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
)
165 loff_t size
, loopsize
;
167 /* Compute loopsize in bytes */
168 size
= i_size_read(file
->f_mapping
->host
);
169 loopsize
= size
- offset
;
170 /* offset is beyond i_size, wierd but possible */
174 if (sizelimit
> 0 && sizelimit
< loopsize
)
175 loopsize
= sizelimit
;
177 * Unfortunately, if we want to do I/O on the device,
178 * the number of 512-byte sectors has to fit into a sector_t.
180 return loopsize
>> 9;
183 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
185 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
189 figure_loop_size(struct loop_device
*lo
, loff_t offset
, loff_t sizelimit
)
191 loff_t size
= get_size(offset
, sizelimit
, lo
->lo_backing_file
);
192 sector_t x
= (sector_t
)size
;
194 if (unlikely((loff_t
)x
!= size
))
196 if (lo
->lo_offset
!= offset
)
197 lo
->lo_offset
= offset
;
198 if (lo
->lo_sizelimit
!= sizelimit
)
199 lo
->lo_sizelimit
= sizelimit
;
200 set_capacity(lo
->lo_disk
, x
);
205 lo_do_transfer(struct loop_device
*lo
, int cmd
,
206 struct page
*rpage
, unsigned roffs
,
207 struct page
*lpage
, unsigned loffs
,
208 int size
, sector_t rblock
)
210 if (unlikely(!lo
->transfer
))
213 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
217 * __do_lo_send_write - helper for writing data to a loop device
219 * This helper just factors out common code between do_lo_send_direct_write()
220 * and do_lo_send_write().
222 static int __do_lo_send_write(struct file
*file
,
223 u8
*buf
, const int len
, loff_t pos
)
226 mm_segment_t old_fs
= get_fs();
229 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
231 if (likely(bw
== len
))
233 printk(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
234 (unsigned long long)pos
, len
);
241 * do_lo_send_direct_write - helper for writing data to a loop device
243 * This is the fast, non-transforming version that does not need double
246 static int do_lo_send_direct_write(struct loop_device
*lo
,
247 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
249 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
250 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
252 kunmap(bvec
->bv_page
);
258 * do_lo_send_write - helper for writing data to a loop device
260 * This is the slow, transforming version that needs to double buffer the
261 * data as it cannot do the transformations in place without having direct
262 * access to the destination pages of the backing file.
264 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
265 loff_t pos
, struct page
*page
)
267 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
268 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
270 return __do_lo_send_write(lo
->lo_backing_file
,
271 page_address(page
), bvec
->bv_len
,
273 printk(KERN_ERR
"loop: Transfer error at byte offset %llu, "
274 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
280 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
282 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
284 struct bio_vec
*bvec
;
285 struct page
*page
= NULL
;
288 if (lo
->transfer
!= transfer_none
) {
289 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
293 do_lo_send
= do_lo_send_write
;
295 do_lo_send
= do_lo_send_direct_write
;
298 bio_for_each_segment(bvec
, bio
, i
) {
299 ret
= do_lo_send(lo
, bvec
, pos
, page
);
311 printk(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
316 struct lo_read_data
{
317 struct loop_device
*lo
;
324 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
325 struct splice_desc
*sd
)
327 struct lo_read_data
*p
= sd
->u
.data
;
328 struct loop_device
*lo
= p
->lo
;
329 struct page
*page
= buf
->page
;
333 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
339 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
340 printk(KERN_ERR
"loop: transfer error block %ld\n",
345 flush_dcache_page(p
->page
);
354 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
356 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
360 do_lo_receive(struct loop_device
*lo
,
361 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
363 struct lo_read_data cookie
;
364 struct splice_desc sd
;
369 cookie
.page
= bvec
->bv_page
;
370 cookie
.offset
= bvec
->bv_offset
;
371 cookie
.bsize
= bsize
;
374 sd
.total_len
= bvec
->bv_len
;
379 file
= lo
->lo_backing_file
;
380 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
386 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
388 struct bio_vec
*bvec
;
392 bio_for_each_segment(bvec
, bio
, i
) {
393 s
= do_lo_receive(lo
, bvec
, bsize
, pos
);
397 if (s
!= bvec
->bv_len
) {
406 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
411 pos
= ((loff_t
) bio
->bi_sector
<< 9) + lo
->lo_offset
;
413 if (bio_rw(bio
) == WRITE
) {
414 struct file
*file
= lo
->lo_backing_file
;
416 if (bio
->bi_rw
& REQ_FLUSH
) {
417 ret
= vfs_fsync(file
, 0);
418 if (unlikely(ret
&& ret
!= -EINVAL
)) {
425 * We use punch hole to reclaim the free space used by the
426 * image a.k.a. discard. However we do not support discard if
427 * encryption is enabled, because it may give an attacker
428 * useful information.
430 if (bio
->bi_rw
& REQ_DISCARD
) {
431 struct file
*file
= lo
->lo_backing_file
;
432 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
434 if ((!file
->f_op
->fallocate
) ||
435 lo
->lo_encrypt_key_size
) {
439 ret
= file
->f_op
->fallocate(file
, mode
, pos
,
441 if (unlikely(ret
&& ret
!= -EINVAL
&&
447 ret
= lo_send(lo
, bio
, pos
);
449 if ((bio
->bi_rw
& REQ_FUA
) && !ret
) {
450 ret
= vfs_fsync(file
, 0);
451 if (unlikely(ret
&& ret
!= -EINVAL
))
455 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
462 * Add bio to back of pending list
464 static void loop_add_bio(struct loop_device
*lo
, struct bio
*bio
)
467 bio_list_add(&lo
->lo_bio_list
, bio
);
471 * Grab first pending buffer
473 static struct bio
*loop_get_bio(struct loop_device
*lo
)
476 return bio_list_pop(&lo
->lo_bio_list
);
479 static void loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
481 struct loop_device
*lo
= q
->queuedata
;
482 int rw
= bio_rw(old_bio
);
487 BUG_ON(!lo
|| (rw
!= READ
&& rw
!= WRITE
));
489 spin_lock_irq(&lo
->lo_lock
);
490 if (lo
->lo_state
!= Lo_bound
)
492 if (unlikely(rw
== WRITE
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)))
494 if (lo
->lo_bio_count
>= q
->nr_congestion_on
)
495 wait_event_lock_irq(lo
->lo_req_wait
,
496 lo
->lo_bio_count
< q
->nr_congestion_off
,
498 loop_add_bio(lo
, old_bio
);
499 wake_up(&lo
->lo_event
);
500 spin_unlock_irq(&lo
->lo_lock
);
504 spin_unlock_irq(&lo
->lo_lock
);
505 bio_io_error(old_bio
);
508 struct switch_request
{
510 struct completion wait
;
513 static void do_loop_switch(struct loop_device
*, struct switch_request
*);
515 static inline void loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
517 if (unlikely(!bio
->bi_bdev
)) {
518 do_loop_switch(lo
, bio
->bi_private
);
521 int ret
= do_bio_filebacked(lo
, bio
);
527 * worker thread that handles reads/writes to file backed loop devices,
528 * to avoid blocking in our make_request_fn. it also does loop decrypting
529 * on reads for block backed loop, as that is too heavy to do from
530 * b_end_io context where irqs may be disabled.
532 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
533 * calling kthread_stop(). Therefore once kthread_should_stop() is
534 * true, make_request will not place any more requests. Therefore
535 * once kthread_should_stop() is true and lo_bio is NULL, we are
536 * done with the loop.
538 static int loop_thread(void *data
)
540 struct loop_device
*lo
= data
;
543 set_user_nice(current
, -20);
545 while (!kthread_should_stop() || !bio_list_empty(&lo
->lo_bio_list
)) {
547 wait_event_interruptible(lo
->lo_event
,
548 !bio_list_empty(&lo
->lo_bio_list
) ||
549 kthread_should_stop());
551 if (bio_list_empty(&lo
->lo_bio_list
))
553 spin_lock_irq(&lo
->lo_lock
);
554 bio
= loop_get_bio(lo
);
555 if (lo
->lo_bio_count
< lo
->lo_queue
->nr_congestion_off
)
556 wake_up(&lo
->lo_req_wait
);
557 spin_unlock_irq(&lo
->lo_lock
);
560 loop_handle_bio(lo
, bio
);
567 * loop_switch performs the hard work of switching a backing store.
568 * First it needs to flush existing IO, it does this by sending a magic
569 * BIO down the pipe. The completion of this BIO does the actual switch.
571 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
573 struct switch_request w
;
574 struct bio
*bio
= bio_alloc(GFP_KERNEL
, 0);
577 init_completion(&w
.wait
);
579 bio
->bi_private
= &w
;
581 loop_make_request(lo
->lo_queue
, bio
);
582 wait_for_completion(&w
.wait
);
587 * Helper to flush the IOs in loop, but keeping loop thread running
589 static int loop_flush(struct loop_device
*lo
)
591 /* loop not yet configured, no running thread, nothing to flush */
595 return loop_switch(lo
, NULL
);
599 * Do the actual switch; called from the BIO completion routine
601 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
603 struct file
*file
= p
->file
;
604 struct file
*old_file
= lo
->lo_backing_file
;
605 struct address_space
*mapping
;
607 /* if no new file, only flush of queued bios requested */
611 mapping
= file
->f_mapping
;
612 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
613 lo
->lo_backing_file
= file
;
614 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
615 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
616 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
617 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
624 * loop_change_fd switched the backing store of a loopback device to
625 * a new file. This is useful for operating system installers to free up
626 * the original file and in High Availability environments to switch to
627 * an alternative location for the content in case of server meltdown.
628 * This can only work if the loop device is used read-only, and if the
629 * new backing store is the same size and type as the old backing store.
631 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
634 struct file
*file
, *old_file
;
639 if (lo
->lo_state
!= Lo_bound
)
642 /* the loop device has to be read-only */
644 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
652 inode
= file
->f_mapping
->host
;
653 old_file
= lo
->lo_backing_file
;
657 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
660 /* size of the new backing store needs to be the same */
661 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
665 error
= loop_switch(lo
, file
);
670 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
671 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
680 static inline int is_loop_device(struct file
*file
)
682 struct inode
*i
= file
->f_mapping
->host
;
684 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
687 /* loop sysfs attributes */
689 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
690 ssize_t (*callback
)(struct loop_device
*, char *))
692 struct gendisk
*disk
= dev_to_disk(dev
);
693 struct loop_device
*lo
= disk
->private_data
;
695 return callback(lo
, page
);
698 #define LOOP_ATTR_RO(_name) \
699 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
700 static ssize_t loop_attr_do_show_##_name(struct device *d, \
701 struct device_attribute *attr, char *b) \
703 return loop_attr_show(d, b, loop_attr_##_name##_show); \
705 static struct device_attribute loop_attr_##_name = \
706 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
708 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
713 spin_lock_irq(&lo
->lo_lock
);
714 if (lo
->lo_backing_file
)
715 p
= d_path(&lo
->lo_backing_file
->f_path
, buf
, PAGE_SIZE
- 1);
716 spin_unlock_irq(&lo
->lo_lock
);
718 if (IS_ERR_OR_NULL(p
))
722 memmove(buf
, p
, ret
);
730 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
732 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
735 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
737 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
740 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
742 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
744 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
747 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
749 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
751 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
754 LOOP_ATTR_RO(backing_file
);
755 LOOP_ATTR_RO(offset
);
756 LOOP_ATTR_RO(sizelimit
);
757 LOOP_ATTR_RO(autoclear
);
758 LOOP_ATTR_RO(partscan
);
760 static struct attribute
*loop_attrs
[] = {
761 &loop_attr_backing_file
.attr
,
762 &loop_attr_offset
.attr
,
763 &loop_attr_sizelimit
.attr
,
764 &loop_attr_autoclear
.attr
,
765 &loop_attr_partscan
.attr
,
769 static struct attribute_group loop_attribute_group
= {
774 static int loop_sysfs_init(struct loop_device
*lo
)
776 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
777 &loop_attribute_group
);
780 static void loop_sysfs_exit(struct loop_device
*lo
)
782 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
783 &loop_attribute_group
);
786 static void loop_config_discard(struct loop_device
*lo
)
788 struct file
*file
= lo
->lo_backing_file
;
789 struct inode
*inode
= file
->f_mapping
->host
;
790 struct request_queue
*q
= lo
->lo_queue
;
793 * We use punch hole to reclaim the free space used by the
794 * image a.k.a. discard. However we do support discard if
795 * encryption is enabled, because it may give an attacker
796 * useful information.
798 if ((!file
->f_op
->fallocate
) ||
799 lo
->lo_encrypt_key_size
) {
800 q
->limits
.discard_granularity
= 0;
801 q
->limits
.discard_alignment
= 0;
802 q
->limits
.max_discard_sectors
= 0;
803 q
->limits
.discard_zeroes_data
= 0;
804 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
808 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
809 q
->limits
.discard_alignment
= 0;
810 q
->limits
.max_discard_sectors
= UINT_MAX
>> 9;
811 q
->limits
.discard_zeroes_data
= 1;
812 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
815 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
816 struct block_device
*bdev
, unsigned int arg
)
818 struct file
*file
, *f
;
820 struct address_space
*mapping
;
821 unsigned lo_blocksize
;
826 /* This is safe, since we have a reference from open(). */
827 __module_get(THIS_MODULE
);
835 if (lo
->lo_state
!= Lo_unbound
)
838 /* Avoid recursion */
840 while (is_loop_device(f
)) {
841 struct loop_device
*l
;
843 if (f
->f_mapping
->host
->i_bdev
== bdev
)
846 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
847 if (l
->lo_state
== Lo_unbound
) {
851 f
= l
->lo_backing_file
;
854 mapping
= file
->f_mapping
;
855 inode
= mapping
->host
;
858 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
861 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
863 lo_flags
|= LO_FLAGS_READ_ONLY
;
865 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
866 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
869 size
= get_loop_size(lo
, file
);
870 if ((loff_t
)(sector_t
)size
!= size
)
875 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
877 lo
->lo_blocksize
= lo_blocksize
;
878 lo
->lo_device
= bdev
;
879 lo
->lo_flags
= lo_flags
;
880 lo
->lo_backing_file
= file
;
881 lo
->transfer
= transfer_none
;
883 lo
->lo_sizelimit
= 0;
884 lo
->lo_bio_count
= 0;
885 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
886 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
888 bio_list_init(&lo
->lo_bio_list
);
891 * set queue make_request_fn, and add limits based on lower level
894 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
895 lo
->lo_queue
->queuedata
= lo
;
897 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
898 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
900 set_capacity(lo
->lo_disk
, size
);
901 bd_set_size(bdev
, size
<< 9);
903 /* let user-space know about the new size */
904 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
906 set_blocksize(bdev
, lo_blocksize
);
908 lo
->lo_thread
= kthread_create(loop_thread
, lo
, "loop%d",
910 if (IS_ERR(lo
->lo_thread
)) {
911 error
= PTR_ERR(lo
->lo_thread
);
914 lo
->lo_state
= Lo_bound
;
915 wake_up_process(lo
->lo_thread
);
917 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
918 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
919 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
924 lo
->lo_thread
= NULL
;
925 lo
->lo_device
= NULL
;
926 lo
->lo_backing_file
= NULL
;
928 set_capacity(lo
->lo_disk
, 0);
929 invalidate_bdev(bdev
);
930 bd_set_size(bdev
, 0);
931 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
932 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
);
933 lo
->lo_state
= Lo_unbound
;
937 /* This is safe: open() is still holding a reference. */
938 module_put(THIS_MODULE
);
943 loop_release_xfer(struct loop_device
*lo
)
946 struct loop_func_table
*xfer
= lo
->lo_encryption
;
950 err
= xfer
->release(lo
);
952 lo
->lo_encryption
= NULL
;
953 module_put(xfer
->owner
);
959 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
960 const struct loop_info64
*i
)
965 struct module
*owner
= xfer
->owner
;
967 if (!try_module_get(owner
))
970 err
= xfer
->init(lo
, i
);
974 lo
->lo_encryption
= xfer
;
979 static int loop_clr_fd(struct loop_device
*lo
)
981 struct file
*filp
= lo
->lo_backing_file
;
982 gfp_t gfp
= lo
->old_gfp_mask
;
983 struct block_device
*bdev
= lo
->lo_device
;
985 if (lo
->lo_state
!= Lo_bound
)
989 * If we've explicitly asked to tear down the loop device,
990 * and it has an elevated reference count, set it for auto-teardown when
991 * the last reference goes away. This stops $!~#$@ udev from
992 * preventing teardown because it decided that it needs to run blkid on
993 * the loopback device whenever they appear. xfstests is notorious for
994 * failing tests because blkid via udev races with a losetup
995 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
996 * command to fail with EBUSY.
998 if (lo
->lo_refcnt
> 1) {
999 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1000 mutex_unlock(&lo
->lo_ctl_mutex
);
1007 spin_lock_irq(&lo
->lo_lock
);
1008 lo
->lo_state
= Lo_rundown
;
1009 spin_unlock_irq(&lo
->lo_lock
);
1011 kthread_stop(lo
->lo_thread
);
1013 spin_lock_irq(&lo
->lo_lock
);
1014 lo
->lo_backing_file
= NULL
;
1015 spin_unlock_irq(&lo
->lo_lock
);
1017 loop_release_xfer(lo
);
1018 lo
->transfer
= NULL
;
1020 lo
->lo_device
= NULL
;
1021 lo
->lo_encryption
= NULL
;
1023 lo
->lo_sizelimit
= 0;
1024 lo
->lo_encrypt_key_size
= 0;
1025 lo
->lo_thread
= NULL
;
1026 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1027 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1028 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1030 invalidate_bdev(bdev
);
1031 set_capacity(lo
->lo_disk
, 0);
1032 loop_sysfs_exit(lo
);
1034 bd_set_size(bdev
, 0);
1035 /* let user-space know about this change */
1036 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1038 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1039 lo
->lo_state
= Lo_unbound
;
1040 /* This is safe: open() is still holding a reference. */
1041 module_put(THIS_MODULE
);
1042 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1043 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
1046 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1047 mutex_unlock(&lo
->lo_ctl_mutex
);
1049 * Need not hold lo_ctl_mutex to fput backing file.
1050 * Calling fput holding lo_ctl_mutex triggers a circular
1051 * lock dependency possibility warning as fput can take
1052 * bd_mutex which is usually taken before lo_ctl_mutex.
1059 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1062 struct loop_func_table
*xfer
;
1063 kuid_t uid
= current_uid();
1065 if (lo
->lo_encrypt_key_size
&&
1066 !uid_eq(lo
->lo_key_owner
, uid
) &&
1067 !capable(CAP_SYS_ADMIN
))
1069 if (lo
->lo_state
!= Lo_bound
)
1071 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1074 err
= loop_release_xfer(lo
);
1078 if (info
->lo_encrypt_type
) {
1079 unsigned int type
= info
->lo_encrypt_type
;
1081 if (type
>= MAX_LO_CRYPT
)
1083 xfer
= xfer_funcs
[type
];
1089 err
= loop_init_xfer(lo
, xfer
, info
);
1093 if (lo
->lo_offset
!= info
->lo_offset
||
1094 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1095 if (figure_loop_size(lo
, info
->lo_offset
, info
->lo_sizelimit
))
1098 loop_config_discard(lo
);
1100 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1101 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1102 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1103 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1107 lo
->transfer
= xfer
->transfer
;
1108 lo
->ioctl
= xfer
->ioctl
;
1110 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1111 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1112 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1114 if ((info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1115 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1116 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1117 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1118 ioctl_by_bdev(lo
->lo_device
, BLKRRPART
, 0);
1121 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1122 lo
->lo_init
[0] = info
->lo_init
[0];
1123 lo
->lo_init
[1] = info
->lo_init
[1];
1124 if (info
->lo_encrypt_key_size
) {
1125 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1126 info
->lo_encrypt_key_size
);
1127 lo
->lo_key_owner
= uid
;
1134 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1136 struct file
*file
= lo
->lo_backing_file
;
1140 if (lo
->lo_state
!= Lo_bound
)
1142 error
= vfs_getattr(file
->f_path
.mnt
, file
->f_path
.dentry
, &stat
);
1145 memset(info
, 0, sizeof(*info
));
1146 info
->lo_number
= lo
->lo_number
;
1147 info
->lo_device
= huge_encode_dev(stat
.dev
);
1148 info
->lo_inode
= stat
.ino
;
1149 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1150 info
->lo_offset
= lo
->lo_offset
;
1151 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1152 info
->lo_flags
= lo
->lo_flags
;
1153 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1154 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1155 info
->lo_encrypt_type
=
1156 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1157 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1158 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1159 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1160 lo
->lo_encrypt_key_size
);
1166 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1168 memset(info64
, 0, sizeof(*info64
));
1169 info64
->lo_number
= info
->lo_number
;
1170 info64
->lo_device
= info
->lo_device
;
1171 info64
->lo_inode
= info
->lo_inode
;
1172 info64
->lo_rdevice
= info
->lo_rdevice
;
1173 info64
->lo_offset
= info
->lo_offset
;
1174 info64
->lo_sizelimit
= 0;
1175 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1176 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1177 info64
->lo_flags
= info
->lo_flags
;
1178 info64
->lo_init
[0] = info
->lo_init
[0];
1179 info64
->lo_init
[1] = info
->lo_init
[1];
1180 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1181 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1183 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1184 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1188 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1190 memset(info
, 0, sizeof(*info
));
1191 info
->lo_number
= info64
->lo_number
;
1192 info
->lo_device
= info64
->lo_device
;
1193 info
->lo_inode
= info64
->lo_inode
;
1194 info
->lo_rdevice
= info64
->lo_rdevice
;
1195 info
->lo_offset
= info64
->lo_offset
;
1196 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1197 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1198 info
->lo_flags
= info64
->lo_flags
;
1199 info
->lo_init
[0] = info64
->lo_init
[0];
1200 info
->lo_init
[1] = info64
->lo_init
[1];
1201 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1202 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1204 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1205 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1207 /* error in case values were truncated */
1208 if (info
->lo_device
!= info64
->lo_device
||
1209 info
->lo_rdevice
!= info64
->lo_rdevice
||
1210 info
->lo_inode
!= info64
->lo_inode
||
1211 info
->lo_offset
!= info64
->lo_offset
)
1218 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1220 struct loop_info info
;
1221 struct loop_info64 info64
;
1223 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1225 loop_info64_from_old(&info
, &info64
);
1226 return loop_set_status(lo
, &info64
);
1230 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1232 struct loop_info64 info64
;
1234 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1236 return loop_set_status(lo
, &info64
);
1240 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1241 struct loop_info info
;
1242 struct loop_info64 info64
;
1248 err
= loop_get_status(lo
, &info64
);
1250 err
= loop_info64_to_old(&info64
, &info
);
1251 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1258 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1259 struct loop_info64 info64
;
1265 err
= loop_get_status(lo
, &info64
);
1266 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1272 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1279 if (unlikely(lo
->lo_state
!= Lo_bound
))
1281 err
= figure_loop_size(lo
, lo
->lo_offset
, lo
->lo_sizelimit
);
1284 sec
= get_capacity(lo
->lo_disk
);
1285 /* the width of sector_t may be narrow for bit-shift */
1288 mutex_lock(&bdev
->bd_mutex
);
1289 bd_set_size(bdev
, sz
);
1290 /* let user-space know about the new size */
1291 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1292 mutex_unlock(&bdev
->bd_mutex
);
1298 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1299 unsigned int cmd
, unsigned long arg
)
1301 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1304 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1307 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1309 case LOOP_CHANGE_FD
:
1310 err
= loop_change_fd(lo
, bdev
, arg
);
1313 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1314 err
= loop_clr_fd(lo
);
1318 case LOOP_SET_STATUS
:
1320 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1321 err
= loop_set_status_old(lo
,
1322 (struct loop_info __user
*)arg
);
1324 case LOOP_GET_STATUS
:
1325 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1327 case LOOP_SET_STATUS64
:
1329 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1330 err
= loop_set_status64(lo
,
1331 (struct loop_info64 __user
*) arg
);
1333 case LOOP_GET_STATUS64
:
1334 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1336 case LOOP_SET_CAPACITY
:
1338 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1339 err
= loop_set_capacity(lo
, bdev
);
1342 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1344 mutex_unlock(&lo
->lo_ctl_mutex
);
1350 #ifdef CONFIG_COMPAT
1351 struct compat_loop_info
{
1352 compat_int_t lo_number
; /* ioctl r/o */
1353 compat_dev_t lo_device
; /* ioctl r/o */
1354 compat_ulong_t lo_inode
; /* ioctl r/o */
1355 compat_dev_t lo_rdevice
; /* ioctl r/o */
1356 compat_int_t lo_offset
;
1357 compat_int_t lo_encrypt_type
;
1358 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1359 compat_int_t lo_flags
; /* ioctl r/o */
1360 char lo_name
[LO_NAME_SIZE
];
1361 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1362 compat_ulong_t lo_init
[2];
1367 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1368 * - noinlined to reduce stack space usage in main part of driver
1371 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1372 struct loop_info64
*info64
)
1374 struct compat_loop_info info
;
1376 if (copy_from_user(&info
, arg
, sizeof(info
)))
1379 memset(info64
, 0, sizeof(*info64
));
1380 info64
->lo_number
= info
.lo_number
;
1381 info64
->lo_device
= info
.lo_device
;
1382 info64
->lo_inode
= info
.lo_inode
;
1383 info64
->lo_rdevice
= info
.lo_rdevice
;
1384 info64
->lo_offset
= info
.lo_offset
;
1385 info64
->lo_sizelimit
= 0;
1386 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1387 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1388 info64
->lo_flags
= info
.lo_flags
;
1389 info64
->lo_init
[0] = info
.lo_init
[0];
1390 info64
->lo_init
[1] = info
.lo_init
[1];
1391 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1392 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1394 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1395 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1400 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1401 * - noinlined to reduce stack space usage in main part of driver
1404 loop_info64_to_compat(const struct loop_info64
*info64
,
1405 struct compat_loop_info __user
*arg
)
1407 struct compat_loop_info info
;
1409 memset(&info
, 0, sizeof(info
));
1410 info
.lo_number
= info64
->lo_number
;
1411 info
.lo_device
= info64
->lo_device
;
1412 info
.lo_inode
= info64
->lo_inode
;
1413 info
.lo_rdevice
= info64
->lo_rdevice
;
1414 info
.lo_offset
= info64
->lo_offset
;
1415 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1416 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1417 info
.lo_flags
= info64
->lo_flags
;
1418 info
.lo_init
[0] = info64
->lo_init
[0];
1419 info
.lo_init
[1] = info64
->lo_init
[1];
1420 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1421 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1423 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1424 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1426 /* error in case values were truncated */
1427 if (info
.lo_device
!= info64
->lo_device
||
1428 info
.lo_rdevice
!= info64
->lo_rdevice
||
1429 info
.lo_inode
!= info64
->lo_inode
||
1430 info
.lo_offset
!= info64
->lo_offset
||
1431 info
.lo_init
[0] != info64
->lo_init
[0] ||
1432 info
.lo_init
[1] != info64
->lo_init
[1])
1435 if (copy_to_user(arg
, &info
, sizeof(info
)))
1441 loop_set_status_compat(struct loop_device
*lo
,
1442 const struct compat_loop_info __user
*arg
)
1444 struct loop_info64 info64
;
1447 ret
= loop_info64_from_compat(arg
, &info64
);
1450 return loop_set_status(lo
, &info64
);
1454 loop_get_status_compat(struct loop_device
*lo
,
1455 struct compat_loop_info __user
*arg
)
1457 struct loop_info64 info64
;
1463 err
= loop_get_status(lo
, &info64
);
1465 err
= loop_info64_to_compat(&info64
, arg
);
1469 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1470 unsigned int cmd
, unsigned long arg
)
1472 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1476 case LOOP_SET_STATUS
:
1477 mutex_lock(&lo
->lo_ctl_mutex
);
1478 err
= loop_set_status_compat(
1479 lo
, (const struct compat_loop_info __user
*) arg
);
1480 mutex_unlock(&lo
->lo_ctl_mutex
);
1482 case LOOP_GET_STATUS
:
1483 mutex_lock(&lo
->lo_ctl_mutex
);
1484 err
= loop_get_status_compat(
1485 lo
, (struct compat_loop_info __user
*) arg
);
1486 mutex_unlock(&lo
->lo_ctl_mutex
);
1488 case LOOP_SET_CAPACITY
:
1490 case LOOP_GET_STATUS64
:
1491 case LOOP_SET_STATUS64
:
1492 arg
= (unsigned long) compat_ptr(arg
);
1494 case LOOP_CHANGE_FD
:
1495 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1505 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1507 struct loop_device
*lo
;
1510 mutex_lock(&loop_index_mutex
);
1511 lo
= bdev
->bd_disk
->private_data
;
1517 mutex_lock(&lo
->lo_ctl_mutex
);
1519 mutex_unlock(&lo
->lo_ctl_mutex
);
1521 mutex_unlock(&loop_index_mutex
);
1525 static int lo_release(struct gendisk
*disk
, fmode_t mode
)
1527 struct loop_device
*lo
= disk
->private_data
;
1530 mutex_lock(&lo
->lo_ctl_mutex
);
1532 if (--lo
->lo_refcnt
)
1535 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1537 * In autoclear mode, stop the loop thread
1538 * and remove configuration after last close.
1540 err
= loop_clr_fd(lo
);
1545 * Otherwise keep thread (if running) and config,
1546 * but flush possible ongoing bios in thread.
1552 mutex_unlock(&lo
->lo_ctl_mutex
);
1557 static const struct block_device_operations lo_fops
= {
1558 .owner
= THIS_MODULE
,
1560 .release
= lo_release
,
1562 #ifdef CONFIG_COMPAT
1563 .compat_ioctl
= lo_compat_ioctl
,
1568 * And now the modules code and kernel interface.
1570 static int max_loop
;
1571 module_param(max_loop
, int, S_IRUGO
);
1572 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1573 module_param(max_part
, int, S_IRUGO
);
1574 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1575 MODULE_LICENSE("GPL");
1576 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1578 int loop_register_transfer(struct loop_func_table
*funcs
)
1580 unsigned int n
= funcs
->number
;
1582 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1584 xfer_funcs
[n
] = funcs
;
1588 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1590 struct loop_device
*lo
= ptr
;
1591 struct loop_func_table
*xfer
= data
;
1593 mutex_lock(&lo
->lo_ctl_mutex
);
1594 if (lo
->lo_encryption
== xfer
)
1595 loop_release_xfer(lo
);
1596 mutex_unlock(&lo
->lo_ctl_mutex
);
1600 int loop_unregister_transfer(int number
)
1602 unsigned int n
= number
;
1603 struct loop_func_table
*xfer
;
1605 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1608 xfer_funcs
[n
] = NULL
;
1609 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1613 EXPORT_SYMBOL(loop_register_transfer
);
1614 EXPORT_SYMBOL(loop_unregister_transfer
);
1616 static int loop_add(struct loop_device
**l
, int i
)
1618 struct loop_device
*lo
;
1619 struct gendisk
*disk
;
1623 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1627 if (!idr_pre_get(&loop_index_idr
, GFP_KERNEL
))
1633 /* create specific i in the index */
1634 err
= idr_get_new_above(&loop_index_idr
, lo
, i
, &m
);
1635 if (err
>= 0 && i
!= m
) {
1636 idr_remove(&loop_index_idr
, m
);
1639 } else if (i
== -1) {
1642 /* get next free nr */
1643 err
= idr_get_new(&loop_index_idr
, lo
, &m
);
1652 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
1656 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1658 goto out_free_queue
;
1661 * Disable partition scanning by default. The in-kernel partition
1662 * scanning can be requested individually per-device during its
1663 * setup. Userspace can always add and remove partitions from all
1664 * devices. The needed partition minors are allocated from the
1665 * extended minor space, the main loop device numbers will continue
1666 * to match the loop minors, regardless of the number of partitions
1669 * If max_part is given, partition scanning is globally enabled for
1670 * all loop devices. The minors for the main loop devices will be
1671 * multiples of max_part.
1673 * Note: Global-for-all-devices, set-only-at-init, read-only module
1674 * parameteters like 'max_loop' and 'max_part' make things needlessly
1675 * complicated, are too static, inflexible and may surprise
1676 * userspace tools. Parameters like this in general should be avoided.
1679 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1680 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1681 mutex_init(&lo
->lo_ctl_mutex
);
1683 lo
->lo_thread
= NULL
;
1684 init_waitqueue_head(&lo
->lo_event
);
1685 init_waitqueue_head(&lo
->lo_req_wait
);
1686 spin_lock_init(&lo
->lo_lock
);
1687 disk
->major
= LOOP_MAJOR
;
1688 disk
->first_minor
= i
<< part_shift
;
1689 disk
->fops
= &lo_fops
;
1690 disk
->private_data
= lo
;
1691 disk
->queue
= lo
->lo_queue
;
1692 sprintf(disk
->disk_name
, "loop%d", i
);
1695 return lo
->lo_number
;
1698 blk_cleanup_queue(lo
->lo_queue
);
1705 static void loop_remove(struct loop_device
*lo
)
1707 del_gendisk(lo
->lo_disk
);
1708 blk_cleanup_queue(lo
->lo_queue
);
1709 put_disk(lo
->lo_disk
);
1713 static int find_free_cb(int id
, void *ptr
, void *data
)
1715 struct loop_device
*lo
= ptr
;
1716 struct loop_device
**l
= data
;
1718 if (lo
->lo_state
== Lo_unbound
) {
1725 static int loop_lookup(struct loop_device
**l
, int i
)
1727 struct loop_device
*lo
;
1733 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1736 ret
= lo
->lo_number
;
1741 /* lookup and return a specific i */
1742 lo
= idr_find(&loop_index_idr
, i
);
1745 ret
= lo
->lo_number
;
1751 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1753 struct loop_device
*lo
;
1754 struct kobject
*kobj
;
1757 mutex_lock(&loop_index_mutex
);
1758 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1760 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1762 kobj
= ERR_PTR(err
);
1764 kobj
= get_disk(lo
->lo_disk
);
1765 mutex_unlock(&loop_index_mutex
);
1771 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1774 struct loop_device
*lo
;
1777 mutex_lock(&loop_index_mutex
);
1780 ret
= loop_lookup(&lo
, parm
);
1785 ret
= loop_add(&lo
, parm
);
1787 case LOOP_CTL_REMOVE
:
1788 ret
= loop_lookup(&lo
, parm
);
1791 mutex_lock(&lo
->lo_ctl_mutex
);
1792 if (lo
->lo_state
!= Lo_unbound
) {
1794 mutex_unlock(&lo
->lo_ctl_mutex
);
1797 if (lo
->lo_refcnt
> 0) {
1799 mutex_unlock(&lo
->lo_ctl_mutex
);
1802 lo
->lo_disk
->private_data
= NULL
;
1803 mutex_unlock(&lo
->lo_ctl_mutex
);
1804 idr_remove(&loop_index_idr
, lo
->lo_number
);
1807 case LOOP_CTL_GET_FREE
:
1808 ret
= loop_lookup(&lo
, -1);
1811 ret
= loop_add(&lo
, -1);
1813 mutex_unlock(&loop_index_mutex
);
1818 static const struct file_operations loop_ctl_fops
= {
1819 .open
= nonseekable_open
,
1820 .unlocked_ioctl
= loop_control_ioctl
,
1821 .compat_ioctl
= loop_control_ioctl
,
1822 .owner
= THIS_MODULE
,
1823 .llseek
= noop_llseek
,
1826 static struct miscdevice loop_misc
= {
1827 .minor
= LOOP_CTRL_MINOR
,
1828 .name
= "loop-control",
1829 .fops
= &loop_ctl_fops
,
1832 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1833 MODULE_ALIAS("devname:loop-control");
1835 static int __init
loop_init(void)
1838 unsigned long range
;
1839 struct loop_device
*lo
;
1842 err
= misc_register(&loop_misc
);
1848 part_shift
= fls(max_part
);
1851 * Adjust max_part according to part_shift as it is exported
1852 * to user space so that user can decide correct minor number
1853 * if [s]he want to create more devices.
1855 * Note that -1 is required because partition 0 is reserved
1856 * for the whole disk.
1858 max_part
= (1UL << part_shift
) - 1;
1861 if ((1UL << part_shift
) > DISK_MAX_PARTS
)
1864 if (max_loop
> 1UL << (MINORBITS
- part_shift
))
1868 * If max_loop is specified, create that many devices upfront.
1869 * This also becomes a hard limit. If max_loop is not specified,
1870 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1871 * init time. Loop devices can be requested on-demand with the
1872 * /dev/loop-control interface, or be instantiated by accessing
1873 * a 'dead' device node.
1877 range
= max_loop
<< part_shift
;
1879 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
1880 range
= 1UL << MINORBITS
;
1883 if (register_blkdev(LOOP_MAJOR
, "loop"))
1886 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1887 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1889 /* pre-create number of devices given by config or max_loop */
1890 mutex_lock(&loop_index_mutex
);
1891 for (i
= 0; i
< nr
; i
++)
1893 mutex_unlock(&loop_index_mutex
);
1895 printk(KERN_INFO
"loop: module loaded\n");
1899 static int loop_exit_cb(int id
, void *ptr
, void *data
)
1901 struct loop_device
*lo
= ptr
;
1907 static void __exit
loop_exit(void)
1909 unsigned long range
;
1911 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
1913 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
1914 idr_remove_all(&loop_index_idr
);
1915 idr_destroy(&loop_index_idr
);
1917 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1918 unregister_blkdev(LOOP_MAJOR
, "loop");
1920 misc_deregister(&loop_misc
);
1923 module_init(loop_init
);
1924 module_exit(loop_exit
);
1927 static int __init
max_loop_setup(char *str
)
1929 max_loop
= simple_strtol(str
, NULL
, 0);
1933 __setup("max_loop=", max_loop_setup
);