
);
105 memcpy(raw_buf
, loop_buf
, size
);
107 kunmap_atomic(loop_buf
);
108 kunmap_atomic(raw_buf
);
113 static int transfer_xor(struct loop_device
*lo
, int cmd
,
114 struct page
*raw_page
, unsigned raw_off
,
115 struct page
*loop_page
, unsigned loop_off
,
116 int size
, sector_t real_block
)
118 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
119 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
120 char *in
, *out
, *key
;
131 key
= lo
->lo_encrypt_key
;
132 keysize
= lo
->lo_encrypt_key_size
;
133 for (i
= 0; i
< size
; i
++)
134 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
136 kunmap_atomic(loop_buf
);
137 kunmap_atomic(raw_buf
);
142 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
144 if (unlikely(info
->lo_encrypt_key_size
<= 0))
149 static struct loop_func_table none_funcs
= {
150 .number
= LO_CRYPT_NONE
,
151 .transfer
= transfer_none
,
154 static struct loop_func_table xor_funcs
= {
155 .number
= LO_CRYPT_XOR
,
156 .transfer
= transfer_xor
,
160 /* xfer_funcs[0] is special - its release function is never called */
161 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
166 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
170 /* Compute loopsize in bytes */
171 loopsize
= i_size_read(file
->f_mapping
->host
);
174 /* offset is beyond i_size, weird but possible */
178 if (sizelimit
> 0 && sizelimit
< loopsize
)
179 loopsize
= sizelimit
;
181 * Unfortunately, if we want to do I/O on the device,
182 * the number of 512-byte sectors has to fit into a sector_t.
184 return loopsize
>> 9;
187 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
189 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
193 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
195 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
196 sector_t x
= (sector_t
)size
;
197 struct block_device
*bdev
= lo
->lo_device
;
199 if (unlikely((loff_t
)x
!= size
))
201 if (lo
->lo_offset
!= offset
)
202 lo
->lo_offset
= offset
;
203 if (lo
->lo_sizelimit
!= sizelimit
)
204 lo
->lo_sizelimit
= sizelimit
;
205 set_capacity(lo
->lo_disk
, x
);
206 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
207 /* let user-space know about the new size */
208 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
213 lo_do_transfer(struct loop_device
*lo
, int cmd
,
214 struct page
*rpage
, unsigned roffs
,
215 struct page
*lpage
, unsigned loffs
,
216 int size
, sector_t rblock
)
218 if (unlikely(!lo
->transfer
))
221 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
225 * __do_lo_send_write - helper for writing data to a loop device
227 * This helper just factors out common code between do_lo_send_direct_write()
228 * and do_lo_send_write().
230 static int __do_lo_send_write(struct file
*file
,
231 u8
*buf
, const int len
, loff_t pos
)
234 mm_segment_t old_fs
= get_fs();
236 file_start_write(file
);
238 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
240 file_end_write(file
);
241 if (likely(bw
== len
))
243 printk_ratelimited(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
244 (unsigned long long)pos
, len
);
251 * do_lo_send_direct_write - helper for writing data to a loop device
253 * This is the fast, non-transforming version that does not need double
256 static int do_lo_send_direct_write(struct loop_device
*lo
,
257 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
259 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
260 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
262 kunmap(bvec
->bv_page
);
268 * do_lo_send_write - helper for writing data to a loop device
270 * This is the slow, transforming version that needs to double buffer the
271 * data as it cannot do the transformations in place without having direct
272 * access to the destination pages of the backing file.
274 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
275 loff_t pos
, struct page
*page
)
277 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
278 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
280 return __do_lo_send_write(lo
->lo_backing_file
,
281 page_address(page
), bvec
->bv_len
,
283 printk_ratelimited(KERN_ERR
"loop: Transfer error at byte offset %llu, "
284 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
290 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
292 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
295 struct bvec_iter iter
;
296 struct page
*page
= NULL
;
299 if (lo
->transfer
!= transfer_none
) {
300 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
304 do_lo_send
= do_lo_send_write
;
306 do_lo_send
= do_lo_send_direct_write
;
309 bio_for_each_segment(bvec
, bio
, iter
) {
310 ret
= do_lo_send(lo
, &bvec
, pos
, page
);
322 printk_ratelimited(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
327 struct lo_read_data
{
328 struct loop_device
*lo
;
335 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
336 struct splice_desc
*sd
)
338 struct lo_read_data
*p
= sd
->u
.data
;
339 struct loop_device
*lo
= p
->lo
;
340 struct page
*page
= buf
->page
;
344 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
350 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
351 printk_ratelimited(KERN_ERR
"loop: transfer error block %ld\n",
356 flush_dcache_page(p
->page
);
365 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
367 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
371 do_lo_receive(struct loop_device
*lo
,
372 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
374 struct lo_read_data cookie
;
375 struct splice_desc sd
;
380 cookie
.page
= bvec
->bv_page
;
381 cookie
.offset
= bvec
->bv_offset
;
382 cookie
.bsize
= bsize
;
385 sd
.total_len
= bvec
->bv_len
;
390 file
= lo
->lo_backing_file
;
391 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
397 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
400 struct bvec_iter iter
;
403 bio_for_each_segment(bvec
, bio
, iter
) {
404 s
= do_lo_receive(lo
, &bvec
, bsize
, pos
);
408 if (s
!= bvec
.bv_len
) {
417 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
422 pos
= ((loff_t
) bio
->bi_iter
.bi_sector
<< 9) + lo
->lo_offset
;
424 if (bio_rw(bio
) == WRITE
) {
425 struct file
*file
= lo
->lo_backing_file
;
427 if (bio
->bi_rw
& REQ_FLUSH
) {
428 ret
= vfs_fsync(file
, 0);
429 if (unlikely(ret
&& ret
!= -EINVAL
)) {
436 * We use punch hole to reclaim the free space used by the
437 * image a.k.a. discard. However we do not support discard if
438 * encryption is enabled, because it may give an attacker
439 * useful information.
441 if (bio
->bi_rw
& REQ_DISCARD
) {
442 struct file
*file
= lo
->lo_backing_file
;
443 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
445 if ((!file
->f_op
->fallocate
) ||
446 lo
->lo_encrypt_key_size
) {
450 ret
= file
->f_op
->fallocate(file
, mode
, pos
,
451 bio
->bi_iter
.bi_size
);
452 if (unlikely(ret
&& ret
!= -EINVAL
&&
458 ret
= lo_send(lo
, bio
, pos
);
460 if ((bio
->bi_rw
& REQ_FUA
) && !ret
) {
461 ret
= vfs_fsync(file
, 0);
462 if (unlikely(ret
&& ret
!= -EINVAL
))
466 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
472 struct switch_request
{
474 struct completion wait
;
477 static inline int loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
479 return do_bio_filebacked(lo
, bio
);
483 * Do the actual switch; called from the BIO completion routine
485 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
487 struct file
*file
= p
->file
;
488 struct file
*old_file
= lo
->lo_backing_file
;
489 struct address_space
*mapping
;
491 /* if no new file, only flush of queued bios requested */
495 mapping
= file
->f_mapping
;
496 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
497 lo
->lo_backing_file
= file
;
498 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
499 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
500 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
501 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
505 * loop_switch performs the hard work of switching a backing store.
506 * First it needs to flush existing IO, it does this by sending a magic
507 * BIO down the pipe. The completion of this BIO does the actual switch.
509 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
511 struct switch_request w
;
515 /* freeze queue and wait for completion of scheduled requests */
516 blk_mq_freeze_queue(lo
->lo_queue
);
518 /* do the switch action */
519 do_loop_switch(lo
, &w
);
522 blk_mq_unfreeze_queue(lo
->lo_queue
);
528 * Helper to flush the IOs in loop, but keeping loop thread running
530 static int loop_flush(struct loop_device
*lo
)
532 return loop_switch(lo
, NULL
);
536 * loop_change_fd switched the backing store of a loopback device to
537 * a new file. This is useful for operating system installers to free up
538 * the original file and in High Availability environments to switch to
539 * an alternative location for the content in case of server meltdown.
540 * This can only work if the loop device is used read-only, and if the
541 * new backing store is the same size and type as the old backing store.
543 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
546 struct file
*file
, *old_file
;
551 if (lo
->lo_state
!= Lo_bound
)
554 /* the loop device has to be read-only */
556 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
564 inode
= file
->f_mapping
->host
;
565 old_file
= lo
->lo_backing_file
;
569 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
572 /* size of the new backing store needs to be the same */
573 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
577 error
= loop_switch(lo
, file
);
582 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
583 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
592 static inline int is_loop_device(struct file
*file
)
594 struct inode
*i
= file
->f_mapping
->host
;
596 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
599 /* loop sysfs attributes */
601 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
602 ssize_t (*callback
)(struct loop_device
*, char *))
604 struct gendisk
*disk
= dev_to_disk(dev
);
605 struct loop_device
*lo
= disk
->private_data
;
607 return callback(lo
, page
);
610 #define LOOP_ATTR_RO(_name) \
611 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
612 static ssize_t loop_attr_do_show_##_name(struct device *d, \
613 struct device_attribute *attr, char *b) \
615 return loop_attr_show(d, b, loop_attr_##_name##_show); \
617 static struct device_attribute loop_attr_##_name = \
618 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
620 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
625 spin_lock_irq(&lo
->lo_lock
);
626 if (lo
->lo_backing_file
)
627 p
= d_path(&lo
->lo_backing_file
->f_path
, buf
, PAGE_SIZE
- 1);
628 spin_unlock_irq(&lo
->lo_lock
);
630 if (IS_ERR_OR_NULL(p
))
634 memmove(buf
, p
, ret
);
642 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
644 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
647 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
649 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
652 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
654 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
656 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
659 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
661 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
663 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
666 LOOP_ATTR_RO(backing_file
);
667 LOOP_ATTR_RO(offset
);
668 LOOP_ATTR_RO(sizelimit
);
669 LOOP_ATTR_RO(autoclear
);
670 LOOP_ATTR_RO(partscan
);
672 static struct attribute
*loop_attrs
[] = {
673 &loop_attr_backing_file
.attr
,
674 &loop_attr_offset
.attr
,
675 &loop_attr_sizelimit
.attr
,
676 &loop_attr_autoclear
.attr
,
677 &loop_attr_partscan
.attr
,
681 static struct attribute_group loop_attribute_group
= {
686 static int loop_sysfs_init(struct loop_device
*lo
)
688 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
689 &loop_attribute_group
);
692 static void loop_sysfs_exit(struct loop_device
*lo
)
694 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
695 &loop_attribute_group
);
698 static void loop_config_discard(struct loop_device
*lo
)
700 struct file
*file
= lo
->lo_backing_file
;
701 struct inode
*inode
= file
->f_mapping
->host
;
702 struct request_queue
*q
= lo
->lo_queue
;
705 * We use punch hole to reclaim the free space used by the
706 * image a.k.a. discard. However we do not support discard if
707 * encryption is enabled, because it may give an attacker
708 * useful information.
710 if ((!file
->f_op
->fallocate
) ||
711 lo
->lo_encrypt_key_size
) {
712 q
->limits
.discard_granularity
= 0;
713 q
->limits
.discard_alignment
= 0;
714 q
->limits
.max_discard_sectors
= 0;
715 q
->limits
.discard_zeroes_data
= 0;
716 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
720 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
721 q
->limits
.discard_alignment
= 0;
722 q
->limits
.max_discard_sectors
= UINT_MAX
>> 9;
723 q
->limits
.discard_zeroes_data
= 1;
724 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
727 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
728 struct block_device
*bdev
, unsigned int arg
)
730 struct file
*file
, *f
;
732 struct address_space
*mapping
;
733 unsigned lo_blocksize
;
738 /* This is safe, since we have a reference from open(). */
739 __module_get(THIS_MODULE
);
747 if (lo
->lo_state
!= Lo_unbound
)
750 /* Avoid recursion */
752 while (is_loop_device(f
)) {
753 struct loop_device
*l
;
755 if (f
->f_mapping
->host
->i_bdev
== bdev
)
758 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
759 if (l
->lo_state
== Lo_unbound
) {
763 f
= l
->lo_backing_file
;
766 mapping
= file
->f_mapping
;
767 inode
= mapping
->host
;
770 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
773 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
775 lo_flags
|= LO_FLAGS_READ_ONLY
;
777 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
778 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
781 size
= get_loop_size(lo
, file
);
782 if ((loff_t
)(sector_t
)size
!= size
)
787 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
789 lo
->lo_blocksize
= lo_blocksize
;
790 lo
->lo_device
= bdev
;
791 lo
->lo_flags
= lo_flags
;
792 lo
->lo_backing_file
= file
;
793 lo
->transfer
= transfer_none
;
795 lo
->lo_sizelimit
= 0;
796 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
797 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
799 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
800 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
802 set_capacity(lo
->lo_disk
, size
);
803 bd_set_size(bdev
, size
<< 9);
805 /* let user-space know about the new size */
806 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
808 set_blocksize(bdev
, lo_blocksize
);
810 lo
->lo_state
= Lo_bound
;
812 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
813 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
814 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
816 /* Grab the block_device to prevent its destruction after we
817 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
825 /* This is safe: open() is still holding a reference. */
826 module_put(THIS_MODULE
);
831 loop_release_xfer(struct loop_device
*lo
)
834 struct loop_func_table
*xfer
= lo
->lo_encryption
;
838 err
= xfer
->release(lo
);
840 lo
->lo_encryption
= NULL
;
841 module_put(xfer
->owner
);
847 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
848 const struct loop_info64
*i
)
853 struct module
*owner
= xfer
->owner
;
855 if (!try_module_get(owner
))
858 err
= xfer
->init(lo
, i
);
862 lo
->lo_encryption
= xfer
;
867 static int loop_clr_fd(struct loop_device
*lo
)
869 struct file
*filp
= lo
->lo_backing_file
;
870 gfp_t gfp
= lo
->old_gfp_mask
;
871 struct block_device
*bdev
= lo
->lo_device
;
873 if (lo
->lo_state
!= Lo_bound
)
877 * If we've explicitly asked to tear down the loop device,
878 * and it has an elevated reference count, set it for auto-teardown when
879 * the last reference goes away. This stops $!~#$@ udev from
880 * preventing teardown because it decided that it needs to run blkid on
881 * the loopback device whenever they appear. xfstests is notorious for
882 * failing tests because blkid via udev races with a losetup
883 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
884 * command to fail with EBUSY.
886 if (lo
->lo_refcnt
> 1) {
887 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
888 mutex_unlock(&lo
->lo_ctl_mutex
);
895 spin_lock_irq(&lo
->lo_lock
);
896 lo
->lo_state
= Lo_rundown
;
897 lo
->lo_backing_file
= NULL
;
898 spin_unlock_irq(&lo
->lo_lock
);
900 loop_release_xfer(lo
);
903 lo
->lo_device
= NULL
;
904 lo
->lo_encryption
= NULL
;
906 lo
->lo_sizelimit
= 0;
907 lo
->lo_encrypt_key_size
= 0;
908 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
909 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
910 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
913 invalidate_bdev(bdev
);
915 set_capacity(lo
->lo_disk
, 0);
918 bd_set_size(bdev
, 0);
919 /* let user-space know about this change */
920 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
922 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
923 lo
->lo_state
= Lo_unbound
;
924 /* This is safe: open() is still holding a reference. */
925 module_put(THIS_MODULE
);
926 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
927 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
930 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
931 mutex_unlock(&lo
->lo_ctl_mutex
);
933 * Need not hold lo_ctl_mutex to fput backing file.
934 * Calling fput holding lo_ctl_mutex triggers a circular
935 * lock dependency possibility warning as fput can take
936 * bd_mutex which is usually taken before lo_ctl_mutex.
943 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
946 struct loop_func_table
*xfer
;
947 kuid_t uid
= current_uid();
949 if (lo
->lo_encrypt_key_size
&&
950 !uid_eq(lo
->lo_key_owner
, uid
) &&
951 !capable(CAP_SYS_ADMIN
))
953 if (lo
->lo_state
!= Lo_bound
)
955 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
958 err
= loop_release_xfer(lo
);
962 if (info
->lo_encrypt_type
) {
963 unsigned int type
= info
->lo_encrypt_type
;
965 if (type
>= MAX_LO_CRYPT
)
967 xfer
= xfer_funcs
[type
];
973 err
= loop_init_xfer(lo
, xfer
, info
);
977 if (lo
->lo_offset
!= info
->lo_offset
||
978 lo
->lo_sizelimit
!= info
->lo_sizelimit
)
979 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
))
982 loop_config_discard(lo
);
984 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
985 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
986 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
987 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
991 lo
->transfer
= xfer
->transfer
;
992 lo
->ioctl
= xfer
->ioctl
;
994 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
995 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
996 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
998 if ((info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
999 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1000 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1001 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1002 ioctl_by_bdev(lo
->lo_device
, BLKRRPART
, 0);
1005 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1006 lo
->lo_init
[0] = info
->lo_init
[0];
1007 lo
->lo_init
[1] = info
->lo_init
[1];
1008 if (info
->lo_encrypt_key_size
) {
1009 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1010 info
->lo_encrypt_key_size
);
1011 lo
->lo_key_owner
= uid
;
1018 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1020 struct file
*file
= lo
->lo_backing_file
;
1024 if (lo
->lo_state
!= Lo_bound
)
1026 error
= vfs_getattr(&file
->f_path
, &stat
);
1029 memset(info
, 0, sizeof(*info
));
1030 info
->lo_number
= lo
->lo_number
;
1031 info
->lo_device
= huge_encode_dev(stat
.dev
);
1032 info
->lo_inode
= stat
.ino
;
1033 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1034 info
->lo_offset
= lo
->lo_offset
;
1035 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1036 info
->lo_flags
= lo
->lo_flags
;
1037 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1038 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1039 info
->lo_encrypt_type
=
1040 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1041 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1042 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1043 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1044 lo
->lo_encrypt_key_size
);
1050 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1052 memset(info64
, 0, sizeof(*info64
));
1053 info64
->lo_number
= info
->lo_number
;
1054 info64
->lo_device
= info
->lo_device
;
1055 info64
->lo_inode
= info
->lo_inode
;
1056 info64
->lo_rdevice
= info
->lo_rdevice
;
1057 info64
->lo_offset
= info
->lo_offset
;
1058 info64
->lo_sizelimit
= 0;
1059 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1060 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1061 info64
->lo_flags
= info
->lo_flags
;
1062 info64
->lo_init
[0] = info
->lo_init
[0];
1063 info64
->lo_init
[1] = info
->lo_init
[1];
1064 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1065 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1067 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1068 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1072 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1074 memset(info
, 0, sizeof(*info
));
1075 info
->lo_number
= info64
->lo_number
;
1076 info
->lo_device
= info64
->lo_device
;
1077 info
->lo_inode
= info64
->lo_inode
;
1078 info
->lo_rdevice
= info64
->lo_rdevice
;
1079 info
->lo_offset
= info64
->lo_offset
;
1080 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1081 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1082 info
->lo_flags
= info64
->lo_flags
;
1083 info
->lo_init
[0] = info64
->lo_init
[0];
1084 info
->lo_init
[1] = info64
->lo_init
[1];
1085 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1086 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1088 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1089 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1091 /* error in case values were truncated */
1092 if (info
->lo_device
!= info64
->lo_device
||
1093 info
->lo_rdevice
!= info64
->lo_rdevice
||
1094 info
->lo_inode
!= info64
->lo_inode
||
1095 info
->lo_offset
!= info64
->lo_offset
)
1102 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1104 struct loop_info info
;
1105 struct loop_info64 info64
;
1107 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1109 loop_info64_from_old(&info
, &info64
);
1110 return loop_set_status(lo
, &info64
);
1114 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1116 struct loop_info64 info64
;
1118 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1120 return loop_set_status(lo
, &info64
);
1124 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1125 struct loop_info info
;
1126 struct loop_info64 info64
;
1132 err
= loop_get_status(lo
, &info64
);
1134 err
= loop_info64_to_old(&info64
, &info
);
1135 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1142 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1143 struct loop_info64 info64
;
1149 err
= loop_get_status(lo
, &info64
);
1150 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1156 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1158 if (unlikely(lo
->lo_state
!= Lo_bound
))
1161 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1164 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1165 unsigned int cmd
, unsigned long arg
)
1167 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1170 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1173 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1175 case LOOP_CHANGE_FD
:
1176 err
= loop_change_fd(lo
, bdev
, arg
);
1179 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1180 err
= loop_clr_fd(lo
);
1184 case LOOP_SET_STATUS
:
1186 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1187 err
= loop_set_status_old(lo
,
1188 (struct loop_info __user
*)arg
);
1190 case LOOP_GET_STATUS
:
1191 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1193 case LOOP_SET_STATUS64
:
1195 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1196 err
= loop_set_status64(lo
,
1197 (struct loop_info64 __user
*) arg
);
1199 case LOOP_GET_STATUS64
:
1200 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1202 case LOOP_SET_CAPACITY
:
1204 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1205 err
= loop_set_capacity(lo
, bdev
);
1208 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1210 mutex_unlock(&lo
->lo_ctl_mutex
);
1216 #ifdef CONFIG_COMPAT
1217 struct compat_loop_info
{
1218 compat_int_t lo_number
; /* ioctl r/o */
1219 compat_dev_t lo_device
; /* ioctl r/o */
1220 compat_ulong_t lo_inode
; /* ioctl r/o */
1221 compat_dev_t lo_rdevice
; /* ioctl r/o */
1222 compat_int_t lo_offset
;
1223 compat_int_t lo_encrypt_type
;
1224 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1225 compat_int_t lo_flags
; /* ioctl r/o */
1226 char lo_name
[LO_NAME_SIZE
];
1227 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1228 compat_ulong_t lo_init
[2];
1233 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1234 * - noinlined to reduce stack space usage in main part of driver
1237 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1238 struct loop_info64
*info64
)
1240 struct compat_loop_info info
;
1242 if (copy_from_user(&info
, arg
, sizeof(info
)))
1245 memset(info64
, 0, sizeof(*info64
));
1246 info64
->lo_number
= info
.lo_number
;
1247 info64
->lo_device
= info
.lo_device
;
1248 info64
->lo_inode
= info
.lo_inode
;
1249 info64
->lo_rdevice
= info
.lo_rdevice
;
1250 info64
->lo_offset
= info
.lo_offset
;
1251 info64
->lo_sizelimit
= 0;
1252 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1253 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1254 info64
->lo_flags
= info
.lo_flags
;
1255 info64
->lo_init
[0] = info
.lo_init
[0];
1256 info64
->lo_init
[1] = info
.lo_init
[1];
1257 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1258 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1260 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1261 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1266 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1267 * - noinlined to reduce stack space usage in main part of driver
1270 loop_info64_to_compat(const struct loop_info64
*info64
,
1271 struct compat_loop_info __user
*arg
)
1273 struct compat_loop_info info
;
1275 memset(&info
, 0, sizeof(info
));
1276 info
.lo_number
= info64
->lo_number
;
1277 info
.lo_device
= info64
->lo_device
;
1278 info
.lo_inode
= info64
->lo_inode
;
1279 info
.lo_rdevice
= info64
->lo_rdevice
;
1280 info
.lo_offset
= info64
->lo_offset
;
1281 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1282 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1283 info
.lo_flags
= info64
->lo_flags
;
1284 info
.lo_init
[0] = info64
->lo_init
[0];
1285 info
.lo_init
[1] = info64
->lo_init
[1];
1286 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1287 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1289 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1290 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1292 /* error in case values were truncated */
1293 if (info
.lo_device
!= info64
->lo_device
||
1294 info
.lo_rdevice
!= info64
->lo_rdevice
||
1295 info
.lo_inode
!= info64
->lo_inode
||
1296 info
.lo_offset
!= info64
->lo_offset
||
1297 info
.lo_init
[0] != info64
->lo_init
[0] ||
1298 info
.lo_init
[1] != info64
->lo_init
[1])
1301 if (copy_to_user(arg
, &info
, sizeof(info
)))
1307 loop_set_status_compat(struct loop_device
*lo
,
1308 const struct compat_loop_info __user
*arg
)
1310 struct loop_info64 info64
;
1313 ret
= loop_info64_from_compat(arg
, &info64
);
1316 return loop_set_status(lo
, &info64
);
1320 loop_get_status_compat(struct loop_device
*lo
,
1321 struct compat_loop_info __user
*arg
)
1323 struct loop_info64 info64
;
1329 err
= loop_get_status(lo
, &info64
);
1331 err
= loop_info64_to_compat(&info64
, arg
);
1335 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1336 unsigned int cmd
, unsigned long arg
)
1338 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1342 case LOOP_SET_STATUS
:
1343 mutex_lock(&lo
->lo_ctl_mutex
);
1344 err
= loop_set_status_compat(
1345 lo
, (const struct compat_loop_info __user
*) arg
);
1346 mutex_unlock(&lo
->lo_ctl_mutex
);
1348 case LOOP_GET_STATUS
:
1349 mutex_lock(&lo
->lo_ctl_mutex
);
1350 err
= loop_get_status_compat(
1351 lo
, (struct compat_loop_info __user
*) arg
);
1352 mutex_unlock(&lo
->lo_ctl_mutex
);
1354 case LOOP_SET_CAPACITY
:
1356 case LOOP_GET_STATUS64
:
1357 case LOOP_SET_STATUS64
:
1358 arg
= (unsigned long) compat_ptr(arg
);
1360 case LOOP_CHANGE_FD
:
1361 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1371 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1373 struct loop_device
*lo
;
1376 mutex_lock(&loop_index_mutex
);
1377 lo
= bdev
->bd_disk
->private_data
;
1383 mutex_lock(&lo
->lo_ctl_mutex
);
1385 mutex_unlock(&lo
->lo_ctl_mutex
);
1387 mutex_unlock(&loop_index_mutex
);
1391 static void lo_release(struct gendisk
*disk
, fmode_t mode
)
1393 struct loop_device
*lo
= disk
->private_data
;
1396 mutex_lock(&lo
->lo_ctl_mutex
);
1398 if (--lo
->lo_refcnt
)
1401 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1403 * In autoclear mode, stop the loop thread
1404 * and remove configuration after last close.
1406 err
= loop_clr_fd(lo
);
1411 * Otherwise keep thread (if running) and config,
1412 * but flush possible ongoing bios in thread.
1418 mutex_unlock(&lo
->lo_ctl_mutex
);
1421 static const struct block_device_operations lo_fops
= {
1422 .owner
= THIS_MODULE
,
1424 .release
= lo_release
,
1426 #ifdef CONFIG_COMPAT
1427 .compat_ioctl
= lo_compat_ioctl
,
1432 * And now the modules code and kernel interface.
1434 static int max_loop
;
1435 module_param(max_loop
, int, S_IRUGO
);
1436 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1437 module_param(max_part
, int, S_IRUGO
);
1438 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1439 MODULE_LICENSE("GPL");
1440 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1442 int loop_register_transfer(struct loop_func_table
*funcs
)
1444 unsigned int n
= funcs
->number
;
1446 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1448 xfer_funcs
[n
] = funcs
;
1452 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1454 struct loop_device
*lo
= ptr
;
1455 struct loop_func_table
*xfer
= data
;
1457 mutex_lock(&lo
->lo_ctl_mutex
);
1458 if (lo
->lo_encryption
== xfer
)
1459 loop_release_xfer(lo
);
1460 mutex_unlock(&lo
->lo_ctl_mutex
);
1464 int loop_unregister_transfer(int number
)
1466 unsigned int n
= number
;
1467 struct loop_func_table
*xfer
;
1469 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1472 xfer_funcs
[n
] = NULL
;
1473 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1477 EXPORT_SYMBOL(loop_register_transfer
);
1478 EXPORT_SYMBOL(loop_unregister_transfer
);
1480 static int loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1481 const struct blk_mq_queue_data
*bd
)
1483 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1485 blk_mq_start_request(bd
->rq
);
1487 if (cmd
->rq
->cmd_flags
& REQ_WRITE
) {
1488 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1489 bool need_sched
= true;
1491 spin_lock_irq(&lo
->lo_lock
);
1492 if (lo
->write_started
)
1495 lo
->write_started
= true;
1496 list_add_tail(&cmd
->list
, &lo
->write_cmd_head
);
1497 spin_unlock_irq(&lo
->lo_lock
);
1500 queue_work(loop_wq
, &lo
->write_work
);
1502 queue_work(loop_wq
, &cmd
->read_work
);
1505 return BLK_MQ_RQ_QUEUE_OK
;
1508 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1510 const bool write
= cmd
->rq
->cmd_flags
& REQ_WRITE
;
1511 struct loop_device
*lo
= cmd
->rq
->q
->queuedata
;
1515 if (lo
->lo_state
!= Lo_bound
)
1518 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
1522 __rq_for_each_bio(bio
, cmd
->rq
)
1523 ret
|= loop_handle_bio(lo
, bio
);
1527 cmd
->rq
->errors
= -EIO
;
1528 blk_mq_complete_request(cmd
->rq
);
1531 static void loop_queue_write_work(struct work_struct
*work
)
1533 struct loop_device
*lo
=
1534 container_of(work
, struct loop_device
, write_work
);
1535 LIST_HEAD(cmd_list
);
1537 spin_lock_irq(&lo
->lo_lock
);
1539 list_splice_init(&lo
->write_cmd_head
, &cmd_list
);
1540 spin_unlock_irq(&lo
->lo_lock
);
1542 while (!list_empty(&cmd_list
)) {
1543 struct loop_cmd
*cmd
= list_first_entry(&cmd_list
,
1544 struct loop_cmd
, list
);
1545 list_del_init(&cmd
->list
);
1546 loop_handle_cmd(cmd
);
1549 spin_lock_irq(&lo
->lo_lock
);
1550 if (!list_empty(&lo
->write_cmd_head
))
1552 lo
->write_started
= false;
1553 spin_unlock_irq(&lo
->lo_lock
);
1556 static void loop_queue_read_work(struct work_struct
*work
)
1558 struct loop_cmd
*cmd
=
1559 container_of(work
, struct loop_cmd
, read_work
);
1561 loop_handle_cmd(cmd
);
1564 static int loop_init_request(void *data
, struct request
*rq
,
1565 unsigned int hctx_idx
, unsigned int request_idx
,
1566 unsigned int numa_node
)
1568 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1571 INIT_WORK(&cmd
->read_work
, loop_queue_read_work
);
1576 static struct blk_mq_ops loop_mq_ops
= {
1577 .queue_rq
= loop_queue_rq
,
1578 .map_queue
= blk_mq_map_queue
,
1579 .init_request
= loop_init_request
,
1582 static int loop_add(struct loop_device
**l
, int i
)
1584 struct loop_device
*lo
;
1585 struct gendisk
*disk
;
1589 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1593 lo
->lo_state
= Lo_unbound
;
1595 /* allocate id, if @id >= 0, we're requesting that specific id */
1597 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1601 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1608 lo
->tag_set
.ops
= &loop_mq_ops
;
1609 lo
->tag_set
.nr_hw_queues
= 1;
1610 lo
->tag_set
.queue_depth
= 128;
1611 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
1612 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
1613 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
1614 lo
->tag_set
.driver_data
= lo
;
1616 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
1620 lo
->lo_queue
= blk_mq_init_queue(&lo
->tag_set
);
1621 if (IS_ERR_OR_NULL(lo
->lo_queue
)) {
1622 err
= PTR_ERR(lo
->lo_queue
);
1623 goto out_cleanup_tags
;
1625 lo
->lo_queue
->queuedata
= lo
;
1627 INIT_LIST_HEAD(&lo
->write_cmd_head
);
1628 INIT_WORK(&lo
->write_work
, loop_queue_write_work
);
1630 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1632 goto out_free_queue
;
1635 * Disable partition scanning by default. The in-kernel partition
1636 * scanning can be requested individually per-device during its
1637 * setup. Userspace can always add and remove partitions from all
1638 * devices. The needed partition minors are allocated from the
1639 * extended minor space, the main loop device numbers will continue
1640 * to match the loop minors, regardless of the number of partitions
1643 * If max_part is given, partition scanning is globally enabled for
1644 * all loop devices. The minors for the main loop devices will be
1645 * multiples of max_part.
1647 * Note: Global-for-all-devices, set-only-at-init, read-only module
1648 * parameteters like 'max_loop' and 'max_part' make things needlessly
1649 * complicated, are too static, inflexible and may surprise
1650 * userspace tools. Parameters like this in general should be avoided.
1653 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1654 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1655 mutex_init(&lo
->lo_ctl_mutex
);
1657 spin_lock_init(&lo
->lo_lock
);
1658 disk
->major
= LOOP_MAJOR
;
1659 disk
->first_minor
= i
<< part_shift
;
1660 disk
->fops
= &lo_fops
;
1661 disk
->private_data
= lo
;
1662 disk
->queue
= lo
->lo_queue
;
1663 sprintf(disk
->disk_name
, "loop%d", i
);
1666 return lo
->lo_number
;
1669 blk_cleanup_queue(lo
->lo_queue
);
1671 blk_mq_free_tag_set(&lo
->tag_set
);
1673 idr_remove(&loop_index_idr
, i
);
1680 static void loop_remove(struct loop_device
*lo
)
1682 del_gendisk(lo
->lo_disk
);
1683 blk_cleanup_queue(lo
->lo_queue
);
1684 blk_mq_free_tag_set(&lo
->tag_set
);
1685 put_disk(lo
->lo_disk
);
1689 static int find_free_cb(int id
, void *ptr
, void *data
)
1691 struct loop_device
*lo
= ptr
;
1692 struct loop_device
**l
= data
;
1694 if (lo
->lo_state
== Lo_unbound
) {
1701 static int loop_lookup(struct loop_device
**l
, int i
)
1703 struct loop_device
*lo
;
1709 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1712 ret
= lo
->lo_number
;
1717 /* lookup and return a specific i */
1718 lo
= idr_find(&loop_index_idr
, i
);
1721 ret
= lo
->lo_number
;
1727 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1729 struct loop_device
*lo
;
1730 struct kobject
*kobj
;
1733 mutex_lock(&loop_index_mutex
);
1734 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1736 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1740 kobj
= get_disk(lo
->lo_disk
);
1741 mutex_unlock(&loop_index_mutex
);
1747 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1750 struct loop_device
*lo
;
1753 mutex_lock(&loop_index_mutex
);
1756 ret
= loop_lookup(&lo
, parm
);
1761 ret
= loop_add(&lo
, parm
);
1763 case LOOP_CTL_REMOVE
:
1764 ret
= loop_lookup(&lo
, parm
);
1767 mutex_lock(&lo
->lo_ctl_mutex
);
1768 if (lo
->lo_state
!= Lo_unbound
) {
1770 mutex_unlock(&lo
->lo_ctl_mutex
);
1773 if (lo
->lo_refcnt
> 0) {
1775 mutex_unlock(&lo
->lo_ctl_mutex
);
1778 lo
->lo_disk
->private_data
= NULL
;
1779 mutex_unlock(&lo
->lo_ctl_mutex
);
1780 idr_remove(&loop_index_idr
, lo
->lo_number
);
1783 case LOOP_CTL_GET_FREE
:
1784 ret
= loop_lookup(&lo
, -1);
1787 ret
= loop_add(&lo
, -1);
1789 mutex_unlock(&loop_index_mutex
);
1794 static const struct file_operations loop_ctl_fops
= {
1795 .open
= nonseekable_open
,
1796 .unlocked_ioctl
= loop_control_ioctl
,
1797 .compat_ioctl
= loop_control_ioctl
,
1798 .owner
= THIS_MODULE
,
1799 .llseek
= noop_llseek
,
1802 static struct miscdevice loop_misc
= {
1803 .minor
= LOOP_CTRL_MINOR
,
1804 .name
= "loop-control",
1805 .fops
= &loop_ctl_fops
,
1808 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1809 MODULE_ALIAS("devname:loop-control");
1811 static int __init
loop_init(void)
1814 unsigned long range
;
1815 struct loop_device
*lo
;
1818 err
= misc_register(&loop_misc
);
1824 part_shift
= fls(max_part
);
1827 * Adjust max_part according to part_shift as it is exported
1828 * to user space so that user can decide correct minor number
1829 * if [s]he want to create more devices.
1831 * Note that -1 is required because partition 0 is reserved
1832 * for the whole disk.
1834 max_part
= (1UL << part_shift
) - 1;
1837 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
1842 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
1848 * If max_loop is specified, create that many devices upfront.
1849 * This also becomes a hard limit. If max_loop is not specified,
1850 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1851 * init time. Loop devices can be requested on-demand with the
1852 * /dev/loop-control interface, or be instantiated by accessing
1853 * a 'dead' device node.
1857 range
= max_loop
<< part_shift
;
1859 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
1860 range
= 1UL << MINORBITS
;
1863 if (register_blkdev(LOOP_MAJOR
, "loop")) {
1868 loop_wq
= alloc_workqueue("kloopd",
1869 WQ_MEM_RECLAIM
| WQ_HIGHPRI
| WQ_UNBOUND
, 0);
1875 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1876 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1878 /* pre-create number of devices given by config or max_loop */
1879 mutex_lock(&loop_index_mutex
);
1880 for (i
= 0; i
< nr
; i
++)
1882 mutex_unlock(&loop_index_mutex
);
1884 printk(KERN_INFO
"loop: module loaded\n");
1888 misc_deregister(&loop_misc
);
1892 static int loop_exit_cb(int id
, void *ptr
, void *data
)
1894 struct loop_device
*lo
= ptr
;
1900 static void __exit
loop_exit(void)
1902 unsigned long range
;
1904 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
1906 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
1907 idr_destroy(&loop_index_idr
);
1909 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1910 unregister_blkdev(LOOP_MAJOR
, "loop");
1912 destroy_workqueue(loop_wq
);
1914 misc_deregister(&loop_misc
);
1917 module_init(loop_init
);
1918 module_exit(loop_exit
);
1921 static int __init
max_loop_setup(char *str
)
1923 max_loop
= simple_strtol(str
, NULL
, 0);
1927 __setup("max_loop=", max_loop_setup
);