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/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/kthread.h>
76 #include <linux/splice.h>
77 #include <linux/sysfs.h>
78 #include <linux/miscdevice.h>
79 #include <linux/falloc.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
;
92 static int transfer_none(struct loop_device
*lo
, int cmd
,
93 struct page
*raw_page
, unsigned raw_off
,
94 struct page
*loop_page
, unsigned loop_off
,
95 int size
, sector_t real_block
)
97 char *raw_buf
= kmap_atomic(raw_page
, KM_USER0
) + raw_off
;
98 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + loop_off
;
101 memcpy(loop_buf
, raw_buf
, size
);
103 memcpy(raw_buf
, loop_buf
, size
);
105 kunmap_atomic(loop_buf
, KM_USER1
);
106 kunmap_atomic(raw_buf
, KM_USER0
);
111 static int transfer_xor(struct loop_device
*lo
, int cmd
,
112 struct page
*raw_page
, unsigned raw_off
,
113 struct page
*loop_page
, unsigned loop_off
,
114 int size
, sector_t real_block
)
116 char *raw_buf
= kmap_atomic(raw_page
, KM_USER0
) + raw_off
;
117 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + loop_off
;
118 char *in
, *out
, *key
;
129 key
= lo
->lo_encrypt_key
;
130 keysize
= lo
->lo_encrypt_key_size
;
131 for (i
= 0; i
< size
; i
++)
132 *out
++ = *in
++ ^ key
[(i
& 511) % keysize
];
134 kunmap_atomic(loop_buf
, KM_USER1
);
135 kunmap_atomic(raw_buf
, KM_USER0
);
140 static int xor_init(struct loop_device
*lo
, const struct loop_info64
*info
)
142 if (unlikely(info
->lo_encrypt_key_size
<= 0))
147 static struct loop_func_table none_funcs
= {
148 .number
= LO_CRYPT_NONE
,
149 .transfer
= transfer_none
,
152 static struct loop_func_table xor_funcs
= {
153 .number
= LO_CRYPT_XOR
,
154 .transfer
= transfer_xor
,
158 /* xfer_funcs[0] is special - its release function is never called */
159 static struct loop_func_table
*xfer_funcs
[MAX_LO_CRYPT
] = {
164 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
166 loff_t size
, offset
, loopsize
;
168 /* Compute loopsize in bytes */
169 size
= i_size_read(file
->f_mapping
->host
);
170 offset
= lo
->lo_offset
;
171 loopsize
= size
- offset
;
172 if (lo
->lo_sizelimit
> 0 && lo
->lo_sizelimit
< loopsize
)
173 loopsize
= lo
->lo_sizelimit
;
176 * Unfortunately, if we want to do I/O on the device,
177 * the number of 512-byte sectors has to fit into a sector_t.
179 return loopsize
>> 9;
183 figure_loop_size(struct loop_device
*lo
)
185 loff_t size
= get_loop_size(lo
, lo
->lo_backing_file
);
186 sector_t x
= (sector_t
)size
;
188 if (unlikely((loff_t
)x
!= size
))
191 set_capacity(lo
->lo_disk
, x
);
196 lo_do_transfer(struct loop_device
*lo
, int cmd
,
197 struct page
*rpage
, unsigned roffs
,
198 struct page
*lpage
, unsigned loffs
,
199 int size
, sector_t rblock
)
201 if (unlikely(!lo
->transfer
))
204 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
208 * __do_lo_send_write - helper for writing data to a loop device
210 * This helper just factors out common code between do_lo_send_direct_write()
211 * and do_lo_send_write().
213 static int __do_lo_send_write(struct file
*file
,
214 u8
*buf
, const int len
, loff_t pos
)
217 mm_segment_t old_fs
= get_fs();
220 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
222 if (likely(bw
== len
))
224 printk(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
225 (unsigned long long)pos
, len
);
232 * do_lo_send_direct_write - helper for writing data to a loop device
234 * This is the fast, non-transforming version that does not need double
237 static int do_lo_send_direct_write(struct loop_device
*lo
,
238 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
240 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
241 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
243 kunmap(bvec
->bv_page
);
249 * do_lo_send_write - helper for writing data to a loop device
251 * This is the slow, transforming version that needs to double buffer the
252 * data as it cannot do the transformations in place without having direct
253 * access to the destination pages of the backing file.
255 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
256 loff_t pos
, struct page
*page
)
258 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
259 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
261 return __do_lo_send_write(lo
->lo_backing_file
,
262 page_address(page
), bvec
->bv_len
,
264 printk(KERN_ERR
"loop: Transfer error at byte offset %llu, "
265 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
271 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
273 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
275 struct bio_vec
*bvec
;
276 struct page
*page
= NULL
;
279 if (lo
->transfer
!= transfer_none
) {
280 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
284 do_lo_send
= do_lo_send_write
;
286 do_lo_send
= do_lo_send_direct_write
;
289 bio_for_each_segment(bvec
, bio
, i
) {
290 ret
= do_lo_send(lo
, bvec
, pos
, page
);
302 printk(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
307 struct lo_read_data
{
308 struct loop_device
*lo
;
315 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
316 struct splice_desc
*sd
)
318 struct lo_read_data
*p
= sd
->u
.data
;
319 struct loop_device
*lo
= p
->lo
;
320 struct page
*page
= buf
->page
;
324 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
330 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
331 printk(KERN_ERR
"loop: transfer error block %ld\n",
336 flush_dcache_page(p
->page
);
345 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
347 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
351 do_lo_receive(struct loop_device
*lo
,
352 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
354 struct lo_read_data cookie
;
355 struct splice_desc sd
;
360 cookie
.page
= bvec
->bv_page
;
361 cookie
.offset
= bvec
->bv_offset
;
362 cookie
.bsize
= bsize
;
365 sd
.total_len
= bvec
->bv_len
;
370 file
= lo
->lo_backing_file
;
371 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
380 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
382 struct bio_vec
*bvec
;
385 bio_for_each_segment(bvec
, bio
, i
) {
386 ret
= do_lo_receive(lo
, bvec
, bsize
, pos
);
394 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
399 pos
= ((loff_t
) bio
->bi_sector
<< 9) + lo
->lo_offset
;
401 if (bio_rw(bio
) == WRITE
) {
402 struct file
*file
= lo
->lo_backing_file
;
404 if (bio
->bi_rw
& REQ_FLUSH
) {
405 ret
= vfs_fsync(file
, 0);
406 if (unlikely(ret
&& ret
!= -EINVAL
)) {
413 * We use punch hole to reclaim the free space used by the
414 * image a.k.a. discard. However we do support discard if
415 * encryption is enabled, because it may give an attacker
416 * useful information.
418 if (bio
->bi_rw
& REQ_DISCARD
) {
419 struct file
*file
= lo
->lo_backing_file
;
420 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
422 if ((!file
->f_op
->fallocate
) ||
423 lo
->lo_encrypt_key_size
) {
427 ret
= file
->f_op
->fallocate(file
, mode
, pos
,
429 if (unlikely(ret
&& ret
!= -EINVAL
&&
435 ret
= lo_send(lo
, bio
, pos
);
437 if ((bio
->bi_rw
& REQ_FUA
) && !ret
) {
438 ret
= vfs_fsync(file
, 0);
439 if (unlikely(ret
&& ret
!= -EINVAL
))
443 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
450 * Add bio to back of pending list
452 static void loop_add_bio(struct loop_device
*lo
, struct bio
*bio
)
454 bio_list_add(&lo
->lo_bio_list
, bio
);
458 * Grab first pending buffer
460 static struct bio
*loop_get_bio(struct loop_device
*lo
)
462 return bio_list_pop(&lo
->lo_bio_list
);
465 static void loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
467 struct loop_device
*lo
= q
->queuedata
;
468 int rw
= bio_rw(old_bio
);
473 BUG_ON(!lo
|| (rw
!= READ
&& rw
!= WRITE
));
475 spin_lock_irq(&lo
->lo_lock
);
476 if (lo
->lo_state
!= Lo_bound
)
478 if (unlikely(rw
== WRITE
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)))
480 loop_add_bio(lo
, old_bio
);
481 wake_up(&lo
->lo_event
);
482 spin_unlock_irq(&lo
->lo_lock
);
486 spin_unlock_irq(&lo
->lo_lock
);
487 bio_io_error(old_bio
);
490 struct switch_request
{
492 struct completion wait
;
495 static void do_loop_switch(struct loop_device
*, struct switch_request
*);
497 static inline void loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
499 if (unlikely(!bio
->bi_bdev
)) {
500 do_loop_switch(lo
, bio
->bi_private
);
503 int ret
= do_bio_filebacked(lo
, bio
);
509 * worker thread that handles reads/writes to file backed loop devices,
510 * to avoid blocking in our make_request_fn. it also does loop decrypting
511 * on reads for block backed loop, as that is too heavy to do from
512 * b_end_io context where irqs may be disabled.
514 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
515 * calling kthread_stop(). Therefore once kthread_should_stop() is
516 * true, make_request will not place any more requests. Therefore
517 * once kthread_should_stop() is true and lo_bio is NULL, we are
518 * done with the loop.
520 static int loop_thread(void *data
)
522 struct loop_device
*lo
= data
;
525 set_user_nice(current
, -20);
527 while (!kthread_should_stop() || !bio_list_empty(&lo
->lo_bio_list
)) {
529 wait_event_interruptible(lo
->lo_event
,
530 !bio_list_empty(&lo
->lo_bio_list
) ||
531 kthread_should_stop());
533 if (bio_list_empty(&lo
->lo_bio_list
))
535 spin_lock_irq(&lo
->lo_lock
);
536 bio
= loop_get_bio(lo
);
537 spin_unlock_irq(&lo
->lo_lock
);
540 loop_handle_bio(lo
, bio
);
547 * loop_switch performs the hard work of switching a backing store.
548 * First it needs to flush existing IO, it does this by sending a magic
549 * BIO down the pipe. The completion of this BIO does the actual switch.
551 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
553 struct switch_request w
;
554 struct bio
*bio
= bio_alloc(GFP_KERNEL
, 0);
557 init_completion(&w
.wait
);
559 bio
->bi_private
= &w
;
561 loop_make_request(lo
->lo_queue
, bio
);
562 wait_for_completion(&w
.wait
);
567 * Helper to flush the IOs in loop, but keeping loop thread running
569 static int loop_flush(struct loop_device
*lo
)
571 /* loop not yet configured, no running thread, nothing to flush */
575 return loop_switch(lo
, NULL
);
579 * Do the actual switch; called from the BIO completion routine
581 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
583 struct file
*file
= p
->file
;
584 struct file
*old_file
= lo
->lo_backing_file
;
585 struct address_space
*mapping
;
587 /* if no new file, only flush of queued bios requested */
591 mapping
= file
->f_mapping
;
592 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
593 lo
->lo_backing_file
= file
;
594 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
595 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
596 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
597 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
604 * loop_change_fd switched the backing store of a loopback device to
605 * a new file. This is useful for operating system installers to free up
606 * the original file and in High Availability environments to switch to
607 * an alternative location for the content in case of server meltdown.
608 * This can only work if the loop device is used read-only, and if the
609 * new backing store is the same size and type as the old backing store.
611 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
614 struct file
*file
, *old_file
;
619 if (lo
->lo_state
!= Lo_bound
)
622 /* the loop device has to be read-only */
624 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
632 inode
= file
->f_mapping
->host
;
633 old_file
= lo
->lo_backing_file
;
637 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
640 /* size of the new backing store needs to be the same */
641 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
645 error
= loop_switch(lo
, file
);
650 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
651 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
660 static inline int is_loop_device(struct file
*file
)
662 struct inode
*i
= file
->f_mapping
->host
;
664 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
667 /* loop sysfs attributes */
669 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
670 ssize_t (*callback
)(struct loop_device
*, char *))
672 struct gendisk
*disk
= dev_to_disk(dev
);
673 struct loop_device
*lo
= disk
->private_data
;
675 return callback(lo
, page
);
678 #define LOOP_ATTR_RO(_name) \
679 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
680 static ssize_t loop_attr_do_show_##_name(struct device *d, \
681 struct device_attribute *attr, char *b) \
683 return loop_attr_show(d, b, loop_attr_##_name##_show); \
685 static struct device_attribute loop_attr_##_name = \
686 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
688 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
693 spin_lock_irq(&lo
->lo_lock
);
694 if (lo
->lo_backing_file
)
695 p
= d_path(&lo
->lo_backing_file
->f_path
, buf
, PAGE_SIZE
- 1);
696 spin_unlock_irq(&lo
->lo_lock
);
698 if (IS_ERR_OR_NULL(p
))
702 memmove(buf
, p
, ret
);
710 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
712 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
715 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
717 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
720 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
722 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
724 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
727 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
729 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
731 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
734 LOOP_ATTR_RO(backing_file
);
735 LOOP_ATTR_RO(offset
);
736 LOOP_ATTR_RO(sizelimit
);
737 LOOP_ATTR_RO(autoclear
);
738 LOOP_ATTR_RO(partscan
);
740 static struct attribute
*loop_attrs
[] = {
741 &loop_attr_backing_file
.attr
,
742 &loop_attr_offset
.attr
,
743 &loop_attr_sizelimit
.attr
,
744 &loop_attr_autoclear
.attr
,
745 &loop_attr_partscan
.attr
,
749 static struct attribute_group loop_attribute_group
= {
754 static int loop_sysfs_init(struct loop_device
*lo
)
756 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
757 &loop_attribute_group
);
760 static void loop_sysfs_exit(struct loop_device
*lo
)
762 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
763 &loop_attribute_group
);
766 static void loop_config_discard(struct loop_device
*lo
)
768 struct file
*file
= lo
->lo_backing_file
;
769 struct inode
*inode
= file
->f_mapping
->host
;
770 struct request_queue
*q
= lo
->lo_queue
;
773 * We use punch hole to reclaim the free space used by the
774 * image a.k.a. discard. However we do support discard if
775 * encryption is enabled, because it may give an attacker
776 * useful information.
778 if ((!file
->f_op
->fallocate
) ||
779 lo
->lo_encrypt_key_size
) {
780 q
->limits
.discard_granularity
= 0;
781 q
->limits
.discard_alignment
= 0;
782 q
->limits
.max_discard_sectors
= 0;
783 q
->limits
.discard_zeroes_data
= 0;
784 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
788 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
789 q
->limits
.discard_alignment
= inode
->i_sb
->s_blocksize
;
790 q
->limits
.max_discard_sectors
= UINT_MAX
>> 9;
791 q
->limits
.discard_zeroes_data
= 1;
792 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
795 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
796 struct block_device
*bdev
, unsigned int arg
)
798 struct file
*file
, *f
;
800 struct address_space
*mapping
;
801 unsigned lo_blocksize
;
806 /* This is safe, since we have a reference from open(). */
807 __module_get(THIS_MODULE
);
815 if (lo
->lo_state
!= Lo_unbound
)
818 /* Avoid recursion */
820 while (is_loop_device(f
)) {
821 struct loop_device
*l
;
823 if (f
->f_mapping
->host
->i_bdev
== bdev
)
826 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
827 if (l
->lo_state
== Lo_unbound
) {
831 f
= l
->lo_backing_file
;
834 mapping
= file
->f_mapping
;
835 inode
= mapping
->host
;
838 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
841 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& FMODE_WRITE
) ||
843 lo_flags
|= LO_FLAGS_READ_ONLY
;
845 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
846 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
849 size
= get_loop_size(lo
, file
);
850 if ((loff_t
)(sector_t
)size
!= size
)
855 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
857 lo
->lo_blocksize
= lo_blocksize
;
858 lo
->lo_device
= bdev
;
859 lo
->lo_flags
= lo_flags
;
860 lo
->lo_backing_file
= file
;
861 lo
->transfer
= transfer_none
;
863 lo
->lo_sizelimit
= 0;
864 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
865 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
867 bio_list_init(&lo
->lo_bio_list
);
870 * set queue make_request_fn, and add limits based on lower level
873 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
874 lo
->lo_queue
->queuedata
= lo
;
876 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
877 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
879 set_capacity(lo
->lo_disk
, size
);
880 bd_set_size(bdev
, size
<< 9);
882 /* let user-space know about the new size */
883 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
885 set_blocksize(bdev
, lo_blocksize
);
887 lo
->lo_thread
= kthread_create(loop_thread
, lo
, "loop%d",
889 if (IS_ERR(lo
->lo_thread
)) {
890 error
= PTR_ERR(lo
->lo_thread
);
893 lo
->lo_state
= Lo_bound
;
894 wake_up_process(lo
->lo_thread
);
896 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
897 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
898 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
903 lo
->lo_thread
= NULL
;
904 lo
->lo_device
= NULL
;
905 lo
->lo_backing_file
= NULL
;
907 set_capacity(lo
->lo_disk
, 0);
908 invalidate_bdev(bdev
);
909 bd_set_size(bdev
, 0);
910 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
911 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
);
912 lo
->lo_state
= Lo_unbound
;
916 /* This is safe: open() is still holding a reference. */
917 module_put(THIS_MODULE
);
922 loop_release_xfer(struct loop_device
*lo
)
925 struct loop_func_table
*xfer
= lo
->lo_encryption
;
929 err
= xfer
->release(lo
);
931 lo
->lo_encryption
= NULL
;
932 module_put(xfer
->owner
);
938 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
939 const struct loop_info64
*i
)
944 struct module
*owner
= xfer
->owner
;
946 if (!try_module_get(owner
))
949 err
= xfer
->init(lo
, i
);
953 lo
->lo_encryption
= xfer
;
958 static int loop_clr_fd(struct loop_device
*lo
)
960 struct file
*filp
= lo
->lo_backing_file
;
961 gfp_t gfp
= lo
->old_gfp_mask
;
962 struct block_device
*bdev
= lo
->lo_device
;
964 if (lo
->lo_state
!= Lo_bound
)
967 if (lo
->lo_refcnt
> 1) /* we needed one fd for the ioctl */
973 spin_lock_irq(&lo
->lo_lock
);
974 lo
->lo_state
= Lo_rundown
;
975 spin_unlock_irq(&lo
->lo_lock
);
977 kthread_stop(lo
->lo_thread
);
979 spin_lock_irq(&lo
->lo_lock
);
980 lo
->lo_backing_file
= NULL
;
981 spin_unlock_irq(&lo
->lo_lock
);
983 loop_release_xfer(lo
);
986 lo
->lo_device
= NULL
;
987 lo
->lo_encryption
= NULL
;
989 lo
->lo_sizelimit
= 0;
990 lo
->lo_encrypt_key_size
= 0;
991 lo
->lo_thread
= NULL
;
992 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
993 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
994 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
996 invalidate_bdev(bdev
);
997 set_capacity(lo
->lo_disk
, 0);
1000 bd_set_size(bdev
, 0);
1001 /* let user-space know about this change */
1002 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1004 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1005 lo
->lo_state
= Lo_unbound
;
1006 /* This is safe: open() is still holding a reference. */
1007 module_put(THIS_MODULE
);
1008 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1009 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
1012 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1013 mutex_unlock(&lo
->lo_ctl_mutex
);
1015 * Need not hold lo_ctl_mutex to fput backing file.
1016 * Calling fput holding lo_ctl_mutex triggers a circular
1017 * lock dependency possibility warning as fput can take
1018 * bd_mutex which is usually taken before lo_ctl_mutex.
1025 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1028 struct loop_func_table
*xfer
;
1029 uid_t uid
= current_uid();
1031 if (lo
->lo_encrypt_key_size
&&
1032 lo
->lo_key_owner
!= uid
&&
1033 !capable(CAP_SYS_ADMIN
))
1035 if (lo
->lo_state
!= Lo_bound
)
1037 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1040 err
= loop_release_xfer(lo
);
1044 if (info
->lo_encrypt_type
) {
1045 unsigned int type
= info
->lo_encrypt_type
;
1047 if (type
>= MAX_LO_CRYPT
)
1049 xfer
= xfer_funcs
[type
];
1055 err
= loop_init_xfer(lo
, xfer
, info
);
1059 if (lo
->lo_offset
!= info
->lo_offset
||
1060 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1061 lo
->lo_offset
= info
->lo_offset
;
1062 lo
->lo_sizelimit
= info
->lo_sizelimit
;
1063 if (figure_loop_size(lo
))
1066 loop_config_discard(lo
);
1068 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1069 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1070 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1071 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1075 lo
->transfer
= xfer
->transfer
;
1076 lo
->ioctl
= xfer
->ioctl
;
1078 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1079 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1080 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1082 if ((info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1083 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1084 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1085 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1086 ioctl_by_bdev(lo
->lo_device
, BLKRRPART
, 0);
1089 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1090 lo
->lo_init
[0] = info
->lo_init
[0];
1091 lo
->lo_init
[1] = info
->lo_init
[1];
1092 if (info
->lo_encrypt_key_size
) {
1093 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1094 info
->lo_encrypt_key_size
);
1095 lo
->lo_key_owner
= uid
;
1102 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1104 struct file
*file
= lo
->lo_backing_file
;
1108 if (lo
->lo_state
!= Lo_bound
)
1110 error
= vfs_getattr(file
->f_path
.mnt
, file
->f_path
.dentry
, &stat
);
1113 memset(info
, 0, sizeof(*info
));
1114 info
->lo_number
= lo
->lo_number
;
1115 info
->lo_device
= huge_encode_dev(stat
.dev
);
1116 info
->lo_inode
= stat
.ino
;
1117 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1118 info
->lo_offset
= lo
->lo_offset
;
1119 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1120 info
->lo_flags
= lo
->lo_flags
;
1121 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1122 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1123 info
->lo_encrypt_type
=
1124 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1125 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1126 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1127 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1128 lo
->lo_encrypt_key_size
);
1134 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1136 memset(info64
, 0, sizeof(*info64
));
1137 info64
->lo_number
= info
->lo_number
;
1138 info64
->lo_device
= info
->lo_device
;
1139 info64
->lo_inode
= info
->lo_inode
;
1140 info64
->lo_rdevice
= info
->lo_rdevice
;
1141 info64
->lo_offset
= info
->lo_offset
;
1142 info64
->lo_sizelimit
= 0;
1143 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1144 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1145 info64
->lo_flags
= info
->lo_flags
;
1146 info64
->lo_init
[0] = info
->lo_init
[0];
1147 info64
->lo_init
[1] = info
->lo_init
[1];
1148 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1149 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1151 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1152 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1156 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1158 memset(info
, 0, sizeof(*info
));
1159 info
->lo_number
= info64
->lo_number
;
1160 info
->lo_device
= info64
->lo_device
;
1161 info
->lo_inode
= info64
->lo_inode
;
1162 info
->lo_rdevice
= info64
->lo_rdevice
;
1163 info
->lo_offset
= info64
->lo_offset
;
1164 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1165 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1166 info
->lo_flags
= info64
->lo_flags
;
1167 info
->lo_init
[0] = info64
->lo_init
[0];
1168 info
->lo_init
[1] = info64
->lo_init
[1];
1169 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1170 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1172 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1173 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1175 /* error in case values were truncated */
1176 if (info
->lo_device
!= info64
->lo_device
||
1177 info
->lo_rdevice
!= info64
->lo_rdevice
||
1178 info
->lo_inode
!= info64
->lo_inode
||
1179 info
->lo_offset
!= info64
->lo_offset
)
1186 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1188 struct loop_info info
;
1189 struct loop_info64 info64
;
1191 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1193 loop_info64_from_old(&info
, &info64
);
1194 return loop_set_status(lo
, &info64
);
1198 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1200 struct loop_info64 info64
;
1202 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1204 return loop_set_status(lo
, &info64
);
1208 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1209 struct loop_info info
;
1210 struct loop_info64 info64
;
1216 err
= loop_get_status(lo
, &info64
);
1218 err
= loop_info64_to_old(&info64
, &info
);
1219 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1226 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1227 struct loop_info64 info64
;
1233 err
= loop_get_status(lo
, &info64
);
1234 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1240 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1247 if (unlikely(lo
->lo_state
!= Lo_bound
))
1249 err
= figure_loop_size(lo
);
1252 sec
= get_capacity(lo
->lo_disk
);
1253 /* the width of sector_t may be narrow for bit-shift */
1256 mutex_lock(&bdev
->bd_mutex
);
1257 bd_set_size(bdev
, sz
);
1258 /* let user-space know about the new size */
1259 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1260 mutex_unlock(&bdev
->bd_mutex
);
1266 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1267 unsigned int cmd
, unsigned long arg
)
1269 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1272 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1275 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1277 case LOOP_CHANGE_FD
:
1278 err
= loop_change_fd(lo
, bdev
, arg
);
1281 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1282 err
= loop_clr_fd(lo
);
1286 case LOOP_SET_STATUS
:
1287 err
= loop_set_status_old(lo
, (struct loop_info __user
*) arg
);
1289 case LOOP_GET_STATUS
:
1290 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1292 case LOOP_SET_STATUS64
:
1293 err
= loop_set_status64(lo
, (struct loop_info64 __user
*) arg
);
1295 case LOOP_GET_STATUS64
:
1296 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1298 case LOOP_SET_CAPACITY
:
1300 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1301 err
= loop_set_capacity(lo
, bdev
);
1304 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1306 mutex_unlock(&lo
->lo_ctl_mutex
);
1312 #ifdef CONFIG_COMPAT
1313 struct compat_loop_info
{
1314 compat_int_t lo_number
; /* ioctl r/o */
1315 compat_dev_t lo_device
; /* ioctl r/o */
1316 compat_ulong_t lo_inode
; /* ioctl r/o */
1317 compat_dev_t lo_rdevice
; /* ioctl r/o */
1318 compat_int_t lo_offset
;
1319 compat_int_t lo_encrypt_type
;
1320 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1321 compat_int_t lo_flags
; /* ioctl r/o */
1322 char lo_name
[LO_NAME_SIZE
];
1323 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1324 compat_ulong_t lo_init
[2];
1329 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1330 * - noinlined to reduce stack space usage in main part of driver
1333 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1334 struct loop_info64
*info64
)
1336 struct compat_loop_info info
;
1338 if (copy_from_user(&info
, arg
, sizeof(info
)))
1341 memset(info64
, 0, sizeof(*info64
));
1342 info64
->lo_number
= info
.lo_number
;
1343 info64
->lo_device
= info
.lo_device
;
1344 info64
->lo_inode
= info
.lo_inode
;
1345 info64
->lo_rdevice
= info
.lo_rdevice
;
1346 info64
->lo_offset
= info
.lo_offset
;
1347 info64
->lo_sizelimit
= 0;
1348 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1349 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1350 info64
->lo_flags
= info
.lo_flags
;
1351 info64
->lo_init
[0] = info
.lo_init
[0];
1352 info64
->lo_init
[1] = info
.lo_init
[1];
1353 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1354 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1356 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1357 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1362 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1363 * - noinlined to reduce stack space usage in main part of driver
1366 loop_info64_to_compat(const struct loop_info64
*info64
,
1367 struct compat_loop_info __user
*arg
)
1369 struct compat_loop_info info
;
1371 memset(&info
, 0, sizeof(info
));
1372 info
.lo_number
= info64
->lo_number
;
1373 info
.lo_device
= info64
->lo_device
;
1374 info
.lo_inode
= info64
->lo_inode
;
1375 info
.lo_rdevice
= info64
->lo_rdevice
;
1376 info
.lo_offset
= info64
->lo_offset
;
1377 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1378 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1379 info
.lo_flags
= info64
->lo_flags
;
1380 info
.lo_init
[0] = info64
->lo_init
[0];
1381 info
.lo_init
[1] = info64
->lo_init
[1];
1382 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1383 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1385 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1386 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1388 /* error in case values were truncated */
1389 if (info
.lo_device
!= info64
->lo_device
||
1390 info
.lo_rdevice
!= info64
->lo_rdevice
||
1391 info
.lo_inode
!= info64
->lo_inode
||
1392 info
.lo_offset
!= info64
->lo_offset
||
1393 info
.lo_init
[0] != info64
->lo_init
[0] ||
1394 info
.lo_init
[1] != info64
->lo_init
[1])
1397 if (copy_to_user(arg
, &info
, sizeof(info
)))
1403 loop_set_status_compat(struct loop_device
*lo
,
1404 const struct compat_loop_info __user
*arg
)
1406 struct loop_info64 info64
;
1409 ret
= loop_info64_from_compat(arg
, &info64
);
1412 return loop_set_status(lo
, &info64
);
1416 loop_get_status_compat(struct loop_device
*lo
,
1417 struct compat_loop_info __user
*arg
)
1419 struct loop_info64 info64
;
1425 err
= loop_get_status(lo
, &info64
);
1427 err
= loop_info64_to_compat(&info64
, arg
);
1431 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1432 unsigned int cmd
, unsigned long arg
)
1434 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1438 case LOOP_SET_STATUS
:
1439 mutex_lock(&lo
->lo_ctl_mutex
);
1440 err
= loop_set_status_compat(
1441 lo
, (const struct compat_loop_info __user
*) arg
);
1442 mutex_unlock(&lo
->lo_ctl_mutex
);
1444 case LOOP_GET_STATUS
:
1445 mutex_lock(&lo
->lo_ctl_mutex
);
1446 err
= loop_get_status_compat(
1447 lo
, (struct compat_loop_info __user
*) arg
);
1448 mutex_unlock(&lo
->lo_ctl_mutex
);
1450 case LOOP_SET_CAPACITY
:
1452 case LOOP_GET_STATUS64
:
1453 case LOOP_SET_STATUS64
:
1454 arg
= (unsigned long) compat_ptr(arg
);
1456 case LOOP_CHANGE_FD
:
1457 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1467 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1469 struct loop_device
*lo
;
1472 mutex_lock(&loop_index_mutex
);
1473 lo
= bdev
->bd_disk
->private_data
;
1479 mutex_lock(&lo
->lo_ctl_mutex
);
1481 mutex_unlock(&lo
->lo_ctl_mutex
);
1483 mutex_unlock(&loop_index_mutex
);
1487 static int lo_release(struct gendisk
*disk
, fmode_t mode
)
1489 struct loop_device
*lo
= disk
->private_data
;
1492 mutex_lock(&lo
->lo_ctl_mutex
);
1494 if (--lo
->lo_refcnt
)
1497 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1499 * In autoclear mode, stop the loop thread
1500 * and remove configuration after last close.
1502 err
= loop_clr_fd(lo
);
1507 * Otherwise keep thread (if running) and config,
1508 * but flush possible ongoing bios in thread.
1514 mutex_unlock(&lo
->lo_ctl_mutex
);
1519 static const struct block_device_operations lo_fops
= {
1520 .owner
= THIS_MODULE
,
1522 .release
= lo_release
,
1524 #ifdef CONFIG_COMPAT
1525 .compat_ioctl
= lo_compat_ioctl
,
1530 * And now the modules code and kernel interface.
1532 static int max_loop
;
1533 module_param(max_loop
, int, S_IRUGO
);
1534 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1535 module_param(max_part
, int, S_IRUGO
);
1536 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1537 MODULE_LICENSE("GPL");
1538 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1540 int loop_register_transfer(struct loop_func_table
*funcs
)
1542 unsigned int n
= funcs
->number
;
1544 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1546 xfer_funcs
[n
] = funcs
;
1550 static int unregister_transfer_cb(int id
, void *ptr
, void *data
)
1552 struct loop_device
*lo
= ptr
;
1553 struct loop_func_table
*xfer
= data
;
1555 mutex_lock(&lo
->lo_ctl_mutex
);
1556 if (lo
->lo_encryption
== xfer
)
1557 loop_release_xfer(lo
);
1558 mutex_unlock(&lo
->lo_ctl_mutex
);
1562 int loop_unregister_transfer(int number
)
1564 unsigned int n
= number
;
1565 struct loop_func_table
*xfer
;
1567 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1570 xfer_funcs
[n
] = NULL
;
1571 idr_for_each(&loop_index_idr
, &unregister_transfer_cb
, xfer
);
1575 EXPORT_SYMBOL(loop_register_transfer
);
1576 EXPORT_SYMBOL(loop_unregister_transfer
);
1578 static int loop_add(struct loop_device
**l
, int i
)
1580 struct loop_device
*lo
;
1581 struct gendisk
*disk
;
1584 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1590 err
= idr_pre_get(&loop_index_idr
, GFP_KERNEL
);
1597 /* create specific i in the index */
1598 err
= idr_get_new_above(&loop_index_idr
, lo
, i
, &m
);
1599 if (err
>= 0 && i
!= m
) {
1600 idr_remove(&loop_index_idr
, m
);
1603 } else if (i
== -1) {
1606 /* get next free nr */
1607 err
= idr_get_new(&loop_index_idr
, lo
, &m
);
1616 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
1620 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1622 goto out_free_queue
;
1625 * Disable partition scanning by default. The in-kernel partition
1626 * scanning can be requested individually per-device during its
1627 * setup. Userspace can always add and remove partitions from all
1628 * devices. The needed partition minors are allocated from the
1629 * extended minor space, the main loop device numbers will continue
1630 * to match the loop minors, regardless of the number of partitions
1633 * If max_part is given, partition scanning is globally enabled for
1634 * all loop devices. The minors for the main loop devices will be
1635 * multiples of max_part.
1637 * Note: Global-for-all-devices, set-only-at-init, read-only module
1638 * parameteters like 'max_loop' and 'max_part' make things needlessly
1639 * complicated, are too static, inflexible and may surprise
1640 * userspace tools. Parameters like this in general should be avoided.
1643 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1644 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1645 mutex_init(&lo
->lo_ctl_mutex
);
1647 lo
->lo_thread
= NULL
;
1648 init_waitqueue_head(&lo
->lo_event
);
1649 spin_lock_init(&lo
->lo_lock
);
1650 disk
->major
= LOOP_MAJOR
;
1651 disk
->first_minor
= i
<< part_shift
;
1652 disk
->fops
= &lo_fops
;
1653 disk
->private_data
= lo
;
1654 disk
->queue
= lo
->lo_queue
;
1655 sprintf(disk
->disk_name
, "loop%d", i
);
1658 return lo
->lo_number
;
1661 blk_cleanup_queue(lo
->lo_queue
);
1668 static void loop_remove(struct loop_device
*lo
)
1670 del_gendisk(lo
->lo_disk
);
1671 blk_cleanup_queue(lo
->lo_queue
);
1672 put_disk(lo
->lo_disk
);
1676 static int find_free_cb(int id
, void *ptr
, void *data
)
1678 struct loop_device
*lo
= ptr
;
1679 struct loop_device
**l
= data
;
1681 if (lo
->lo_state
== Lo_unbound
) {
1688 static int loop_lookup(struct loop_device
**l
, int i
)
1690 struct loop_device
*lo
;
1696 err
= idr_for_each(&loop_index_idr
, &find_free_cb
, &lo
);
1699 ret
= lo
->lo_number
;
1704 /* lookup and return a specific i */
1705 lo
= idr_find(&loop_index_idr
, i
);
1708 ret
= lo
->lo_number
;
1714 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1716 struct loop_device
*lo
;
1717 struct kobject
*kobj
;
1720 mutex_lock(&loop_index_mutex
);
1721 err
= loop_lookup(&lo
, MINOR(dev
) >> part_shift
);
1723 err
= loop_add(&lo
, MINOR(dev
) >> part_shift
);
1725 kobj
= ERR_PTR(err
);
1727 kobj
= get_disk(lo
->lo_disk
);
1728 mutex_unlock(&loop_index_mutex
);
1734 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
1737 struct loop_device
*lo
;
1740 mutex_lock(&loop_index_mutex
);
1743 ret
= loop_lookup(&lo
, parm
);
1748 ret
= loop_add(&lo
, parm
);
1750 case LOOP_CTL_REMOVE
:
1751 ret
= loop_lookup(&lo
, parm
);
1754 mutex_lock(&lo
->lo_ctl_mutex
);
1755 if (lo
->lo_state
!= Lo_unbound
) {
1757 mutex_unlock(&lo
->lo_ctl_mutex
);
1760 if (lo
->lo_refcnt
> 0) {
1762 mutex_unlock(&lo
->lo_ctl_mutex
);
1765 lo
->lo_disk
->private_data
= NULL
;
1766 mutex_unlock(&lo
->lo_ctl_mutex
);
1767 idr_remove(&loop_index_idr
, lo
->lo_number
);
1770 case LOOP_CTL_GET_FREE
:
1771 ret
= loop_lookup(&lo
, -1);
1774 ret
= loop_add(&lo
, -1);
1776 mutex_unlock(&loop_index_mutex
);
1781 static const struct file_operations loop_ctl_fops
= {
1782 .open
= nonseekable_open
,
1783 .unlocked_ioctl
= loop_control_ioctl
,
1784 .compat_ioctl
= loop_control_ioctl
,
1785 .owner
= THIS_MODULE
,
1786 .llseek
= noop_llseek
,
1789 static struct miscdevice loop_misc
= {
1790 .minor
= LOOP_CTRL_MINOR
,
1791 .name
= "loop-control",
1792 .fops
= &loop_ctl_fops
,
1795 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
1796 MODULE_ALIAS("devname:loop-control");
1798 static int __init
loop_init(void)
1801 unsigned long range
;
1802 struct loop_device
*lo
;
1805 err
= misc_register(&loop_misc
);
1811 part_shift
= fls(max_part
);
1814 * Adjust max_part according to part_shift as it is exported
1815 * to user space so that user can decide correct minor number
1816 * if [s]he want to create more devices.
1818 * Note that -1 is required because partition 0 is reserved
1819 * for the whole disk.
1821 max_part
= (1UL << part_shift
) - 1;
1824 if ((1UL << part_shift
) > DISK_MAX_PARTS
)
1827 if (max_loop
> 1UL << (MINORBITS
- part_shift
))
1831 * If max_loop is specified, create that many devices upfront.
1832 * This also becomes a hard limit. If max_loop is not specified,
1833 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1834 * init time. Loop devices can be requested on-demand with the
1835 * /dev/loop-control interface, or be instantiated by accessing
1836 * a 'dead' device node.
1840 range
= max_loop
<< part_shift
;
1842 nr
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
1843 range
= 1UL << MINORBITS
;
1846 if (register_blkdev(LOOP_MAJOR
, "loop"))
1849 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1850 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1852 /* pre-create number of devices given by config or max_loop */
1853 mutex_lock(&loop_index_mutex
);
1854 for (i
= 0; i
< nr
; i
++)
1856 mutex_unlock(&loop_index_mutex
);
1858 printk(KERN_INFO
"loop: module loaded\n");
1862 static int loop_exit_cb(int id
, void *ptr
, void *data
)
1864 struct loop_device
*lo
= ptr
;
1870 static void __exit
loop_exit(void)
1872 unsigned long range
;
1874 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
1876 idr_for_each(&loop_index_idr
, &loop_exit_cb
, NULL
);
1877 idr_remove_all(&loop_index_idr
);
1878 idr_destroy(&loop_index_idr
);
1880 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1881 unregister_blkdev(LOOP_MAJOR
, "loop");
1883 misc_deregister(&loop_misc
);
1886 module_init(loop_init
);
1887 module_exit(loop_exit
);
1890 static int __init
max_loop_setup(char *str
)
1892 max_loop
= simple_strtol(str
, NULL
, 0);
1896 __setup("max_loop=", max_loop_setup
);