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/loop.h>
67 #include <linux/compat.h>
68 #include <linux/suspend.h>
69 #include <linux/freezer.h>
70 #include <linux/mutex.h>
71 #include <linux/writeback.h>
72 #include <linux/completion.h>
73 #include <linux/highmem.h>
74 #include <linux/kthread.h>
75 #include <linux/splice.h>
76 #include <linux/sysfs.h>
77 #include <linux/miscdevice.h>
78 #include <linux/falloc.h>
80 #include <asm/uaccess.h>
82 static DEFINE_IDR(loop_index_idr
);
83 static DEFINE_MUTEX(loop_index_mutex
);
86 static int part_shift
;
91 static int transfer_none(struct loop_device
*lo
, int cmd
,
92 struct page
*raw_page
, unsigned raw_off
,
93 struct page
*loop_page
, unsigned loop_off
,
94 int size
, sector_t real_block
)
96 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
97 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
100 memcpy(loop_buf
, raw_buf
, size
);
102 memcpy(raw_buf
, loop_buf
, size
);
104 kunmap_atomic(loop_buf
);
105 kunmap_atomic(raw_buf
);
110 static int transfer_xor(struct loop_device
*lo
, int cmd
,
111 struct page
*raw_page
, unsigned raw_off
,
112 struct page
*loop_page
, unsigned loop_off
,
113 int size
, sector_t real_block
)
115 char *raw_buf
= kmap_atomic(raw_page
) + raw_off
;
116 char *loop_buf
= kmap_atomic(loop_page
) + loop_off
;
117 char *in
, *out
, *key
;
128 key
= lo
->lo_encrypt_key
;
129 keysize
= lo
->lo_encrypt_key_size
;
130 for (i
= 0; i
< size
; i
++)
131 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
133 kunmap_atomic(loop_buf
);
134 kunmap_atomic(raw_buf
);
139 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
141 if (unlikely(info
->lo_encrypt_key_size
<= 0))
146 static struct loop_func_table none_funcs
= {
147 .number
= LO_CRYPT_NONE
,
148 .transfer
= transfer_none
,
151 static struct loop_func_table xor_funcs
= {
152 .number
= LO_CRYPT_XOR
,
153 .transfer
= transfer_xor
,
157 /* xfer_funcs[0] is special - its release function is never called */
158 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
163 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
167 /* Compute loopsize in bytes */
168 loopsize
= i_size_read(file
->f_mapping
->host
);
171 /* offset is beyond i_size, weird but possible */
175 if (sizelimit
> 0 && sizelimit
< loopsize
)
176 loopsize
= sizelimit
;
178 * Unfortunately, if we want to do I/O on the device,
179 * the number of 512-byte sectors has to fit into a sector_t.
181 return loopsize
>> 9;
184 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
186 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
190 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
192 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
193 sector_t x
= (sector_t
)size
;
194 struct block_device
*bdev
= lo
->lo_device
;
196 if (unlikely((loff_t
)x
!= size
))
198 if (lo
->lo_offset
!= offset
)
199 lo
->lo_offset
= offset
;
200 if (lo
->lo_sizelimit
!= sizelimit
)
201 lo
->lo_sizelimit
= sizelimit
;
202 set_capacity(lo
->lo_disk
, x
);
203 bd_set_size(bdev
, (loff_t
)get_capacity(bdev
->bd_disk
) << 9);
204 /* let user-space know about the new size */
205 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
210 lo_do_transfer(struct loop_device
*lo
, int cmd
,
211 struct page
*rpage
, unsigned roffs
,
212 struct page
*lpage
, unsigned loffs
,
213 int size
, sector_t rblock
)
215 if (unlikely(!lo
->transfer
))
218 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
222 * __do_lo_send_write - helper for writing data to a loop device
224 * This helper just factors out common code between do_lo_send_direct_write()
225 * and do_lo_send_write().
227 static int __do_lo_send_write(struct file
*file
,
228 u8
*buf
, const int len
, loff_t pos
)
231 mm_segment_t old_fs
= get_fs();
234 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
236 if (likely(bw
== len
))
238 printk(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
239 (unsigned long long)pos
, len
);
246 * do_lo_send_direct_write - helper for writing data to a loop device
248 * This is the fast, non-transforming version that does not need double
251 static int do_lo_send_direct_write(struct loop_device
*lo
,
252 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
254 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
255 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
257 kunmap(bvec
->bv_page
);
263 * do_lo_send_write - helper for writing data to a loop device
265 * This is the slow, transforming version that needs to double buffer the
266 * data as it cannot do the transformations in place without having direct
267 * access to the destination pages of the backing file.
269 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
270 loff_t pos
, struct page
*page
)
272 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
273 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
275 return __do_lo_send_write(lo
->lo_backing_file
,
276 page_address(page
), bvec
->bv_len
,
278 printk(KERN_ERR
"loop: Transfer error at byte offset %llu, "
279 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
285 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
287 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
289 struct bio_vec
*bvec
;
290 struct page
*page
= NULL
;
293 if (lo
->transfer
!= transfer_none
) {
294 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
298 do_lo_send
= do_lo_send_write
;
300 do_lo_send
= do_lo_send_direct_write
;
303 bio_for_each_segment(bvec
, bio
, i
) {
304 ret
= do_lo_send(lo
, bvec
, pos
, page
);
316 printk(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
321 struct lo_read_data
{
322 struct loop_device
*lo
;
329 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
330 struct splice_desc
*sd
)
332 struct lo_read_data
*p
= sd
->u
.data
;
333 struct loop_device
*lo
= p
->lo
;
334 struct page
*page
= buf
->page
;
338 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
344 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
345 printk(KERN_ERR
"loop: transfer error block %ld\n",
350 flush_dcache_page(p
->page
);
359 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
361 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
365 do_lo_receive(struct loop_device
*lo
,
366 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
368 struct lo_read_data cookie
;
369 struct splice_desc sd
;
374 cookie
.page
= bvec
->bv_page
;
375 cookie
.offset
= bvec
->bv_offset
;
376 cookie
.bsize
= bsize
;
379 sd
.total_len
= bvec
->bv_len
;
384 file
= lo
->lo_backing_file
;
385 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
391 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
393 struct bio_vec
*bvec
;
397 bio_for_each_segment(bvec
, bio
, i
) {
398 s
= do_lo_receive(lo
, bvec
, bsize
, pos
);
402 if (s
!= bvec
->bv_len
) {
411 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
416 pos
= ((loff_t
) bio
->bi_sector
<< 9) + lo
->lo_offset
;
418 if (bio_rw(bio
) == WRITE
) {
419 struct file
*file
= lo
->lo_backing_file
;
421 if (bio
->bi_rw
& REQ_FLUSH
) {
422 ret
= vfs_fsync(file
, 0);
423 if (unlikely(ret
&& ret
!= -EINVAL
)) {
430 * We use punch hole to reclaim the free space used by the
431 * image a.k.a. discard. However we do not support discard if
432 * encryption is enabled, because it may give an attacker
433 * useful information.
435 if (bio
->bi_rw
& REQ_DISCARD
) {
436 struct file
*file
= lo
->lo_backing_file
;
437 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
439 if ((!file
->f_op
->fallocate
) ||
440 lo
->lo_encrypt_key_size
) {
444 ret
= file
->f_op
->fallocate(file
, mode
, pos
,
446 if (unlikely(ret
&& ret
!= -EINVAL
&&
452 ret
= lo_send(lo
, bio
, pos
);
454 if ((bio
->bi_rw
& REQ_FUA
) && !ret
) {
455 ret
= vfs_fsync(file
, 0);
456 if (unlikely(ret
&& ret
!= -EINVAL
))
460 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
467 * Add bio to back of pending list
469 static void loop_add_bio(struct loop_device
*lo
, struct bio
*bio
)
472 bio_list_add(&lo
->lo_bio_list
, bio
);
476 * Grab first pending buffer
478 static struct bio
*loop_get_bio(struct loop_device
*lo
)
481 return bio_list_pop(&lo
->lo_bio_list
);
484 static void loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
486 struct loop_device
*lo
= q
->queuedata
;
487 int rw
= bio_rw(old_bio
);
492 BUG_ON(!lo
|| (rw
!= READ
&& rw
!= WRITE
));
494 spin_lock_irq(&lo
->lo_lock
);
495 if (lo
->lo_state
!= Lo_bound
)
497 if (unlikely(rw
== WRITE
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)))
499 if (lo
->lo_bio_count
>= q
->nr_congestion_on
)
500 wait_event_lock_irq(lo
->lo_req_wait
,
501 lo
->lo_bio_count
< q
->nr_congestion_off
,
503 loop_add_bio(lo
, old_bio
);
504 wake_up(&lo
->lo_event
);
505 spin_unlock_irq(&lo
->lo_lock
);
509 spin_unlock_irq(&lo
->lo_lock
);
510 bio_io_error(old_bio
);
513 struct switch_request
{
515 struct completion wait
;
518 static void do_loop_switch(struct loop_device
*, struct switch_request
*);
520 static inline void loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
522 if (unlikely(!bio
->bi_bdev
)) {
523 do_loop_switch(lo
, bio
->bi_private
);
526 int ret
= do_bio_filebacked(lo
, bio
);
532 * worker thread that handles reads/writes to file backed loop devices,
533 * to avoid blocking in our make_request_fn. it also does loop decrypting
534 * on reads for block backed loop, as that is too heavy to do from
535 * b_end_io context where irqs may be disabled.
537 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
538 * calling kthread_stop(). Therefore once kthread_should_stop() is
539 * true, make_request will not place any more requests. Therefore
540 * once kthread_should_stop() is true and lo_bio is NULL, we are
541 * done with the loop.
543 static int loop_thread(void *data
)
545 struct loop_device
*lo
= data
;
548 set_user_nice(current
, -20);
550 while (!kthread_should_stop() || !bio_list_empty(&lo
->lo_bio_list
)) {
552 wait_event_interruptible(lo
->lo_event
,
553 !bio_list_empty(&lo
->lo_bio_list
) ||
554 kthread_should_stop());
556 if (bio_list_empty(&lo
->lo_bio_list
))
558 spin_lock_irq(&lo
->lo_lock
);
559 bio
= loop_get_bio(lo
);
560 if (lo
->lo_bio_count
< lo
->lo_queue
->nr_congestion_off
)
561 wake_up(&lo
->lo_req_wait
);
562 spin_unlock_irq(&lo
->lo_lock
);
565 loop_handle_bio(lo
, bio
);
572 * loop_switch performs the hard work of switching a backing store.
573 * First it needs to flush existing IO, it does this by sending a magic
574 * BIO down the pipe. The completion of this BIO does the actual switch.
576 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
578 struct switch_request w
;
579 struct bio
*bio
= bio_alloc(GFP_KERNEL
, 0);
582 init_completion(&w
.wait
);
584 bio
->bi_private
= &w
;
586 loop_make_request(lo
->lo_queue
, bio
);
587 wait_for_completion(&w
.wait
);
592 * Helper to flush the IOs in loop, but keeping loop thread running
594 static int loop_flush(struct loop_device
*lo
)
596 /* loop not yet configured, no running thread, nothing to flush */
600 return loop_switch(lo
, NULL
);
604 * Do the actual switch; called from the BIO completion routine
606 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
608 struct file
*file
= p
->file
;
609 struct file
*old_file
= lo
->lo_backing_file
;
610 struct address_space
*mapping
;
612 /* if no new file, only flush of queued bios requested */
616 mapping
= file
->f_mapping
;
617 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
618 lo
->lo_backing_file
= file
;
619 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
620 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
621 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
622 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
629 * loop_change_fd switched the backing store of a loopback device to
630 * a new file. This is useful for operating system installers to free up
631 * the original file and in High Availability environments to switch to
632 * an alternative location for the content in case of server meltdown.
633 * This can only work if the loop device is used read-only, and if the
634 * new backing store is the same size and type as the old backing store.
636 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
639 struct file
*file
, *old_file
;
644 if (lo
->lo_state
!= Lo_bound
)
647 /* the loop device has to be read-only */
649 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
657 inode
= file
->f_mapping
->host
;
658 old_file
= lo
->lo_backing_file
;
662 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
665 /* size of the new backing store needs to be the same */
666 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
670 error
= loop_switch(lo
, file
);
675 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
676 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
685 static inline int is_loop_device(struct file
*file
)
687 struct inode
*i
= file
->f_mapping
->host
;
689 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
692 /* loop sysfs attributes */
694 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
695 ssize_t (*callback
)(struct loop_device
*, char *))
697 struct gendisk
*disk
= dev_to_disk(dev
);
698 struct loop_device
*lo
= disk
->private_data
;
700 return callback(lo
, page
);
703 #define LOOP_ATTR_RO(_name) \
704 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
705 static ssize_t loop_attr_do_show_##_name(struct device *d, \
706 struct device_attribute *attr, char *b) \
708 return loop_attr_show(d, b, loop_attr_##_name##_show); \
710 static struct device_attribute loop_attr_##_name = \
711 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
713 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
718 spin_lock_irq(&lo
->lo_lock
);
719 if (lo
->lo_backing_file
)
720 p
= d_path(&lo
->lo_backing_file
->f_path
, buf
, PAGE_SIZE
- 1);
721 spin_unlock_irq(&lo
->lo_lock
);
723 if (IS_ERR_OR_NULL(p
))
727 memmove(buf
, p
, ret
);
735 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
737 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
740 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
742 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
745 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
747 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
749 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
752 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
754 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
756 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
759 LOOP_ATTR_RO(backing_file
);
760 LOOP_ATTR_RO(offset
);
761 LOOP_ATTR_RO(sizelimit
);
762 LOOP_ATTR_RO(autoclear
);
763 LOOP_ATTR_RO(partscan
);
765 static struct attribute
*loop_attrs
[] = {
766 &loop_attr_backing_file
.attr
,
767 &loop_attr_offset
.attr
,
768 &loop_attr_sizelimit
.attr
,
769 &loop_attr_autoclear
.attr
,
770 &loop_attr_partscan
.attr
,
774 static struct attribute_group loop_attribute_group
= {
779 static int loop_sysfs_init(struct loop_device
*lo
)
781 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
782 &loop_attribute_group
);
785 static void loop_sysfs_exit(struct loop_device
*lo
)
787 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
788 &loop_attribute_group
);
791 static void loop_config_discard(struct loop_device
*lo
)
793 struct file
*file
= lo
->lo_backing_file
;
794 struct inode
*inode
= file
->f_mapping
->host
;
795 struct request_queue
*q
= lo
->lo_queue
;
798 * We use punch hole to reclaim the free space used by the
799 * image a.k.a. discard. However we do support discard if
800 * encryption is enabled, because it may give an attacker
801 * useful information.
803 if ((!file
->f_op
->fallocate
) ||
804 lo
->lo_encrypt_key_size
) {
805 q
->limits
.discard_granularity
= 0;
806 q
->limits
.discard_alignment
= 0;
807 q
->limits
.max_discard_sectors
= 0;
808 q
->limits
.discard_zeroes_data
= 0;
809 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
813 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
814 q
->limits
.discard_alignment
= 0;
815 q
->limits
.max_discard_sectors
= UINT_MAX
>> 9;
816 q
->limits
.discard_zeroes_data
= 1;
817 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
820 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
821 struct block_device
*bdev
, unsigned int arg
)
823 struct file
*file
, *f
;
825 struct address_space
*mapping
;
826 unsigned lo_blocksize
;
831 /* This is safe, since we have a reference from open(). */
832 __module_get(THIS_MODULE
);
840 if (lo
->lo_state
!= Lo_unbound
)
843 /* Avoid recursion */
845 while (is_loop_device(f
)) {
846 struct loop_device
*l
;
848 if (f
->f_mapping
->host
->i_bdev
== bdev
)
851 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
852 if (l
->lo_state
== Lo_unbound
) {
856 f
= l
->lo_backing_file
;
859 mapping
= file
->f_mapping
;
860 inode
= mapping
->host
;
863 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
866 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
868 lo_flags
|= LO_FLAGS_READ_ONLY
;
870 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
871 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
874 size
= get_loop_size(lo
, file
);
875 if ((loff_t
)(sector_t
)size
!= size
)
880 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
882 lo
->lo_blocksize
= lo_blocksize
;
883 lo
->lo_device
= bdev
;
884 lo
->lo_flags
= lo_flags
;
885 lo
->lo_backing_file
= file
;
886 lo
->transfer
= transfer_none
;
888 lo
->lo_sizelimit
= 0;
889 lo
->lo_bio_count
= 0;
890 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
891 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
893 bio_list_init(&lo
->lo_bio_list
);
896 * set queue make_request_fn, and add limits based on lower level
899 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
900 lo
->lo_queue
->queuedata
= lo
;
902 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
903 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
905 set_capacity(lo
->lo_disk
, size
);
906 bd_set_size(bdev
, size
<< 9);
908 /* let user-space know about the new size */
909 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
911 set_blocksize(bdev
, lo_blocksize
);
913 lo
->lo_thread
= kthread_create(loop_thread
, lo
, "loop%d",
915 if (IS_ERR(lo
->lo_thread
)) {
916 error
= PTR_ERR(lo
->lo_thread
);
919 lo
->lo_state
= Lo_bound
;
920 wake_up_process(lo
->lo_thread
);
922 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
923 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
924 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
926 /* Grab the block_device to prevent its destruction after we
927 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
934 lo
->lo_thread
= NULL
;
935 lo
->lo_device
= NULL
;
936 lo
->lo_backing_file
= NULL
;
938 set_capacity(lo
->lo_disk
, 0);
939 invalidate_bdev(bdev
);
940 bd_set_size(bdev
, 0);
941 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
942 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
);
943 lo
->lo_state
= Lo_unbound
;
947 /* This is safe: open() is still holding a reference. */
948 module_put(THIS_MODULE
);
953 loop_release_xfer(struct loop_device
*lo
)
956 struct loop_func_table
*xfer
= lo
->lo_encryption
;
960 err
= xfer
->release(lo
);
962 lo
->lo_encryption
= NULL
;
963 module_put(xfer
->owner
);
969 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
970 const struct loop_info64
*i
)
975 struct module
*owner
= xfer
->owner
;
977 if (!try_module_get(owner
))
980 err
= xfer
->init(lo
, i
);
984 lo
->lo_encryption
= xfer
;
989 static int loop_clr_fd(struct loop_device
*lo
)
991 struct file
*filp
= lo
->lo_backing_file
;
992 gfp_t gfp
= lo
->old_gfp_mask
;
993 struct block_device
*bdev
= lo
->lo_device
;
995 if (lo
->lo_state
!= Lo_bound
)
999 * If we've explicitly asked to tear down the loop device,
1000 * and it has an elevated reference count, set it for auto-teardown when
1001 * the last reference goes away. This stops $!~#$@ udev from
1002 * preventing teardown because it decided that it needs to run blkid on
1003 * the loopback device whenever they appear. xfstests is notorious for
1004 * failing tests because blkid via udev races with a losetup
1005 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1006 * command to fail with EBUSY.
1008 if (lo
->lo_refcnt
> 1) {
1009 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1010 mutex_unlock(&lo
->lo_ctl_mutex
);
1017 spin_lock_irq(&lo
->lo_lock
);
1018 lo
->lo_state
= Lo_rundown
;
1019 spin_unlock_irq(&lo
->lo_lock
);
1021 kthread_stop(lo
->lo_thread
);
1023 spin_lock_irq(&lo
->lo_lock
);
1024 lo
->lo_backing_file
= NULL
;
1025 spin_unlock_irq(&lo
->lo_lock
);
1027 loop_release_xfer(lo
);
1028 lo
->transfer
= NULL
;
1030 lo
->lo_device
= NULL
;
1031 lo
->lo_encryption
= NULL
;
1033 lo
->lo_sizelimit
= 0;
1034 lo
->lo_encrypt_key_size
= 0;
1035 lo
->lo_thread
= NULL
;
1036 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1037 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1038 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1041 invalidate_bdev(bdev
);
1043 set_capacity(lo
->lo_disk
, 0);
1044 loop_sysfs_exit(lo
);
1046 bd_set_size(bdev
, 0);
1047 /* let user-space know about this change */
1048 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1050 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1051 lo
->lo_state
= Lo_unbound
;
1052 /* This is safe: open() is still holding a reference. */
1053 module_put(THIS_MODULE
);
1054 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1055 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
1058 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1059 mutex_unlock(&lo
->lo_ctl_mutex
);
1061 * Need not hold lo_ctl_mutex to fput backing file.
1062 * Calling fput holding lo_ctl_mutex triggers a circular
1063 * lock dependency possibility warning as fput can take
1064 * bd_mutex which is usually taken before lo_ctl_mutex.
1071 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1074 struct loop_func_table
*xfer
;
1075 kuid_t uid
= current_uid();
1077 if (lo
->lo_encrypt_key_size
&&
1078 !uid_eq(lo
->lo_key_owner
, uid
) &&
1079 !capable(CAP_SYS_ADMIN
))
1081 if (lo
->lo_state
!= Lo_bound
)
1083 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1086 err
= loop_release_xfer(lo
);
1090 if (info
->lo_encrypt_type
) {
1091 unsigned int type
= info
->lo_encrypt_type
;
1093 if (type
>= MAX_LO_CRYPT
)
1095 xfer
= xfer_funcs
[type
];
1101 err
= loop_init_xfer(lo
, xfer
, info
);
1105 if (lo
->lo_offset
!= info
->lo_offset
||
1106 lo
->lo_sizelimit
!= info
->lo_sizelimit
)
1107 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
))
1110 loop_config_discard(lo
);
1112 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1113 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1114 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1115 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1119 lo
->transfer
= xfer
->transfer
;
1120 lo
->ioctl
= xfer
->ioctl
;
1122 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1123 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1124 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1126 if ((info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1127 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1128 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1129 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1130 ioctl_by_bdev(lo
->lo_device
, BLKRRPART
, 0);
1133 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1134 lo
->lo_init
[0] = info
->lo_init
[0];
1135 lo
->lo_init
[1] = info
->lo_init
[1];
1136 if (info
->lo_encrypt_key_size
) {
1137 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1138 info
->lo_encrypt_key_size
);
1139 lo
->lo_key_owner
= uid
;
1146 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1148 struct file
*file
= lo
->lo_backing_file
;
1152 if (lo
->lo_state
!= Lo_bound
)
1154 error
= vfs_getattr(&file
->f_path
, &stat
);
1157 memset(info
, 0, sizeof(*info
));
1158 info
->lo_number
= lo
->lo_number
;
1159 info
->lo_device
= huge_encode_dev(stat
.dev
);
1160 info
->lo_inode
= stat
.ino
;
1161 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1162 info
->lo_offset
= lo
->lo_offset
;
1163 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1164 info
->lo_flags
= lo
->lo_flags
;
1165 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1166 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1167 info
->lo_encrypt_type
=
1168 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1169 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1170 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1171 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1172 lo
->lo_encrypt_key_size
);
1178 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1180 memset(info64
, 0, sizeof(*info64
));
1181 info64
->lo_number
= info
->lo_number
;
1182 info64
->lo_device
= info
->lo_device
;
1183 info64
->lo_inode
= info
->lo_inode
;
1184 info64
->lo_rdevice
= info
->lo_rdevice
;
1185 info64
->lo_offset
= info
->lo_offset
;
1186 info64
->lo_sizelimit
= 0;
1187 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1188 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1189 info64
->lo_flags
= info
->lo_flags
;
1190 info64
->lo_init
[0] = info
->lo_init
[0];
1191 info64
->lo_init
[1] = info
->lo_init
[1];
1192 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1193 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1195 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1196 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1200 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1202 memset(info
, 0, sizeof(*info
));
1203 info
->lo_number
= info64
->lo_number
;
1204 info
->lo_device
= info64
->lo_device
;
1205 info
->lo_inode
= info64
->lo_inode
;
1206 info
->lo_rdevice
= info64
->lo_rdevice
;
1207 info
->lo_offset
= info64
->lo_offset
;
1208 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1209 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1210 info
->lo_flags
= info64
->lo_flags
;
1211 info
->lo_init
[0] = info64
->lo_init
[0];
1212 info
->lo_init
[1] = info64
->lo_init
[1];
1213 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1214 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1216 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1217 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1219 /* error in case values were truncated */
1220 if (info
->lo_device
!= info64
->lo_device
||
1221 info
->lo_rdevice
!= info64
->lo_rdevice
||
1222 info
->lo_inode
!= info64
->lo_inode
||
1223 info
->lo_offset
!= info64
->lo_offset
)
1230 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1232 struct loop_info info
;
1233 struct loop_info64 info64
;
1235 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1237 loop_info64_from_old(&info
, &info64
);
1238 return loop_set_status(lo
, &info64
);
1242 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1244 struct loop_info64 info64
;
1246 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1248 return loop_set_status(lo
, &info64
);
1252 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1253 struct loop_info info
;
1254 struct loop_info64 info64
;
1260 err
= loop_get_status(lo
, &info64
);
1262 err
= loop_info64_to_old(&info64
, &info
);
1263 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1270 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1271 struct loop_info64 info64
;
1277 err
= loop_get_status(lo
, &info64
);
1278 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1284 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1286 if (unlikely(lo
->lo_state
!= Lo_bound
))
1289 return figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1292 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1293 unsigned int cmd
, unsigned long arg
)
1295 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1298 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1301 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1303 case LOOP_CHANGE_FD
:
1304 err
= loop_change_fd(lo
, bdev
, arg
);
1307 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1308 err
= loop_clr_fd(lo
);
1312 case LOOP_SET_STATUS
:
1314 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1315 err
= loop_set_status_old(lo
,
1316 (struct loop_info __user
*)arg
);
1318 case LOOP_GET_STATUS
:
1319 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1321 case LOOP_SET_STATUS64
:
1323 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1324 err
= loop_set_status64(lo
,
1325 (struct loop_info64 __user
*) arg
);
1327 case LOOP_GET_STATUS64
:
1328 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1330 case LOOP_SET_CAPACITY
:
1332 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1333 err
= loop_set_capacity(lo
, bdev
);
1336 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1338 mutex_unlock(&lo
->lo_ctl_mutex
);
1344 #ifdef CONFIG_COMPAT
1345 struct compat_loop_info
{
1346 compat_int_t lo_number
; /* ioctl r/o */
1347 compat_dev_t lo_device
; /* ioctl r/o */
1348 compat_ulong_t lo_inode
; /* ioctl r/o */
1349 compat_dev_t lo_rdevice
; /* ioctl r/o */
1350 compat_int_t lo_offset
;
1351 compat_int_t lo_encrypt_type
;
1352 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1353 compat_int_t lo_flags
; /* ioctl r/o */
1354 char lo_name
[LO_NAME_SIZE
];
1355 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1356 compat_ulong_t lo_init
[2];
1361 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1362 * - noinlined to reduce stack space usage in main part of driver
1365 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1366 struct loop_info64
*info64
)
1368 struct compat_loop_info info
;
1370 if (copy_from_user(&info
, arg
, sizeof(info
)))
1373 memset(info64
, 0, sizeof(*info64
));
1374 info64
->lo_number
= info
.lo_number
;
1375 info64
->lo_device
= info
.lo_device
;
1376 info64
->lo_inode
= info
.lo_inode
;
1377 info64
->lo_rdevice
= info
.lo_rdevice
;
1378 info64
->lo_offset
= info
.lo_offset
;
1379 info64
->lo_sizelimit
= 0;
1380 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1381 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1382 info64
->lo_flags
= info
.lo_flags
;
1383 info64
->lo_init
[0] = info
.lo_init
[0];
1384 info64
->lo_init
[1] = info
.lo_init
[1];
1385 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1386 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1388 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1389 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1394 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1395 * - noinlined to reduce stack space usage in main part of driver
1398 loop_info64_to_compat(const struct loop_info64
*info64
,
1399 struct compat_loop_info __user
*arg
)
1401 struct compat_loop_info info
;
1403 memset(&info
, 0, sizeof(info
));
1404 info
.lo_number
= info64
->lo_number
;
1405 info
.lo_device
= info64
->lo_device
;
1406 info
.lo_inode
= info64
->lo_inode
;
1407 info
.lo_rdevice
= info64
->lo_rdevice
;
1408 info
.lo_offset
= info64
->lo_offset
;
1409 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1410 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1411 info
.lo_flags
= info64
->lo_flags
;
1412 info
.lo_init
[0] = info64
->lo_init
[0];
1413 info
.lo_init
[1] = info64
->lo_init
[1];
1414 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1415 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1417 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1418 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1420 /* error in case values were truncated */
1421 if (info
.lo_device
!= info64
->lo_device
||
1422 info
.lo_rdevice
!= info64
->lo_rdevice
||
1423 info
.lo_inode
!= info64
->lo_inode
||
1424 info
.lo_offset
!= info64
->lo_offset
||
1425 info
.lo_init
[0] != info64
->lo_init
[0] ||
1426 info
.lo_init
[1] != info64
->lo_init
[1])
1429 if (copy_to_user(arg
, &info
, sizeof(info
)))
1435 loop_set_status_compat(struct loop_device
*lo
,
1436 const struct compat_loop_info __user
*arg
)
1438 struct loop_info64 info64
;
1441 ret
= loop_info64_from_compat(arg
, &info64
);
1444 return loop_set_status(lo
, &info64
);
1448 loop_get_status_compat(struct loop_device
*lo
,
1449 struct compat_loop_info __user
*arg
)
1451 struct loop_info64 info64
;
1457 err
= loop_get_status(lo
, &info64
);
1459 err
= loop_info64_to_compat(&info64
, arg
);
1463 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1464 unsigned int cmd
, unsigned long arg
)
1466 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1470 case LOOP_SET_STATUS
:
1471 mutex_lock(&lo
->lo_ctl_mutex
);
1472 err
= loop_set_status_compat(
1473 lo
, (const struct compat_loop_info __user
*) arg
);
1474 mutex_unlock(&lo
->lo_ctl_mutex
);
1476 case LOOP_GET_STATUS
:
1477 mutex_lock(&lo
->lo_ctl_mutex
);
1478 err
= loop_get_status_compat(
1479 lo
, (struct compat_loop_info __user
*) arg
);
1480 mutex_unlock(&lo
->lo_ctl_mutex
);
1482 case LOOP_SET_CAPACITY
:
1484 case LOOP_GET_STATUS64
:
1485 case LOOP_SET_STATUS64
:
1486 arg
= (unsigned long) compat_ptr(arg
);
1488 case LOOP_CHANGE_FD
:
1489 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1499 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1501 struct loop_device
*lo
;
1504 mutex_lock(&loop_index_mutex
);
1505 lo
= bdev
->bd_disk
->private_data
;
1511 mutex_lock(&lo
->lo_ctl_mutex
);
1513 mutex_unlock(&lo
->lo_ctl_mutex
);
1515 mutex_unlock(&loop_index_mutex
);
1519 static int lo_release(struct gendisk
*disk
, fmode_t mode
)
1521 struct loop_device
*lo
= disk
->private_data
;
1524 mutex_lock(&lo
->lo_ctl_mutex
);
1526 if (--lo
->lo_refcnt
)
1529 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1531 * In autoclear mode, stop the loop thread
1532 * and remove configuration after last close.
1534 err
= loop_clr_fd(lo
);
1539 * Otherwise keep thread (if running) and config,
1540 * but flush possible ongoing bios in thread.
1546 mutex_unlock(&lo
->lo_ctl_mutex
);
1551 static const struct block_device_operations lo_fops
= {
1552 .owner
= THIS_MODULE
,
1554 .release
= lo_release
,
1556 #ifdef CONFIG_COMPAT
1557 .compat_ioctl
= lo_compat_ioctl
,
1562 * And now the modules code and kernel interface.
1564 static int max_loop
;
1565 module_param(max_loop
, int, S_IRUGO
);
1566 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1567 module_param(max_part
, int, S_IRUGO
);
1568 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1569 MODULE_LICENSE("GPL");
1570 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1572 int loop_register_transfer(struct loop_func_table
*funcs
)
1574 unsigned int n
= funcs
->number
;
1576 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1578 xfer_funcs
[n
] = funcs
;
1582 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1584 struct loop_device
*lo
= ptr
;
1585 struct loop_func_table
*xfer
= data
;
1587 mutex_lock(&lo
->lo_ctl_mutex
);
1588 if (lo
->lo_encryption
== xfer
)
1589 loop_release_xfer(lo
);
1590 mutex_unlock(&lo
->lo_ctl_mutex
);
1594 int loop_unregister_transfer(int number
)
1596 unsigned int n
= number
;
1597 struct loop_func_table
*xfer
;
1599 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1602 xfer_funcs
[n
] = NULL
;
1603 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1607 EXPORT_SYMBOL(loop_register_transfer
);
1608 EXPORT_SYMBOL(loop_unregister_transfer
);
1610 static int loop_add(struct loop_device
**l
, int i
)
1612 struct loop_device
*lo
;
1613 struct gendisk
*disk
;
1617 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1621 /* allocate id, if @id >= 0, we're requesting that specific id */
1623 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
1627 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
1634 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
1638 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1640 goto out_free_queue
;
1643 * Disable partition scanning by default. The in-kernel partition
1644 * scanning can be requested individually per-device during its
1645 * setup. Userspace can always add and remove partitions from all
1646 * devices. The needed partition minors are allocated from the
1647 * extended minor space, the main loop device numbers will continue
1648 * to match the loop minors, regardless of the number of partitions
1651 * If max_part is given, partition scanning is globally enabled for
1652 * all loop devices. The minors for the main loop devices will be
1653 * multiples of max_part.
1655 * Note: Global-for-all-devices, set-only-at-init, read-only module
1656 * parameteters like 'max_loop' and 'max_part' make things needlessly
1657 * complicated, are too static, inflexible and may surprise
1658 * userspace tools. Parameters like this in general should be avoided.
1661 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1662 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1663 mutex_init(&lo
->lo_ctl_mutex
);
1665 lo
->lo_thread
= NULL
;
1666 init_waitqueue_head(&lo
->lo_event
);
1667 init_waitqueue_head(&lo
->lo_req_wait
);
1668 spin_lock_init(&lo
->lo_lock
);
1669 disk
->major
= LOOP_MAJOR
;
1670 disk
->first_minor
= i
<< part_shift
;
1671 disk
->fops
= &lo_fops
;
1672 disk
->private_data
= lo
;
1673 disk
->queue
= lo
->lo_queue
;
1674 sprintf(disk
->disk_name
, "loop%d", i
);
1677 return lo
->lo_number
;
1680 blk_cleanup_queue(lo
->lo_queue
);
1687 static void loop_remove(struct loop_device
*lo
)
1689 del_gendisk(lo
->lo_disk
);
1690 blk_cleanup_queue(lo
->lo_queue
);
1691 put_disk(lo
->lo_disk
);
1695 static int find_free_cb(int id
, void *ptr
, void *data
)
1697 struct loop_device
*lo
= ptr
;
1698 struct loop_device
**l
= data
;
1700 if (lo
->lo_state
== Lo_unbound
) {
1707 static int loop_lookup(struct loop_device
**l
, int i
)
1709 struct loop_device
*lo
;
1715 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1718 ret
= lo
->lo_number
;
1723 /* lookup and return a specific i */
1724 lo
= idr_find(&loop_index_idr
, i
);
1727 ret
= lo
->lo_number
;
1733 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1735 struct loop_device
*lo
;
1736 struct kobject
*kobj
;
1739 mutex_lock(&loop_index_mutex
);
1740 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1742 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1744 kobj
= ERR_PTR(err
);
1746 kobj
= get_disk(lo
->lo_disk
);
1747 mutex_unlock(&loop_index_mutex
);
1753 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1756 struct loop_device
*lo
;
1759 mutex_lock(&loop_index_mutex
);
1762 ret
= loop_lookup(&lo
, parm
);
1767 ret
= loop_add(&lo
, parm
);
1769 case LOOP_CTL_REMOVE
:
1770 ret
= loop_lookup(&lo
, parm
);
1773 mutex_lock(&lo
->lo_ctl_mutex
);
1774 if (lo
->lo_state
!= Lo_unbound
) {
1776 mutex_unlock(&lo
->lo_ctl_mutex
);
1779 if (lo
->lo_refcnt
> 0) {
1781 mutex_unlock(&lo
->lo_ctl_mutex
);
1784 lo
->lo_disk
->private_data
= NULL
;
1785 mutex_unlock(&lo
->lo_ctl_mutex
);
1786 idr_remove(&loop_index_idr
, lo
->lo_number
);
1789 case LOOP_CTL_GET_FREE
:
1790 ret
= loop_lookup(&lo
, -1);
1793 ret
= loop_add(&lo
, -1);
1795 mutex_unlock(&loop_index_mutex
);
1800 static const struct file_operations loop_ctl_fops
= {
1801 .open
= nonseekable_open
,
1802 .unlocked_ioctl
= loop_control_ioctl
,
1803 .compat_ioctl
= loop_control_ioctl
,
1804 .owner
= THIS_MODULE
,
1805 .llseek
= noop_llseek
,
1808 static struct miscdevice loop_misc
= {
1809 .minor
= LOOP_CTRL_MINOR
,
1810 .name
= "loop-control",
1811 .fops
= &loop_ctl_fops
,
1814 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1815 MODULE_ALIAS("devname:loop-control");
1817 static int __init
loop_init(void)
1820 unsigned long range
;
1821 struct loop_device
*lo
;
1824 err
= misc_register(&loop_misc
);
1830 part_shift
= fls(max_part
);
1833 * Adjust max_part according to part_shift as it is exported
1834 * to user space so that user can decide correct minor number
1835 * if [s]he want to create more devices.
1837 * Note that -1 is required because partition 0 is reserved
1838 * for the whole disk.
1840 max_part
= (1UL << part_shift
) - 1;
1843 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
1848 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
1854 * If max_loop is specified, create that many devices upfront.
1855 * This also becomes a hard limit. If max_loop is not specified,
1856 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1857 * init time. Loop devices can be requested on-demand with the
1858 * /dev/loop-control interface, or be instantiated by accessing
1859 * a 'dead' device node.
1863 range
= max_loop
<< part_shift
;
1865 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
1866 range
= 1UL << MINORBITS
;
1869 if (register_blkdev(LOOP_MAJOR
, "loop")) {
1874 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1875 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1877 /* pre-create number of devices given by config or max_loop */
1878 mutex_lock(&loop_index_mutex
);
1879 for (i
= 0; i
< nr
; i
++)
1881 mutex_unlock(&loop_index_mutex
);
1883 printk(KERN_INFO
"loop: module loaded\n");
1887 misc_deregister(&loop_misc
);
1891 static int loop_exit_cb(int id
, void *ptr
, void *data
)
1893 struct loop_device
*lo
= ptr
;
1899 static void __exit
loop_exit(void)
1901 unsigned long range
;
1903 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
1905 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
1906 idr_destroy(&loop_index_idr
);
1908 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1909 unregister_blkdev(LOOP_MAJOR
, "loop");
1911 misc_deregister(&loop_misc
);
1914 module_init(loop_init
);
1915 module_exit(loop_exit
);
1918 static int __init
max_loop_setup(char *str
)
1920 max_loop
= simple_strtol(str
, NULL
, 0);
1924 __setup("max_loop=", max_loop_setup
);