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/falloc.h>
80 #include <asm/uaccess.h>
82 static LIST_HEAD(loop_devices
);
83 static DEFINE_MUTEX(loop_devices_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
, KM_USER0
) + raw_off
;
97 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + loop_off
;
100 memcpy(loop_buf
, raw_buf
, size
);
102 memcpy(raw_buf
, loop_buf
, size
);
104 kunmap_atomic(loop_buf
, KM_USER1
);
105 kunmap_atomic(raw_buf
, KM_USER0
);
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
, KM_USER0
) + raw_off
;
116 char *loop_buf
= kmap_atomic(loop_page
, KM_USER1
) + 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
, KM_USER1
);
134 kunmap_atomic(raw_buf
, KM_USER0
);
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_loop_size(struct loop_device
*lo
, struct file
*file
)
165 loff_t size
, offset
, loopsize
;
167 /* Compute loopsize in bytes */
168 size
= i_size_read(file
->f_mapping
->host
);
169 offset
= lo
->lo_offset
;
170 loopsize
= size
- offset
;
171 if (lo
->lo_sizelimit
> 0 && lo
->lo_sizelimit
< loopsize
)
172 loopsize
= lo
->lo_sizelimit
;
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
178 return loopsize
>> 9;
182 figure_loop_size(struct loop_device
*lo
)
184 loff_t size
= get_loop_size(lo
, lo
->lo_backing_file
);
185 sector_t x
= (sector_t
)size
;
187 if (unlikely((loff_t
)x
!= size
))
190 set_capacity(lo
->lo_disk
, x
);
195 lo_do_transfer(struct loop_device
*lo
, int cmd
,
196 struct page
*rpage
, unsigned roffs
,
197 struct page
*lpage
, unsigned loffs
,
198 int size
, sector_t rblock
)
200 if (unlikely(!lo
->transfer
))
203 return lo
->transfer(lo
, cmd
, rpage
, roffs
, lpage
, loffs
, size
, rblock
);
207 * do_lo_send_aops - helper for writing data to a loop device
209 * This is the fast version for backing filesystems which implement the address
210 * space operations write_begin and write_end.
212 static int do_lo_send_aops(struct loop_device
*lo
, struct bio_vec
*bvec
,
213 loff_t pos
, struct page
*unused
)
215 struct file
*file
= lo
->lo_backing_file
; /* kudos to NFsckingS */
216 struct address_space
*mapping
= file
->f_mapping
;
218 unsigned offset
, bv_offs
;
221 mutex_lock(&mapping
->host
->i_mutex
);
222 index
= pos
>> PAGE_CACHE_SHIFT
;
223 offset
= pos
& ((pgoff_t
)PAGE_CACHE_SIZE
- 1);
224 bv_offs
= bvec
->bv_offset
;
228 unsigned size
, copied
;
233 IV
= ((sector_t
)index
<< (PAGE_CACHE_SHIFT
- 9))+(offset
>> 9);
234 size
= PAGE_CACHE_SIZE
- offset
;
238 ret
= pagecache_write_begin(file
, mapping
, pos
, size
, 0,
243 file_update_time(file
);
245 transfer_result
= lo_do_transfer(lo
, WRITE
, page
, offset
,
246 bvec
->bv_page
, bv_offs
, size
, IV
);
248 if (unlikely(transfer_result
))
251 ret
= pagecache_write_end(file
, mapping
, pos
, size
, copied
,
253 if (ret
< 0 || ret
!= copied
)
256 if (unlikely(transfer_result
))
267 mutex_unlock(&mapping
->host
->i_mutex
);
275 * __do_lo_send_write - helper for writing data to a loop device
277 * This helper just factors out common code between do_lo_send_direct_write()
278 * and do_lo_send_write().
280 static int __do_lo_send_write(struct file
*file
,
281 u8
*buf
, const int len
, loff_t pos
)
284 mm_segment_t old_fs
= get_fs();
287 bw
= file
->f_op
->write(file
, buf
, len
, &pos
);
289 if (likely(bw
== len
))
291 printk(KERN_ERR
"loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)pos
, len
);
299 * do_lo_send_direct_write - helper for writing data to a loop device
301 * This is the fast, non-transforming version for backing filesystems which do
302 * not implement the address space operations write_begin and write_end.
303 * It uses the write file operation which should be present on all writeable
306 static int do_lo_send_direct_write(struct loop_device
*lo
,
307 struct bio_vec
*bvec
, loff_t pos
, struct page
*page
)
309 ssize_t bw
= __do_lo_send_write(lo
->lo_backing_file
,
310 kmap(bvec
->bv_page
) + bvec
->bv_offset
,
312 kunmap(bvec
->bv_page
);
318 * do_lo_send_write - helper for writing data to a loop device
320 * This is the slow, transforming version for filesystems which do not
321 * implement the address space operations write_begin and write_end. It
322 * uses the write file operation which should be present on all writeable
325 * Using fops->write is slower than using aops->{prepare,commit}_write in the
326 * transforming case because we need to double buffer the data as we cannot do
327 * the transformations in place as we do not have direct access to the
328 * destination pages of the backing file.
330 static int do_lo_send_write(struct loop_device
*lo
, struct bio_vec
*bvec
,
331 loff_t pos
, struct page
*page
)
333 int ret
= lo_do_transfer(lo
, WRITE
, page
, 0, bvec
->bv_page
,
334 bvec
->bv_offset
, bvec
->bv_len
, pos
>> 9);
336 return __do_lo_send_write(lo
->lo_backing_file
,
337 page_address(page
), bvec
->bv_len
,
339 printk(KERN_ERR
"loop: Transfer error at byte offset %llu, "
340 "length %i.\n", (unsigned long long)pos
, bvec
->bv_len
);
346 static int lo_send(struct loop_device
*lo
, struct bio
*bio
, loff_t pos
)
348 int (*do_lo_send
)(struct loop_device
*, struct bio_vec
*, loff_t
,
350 struct bio_vec
*bvec
;
351 struct page
*page
= NULL
;
354 do_lo_send
= do_lo_send_aops
;
355 if (!(lo
->lo_flags
& LO_FLAGS_USE_AOPS
)) {
356 do_lo_send
= do_lo_send_direct_write
;
357 if (lo
->transfer
!= transfer_none
) {
358 page
= alloc_page(GFP_NOIO
| __GFP_HIGHMEM
);
362 do_lo_send
= do_lo_send_write
;
365 bio_for_each_segment(bvec
, bio
, i
) {
366 ret
= do_lo_send(lo
, bvec
, pos
, page
);
378 printk(KERN_ERR
"loop: Failed to allocate temporary page for write.\n");
383 struct lo_read_data
{
384 struct loop_device
*lo
;
391 lo_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
392 struct splice_desc
*sd
)
394 struct lo_read_data
*p
= sd
->u
.data
;
395 struct loop_device
*lo
= p
->lo
;
396 struct page
*page
= buf
->page
;
400 IV
= ((sector_t
) page
->index
<< (PAGE_CACHE_SHIFT
- 9)) +
406 if (lo_do_transfer(lo
, READ
, page
, buf
->offset
, p
->page
, p
->offset
, size
, IV
)) {
407 printk(KERN_ERR
"loop: transfer error block %ld\n",
412 flush_dcache_page(p
->page
);
421 lo_direct_splice_actor(struct pipe_inode_info
*pipe
, struct splice_desc
*sd
)
423 return __splice_from_pipe(pipe
, sd
, lo_splice_actor
);
427 do_lo_receive(struct loop_device
*lo
,
428 struct bio_vec
*bvec
, int bsize
, loff_t pos
)
430 struct lo_read_data cookie
;
431 struct splice_desc sd
;
436 cookie
.page
= bvec
->bv_page
;
437 cookie
.offset
= bvec
->bv_offset
;
438 cookie
.bsize
= bsize
;
441 sd
.total_len
= bvec
->bv_len
;
446 file
= lo
->lo_backing_file
;
447 retval
= splice_direct_to_actor(file
, &sd
, lo_direct_splice_actor
);
456 lo_receive(struct loop_device
*lo
, struct bio
*bio
, int bsize
, loff_t pos
)
458 struct bio_vec
*bvec
;
461 bio_for_each_segment(bvec
, bio
, i
) {
462 ret
= do_lo_receive(lo
, bvec
, bsize
, pos
);
470 static int do_bio_filebacked(struct loop_device
*lo
, struct bio
*bio
)
475 pos
= ((loff_t
) bio
->bi_sector
<< 9) + lo
->lo_offset
;
477 if (bio_rw(bio
) == WRITE
) {
478 struct file
*file
= lo
->lo_backing_file
;
480 if (bio
->bi_rw
& REQ_FLUSH
) {
481 ret
= vfs_fsync(file
, 0);
482 if (unlikely(ret
&& ret
!= -EINVAL
)) {
489 * We use punch hole to reclaim the free space used by the
490 * image a.k.a. discard. However we do support discard if
491 * encryption is enabled, because it may give an attacker
492 * useful information.
494 if (bio
->bi_rw
& REQ_DISCARD
) {
495 struct file
*file
= lo
->lo_backing_file
;
496 int mode
= FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
;
498 if ((!file
->f_op
->fallocate
) ||
499 lo
->lo_encrypt_key_size
) {
503 ret
= file
->f_op
->fallocate(file
, mode
, pos
,
505 if (unlikely(ret
&& ret
!= -EINVAL
&&
511 ret
= lo_send(lo
, bio
, pos
);
513 if ((bio
->bi_rw
& REQ_FUA
) && !ret
) {
514 ret
= vfs_fsync(file
, 0);
515 if (unlikely(ret
&& ret
!= -EINVAL
))
519 ret
= lo_receive(lo
, bio
, lo
->lo_blocksize
, pos
);
526 * Add bio to back of pending list
528 static void loop_add_bio(struct loop_device
*lo
, struct bio
*bio
)
530 bio_list_add(&lo
->lo_bio_list
, bio
);
534 * Grab first pending buffer
536 static struct bio
*loop_get_bio(struct loop_device
*lo
)
538 return bio_list_pop(&lo
->lo_bio_list
);
541 static int loop_make_request(struct request_queue
*q
, struct bio
*old_bio
)
543 struct loop_device
*lo
= q
->queuedata
;
544 int rw
= bio_rw(old_bio
);
549 BUG_ON(!lo
|| (rw
!= READ
&& rw
!= WRITE
));
551 spin_lock_irq(&lo
->lo_lock
);
552 if (lo
->lo_state
!= Lo_bound
)
554 if (unlikely(rw
== WRITE
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)))
556 loop_add_bio(lo
, old_bio
);
557 wake_up(&lo
->lo_event
);
558 spin_unlock_irq(&lo
->lo_lock
);
562 spin_unlock_irq(&lo
->lo_lock
);
563 bio_io_error(old_bio
);
567 struct switch_request
{
569 struct completion wait
;
572 static void do_loop_switch(struct loop_device
*, struct switch_request
*);
574 static inline void loop_handle_bio(struct loop_device
*lo
, struct bio
*bio
)
576 if (unlikely(!bio
->bi_bdev
)) {
577 do_loop_switch(lo
, bio
->bi_private
);
580 int ret
= do_bio_filebacked(lo
, bio
);
586 * worker thread that handles reads/writes to file backed loop devices,
587 * to avoid blocking in our make_request_fn. it also does loop decrypting
588 * on reads for block backed loop, as that is too heavy to do from
589 * b_end_io context where irqs may be disabled.
591 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
592 * calling kthread_stop(). Therefore once kthread_should_stop() is
593 * true, make_request will not place any more requests. Therefore
594 * once kthread_should_stop() is true and lo_bio is NULL, we are
595 * done with the loop.
597 static int loop_thread(void *data
)
599 struct loop_device
*lo
= data
;
602 set_user_nice(current
, -20);
604 while (!kthread_should_stop() || !bio_list_empty(&lo
->lo_bio_list
)) {
606 wait_event_interruptible(lo
->lo_event
,
607 !bio_list_empty(&lo
->lo_bio_list
) ||
608 kthread_should_stop());
610 if (bio_list_empty(&lo
->lo_bio_list
))
612 spin_lock_irq(&lo
->lo_lock
);
613 bio
= loop_get_bio(lo
);
614 spin_unlock_irq(&lo
->lo_lock
);
617 loop_handle_bio(lo
, bio
);
624 * loop_switch performs the hard work of switching a backing store.
625 * First it needs to flush existing IO, it does this by sending a magic
626 * BIO down the pipe. The completion of this BIO does the actual switch.
628 static int loop_switch(struct loop_device
*lo
, struct file
*file
)
630 struct switch_request w
;
631 struct bio
*bio
= bio_alloc(GFP_KERNEL
, 0);
634 init_completion(&w
.wait
);
636 bio
->bi_private
= &w
;
638 loop_make_request(lo
->lo_queue
, bio
);
639 wait_for_completion(&w
.wait
);
644 * Helper to flush the IOs in loop, but keeping loop thread running
646 static int loop_flush(struct loop_device
*lo
)
648 /* loop not yet configured, no running thread, nothing to flush */
652 return loop_switch(lo
, NULL
);
656 * Do the actual switch; called from the BIO completion routine
658 static void do_loop_switch(struct loop_device
*lo
, struct switch_request
*p
)
660 struct file
*file
= p
->file
;
661 struct file
*old_file
= lo
->lo_backing_file
;
662 struct address_space
*mapping
;
664 /* if no new file, only flush of queued bios requested */
668 mapping
= file
->f_mapping
;
669 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
670 lo
->lo_backing_file
= file
;
671 lo
->lo_blocksize
= S_ISBLK(mapping
->host
->i_mode
) ?
672 mapping
->host
->i_bdev
->bd_block_size
: PAGE_SIZE
;
673 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
674 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
681 * loop_change_fd switched the backing store of a loopback device to
682 * a new file. This is useful for operating system installers to free up
683 * the original file and in High Availability environments to switch to
684 * an alternative location for the content in case of server meltdown.
685 * This can only work if the loop device is used read-only, and if the
686 * new backing store is the same size and type as the old backing store.
688 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
691 struct file
*file
, *old_file
;
696 if (lo
->lo_state
!= Lo_bound
)
699 /* the loop device has to be read-only */
701 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
709 inode
= file
->f_mapping
->host
;
710 old_file
= lo
->lo_backing_file
;
714 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
717 /* size of the new backing store needs to be the same */
718 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
722 error
= loop_switch(lo
, file
);
727 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
728 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
737 static inline int is_loop_device(struct file
*file
)
739 struct inode
*i
= file
->f_mapping
->host
;
741 return i
&& S_ISBLK(i
->i_mode
) && MAJOR(i
->i_rdev
) == LOOP_MAJOR
;
744 /* loop sysfs attributes */
746 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
747 ssize_t (*callback
)(struct loop_device
*, char *))
749 struct loop_device
*l
, *lo
= NULL
;
751 mutex_lock(&loop_devices_mutex
);
752 list_for_each_entry(l
, &loop_devices
, lo_list
)
753 if (disk_to_dev(l
->lo_disk
) == dev
) {
757 mutex_unlock(&loop_devices_mutex
);
759 return lo
? callback(lo
, page
) : -EIO
;
762 #define LOOP_ATTR_RO(_name) \
763 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
764 static ssize_t loop_attr_do_show_##_name(struct device *d, \
765 struct device_attribute *attr, char *b) \
767 return loop_attr_show(d, b, loop_attr_##_name##_show); \
769 static struct device_attribute loop_attr_##_name = \
770 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
772 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
777 mutex_lock(&lo
->lo_ctl_mutex
);
778 if (lo
->lo_backing_file
)
779 p
= d_path(&lo
->lo_backing_file
->f_path
, buf
, PAGE_SIZE
- 1);
780 mutex_unlock(&lo
->lo_ctl_mutex
);
782 if (IS_ERR_OR_NULL(p
))
786 memmove(buf
, p
, ret
);
794 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
796 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
799 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
801 return sprintf(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
804 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
806 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
808 return sprintf(buf
, "%s\n", autoclear
? "1" : "0");
811 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
813 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
815 return sprintf(buf
, "%s\n", partscan
? "1" : "0");
818 LOOP_ATTR_RO(backing_file
);
819 LOOP_ATTR_RO(offset
);
820 LOOP_ATTR_RO(sizelimit
);
821 LOOP_ATTR_RO(autoclear
);
822 LOOP_ATTR_RO(partscan
);
824 static struct attribute
*loop_attrs
[] = {
825 &loop_attr_backing_file
.attr
,
826 &loop_attr_offset
.attr
,
827 &loop_attr_sizelimit
.attr
,
828 &loop_attr_autoclear
.attr
,
829 &loop_attr_partscan
.attr
,
833 static struct attribute_group loop_attribute_group
= {
838 static int loop_sysfs_init(struct loop_device
*lo
)
840 return sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
841 &loop_attribute_group
);
844 static void loop_sysfs_exit(struct loop_device
*lo
)
846 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
847 &loop_attribute_group
);
850 static void loop_config_discard(struct loop_device
*lo
)
852 struct file
*file
= lo
->lo_backing_file
;
853 struct inode
*inode
= file
->f_mapping
->host
;
854 struct request_queue
*q
= lo
->lo_queue
;
857 * We use punch hole to reclaim the free space used by the
858 * image a.k.a. discard. However we do support discard if
859 * encryption is enabled, because it may give an attacker
860 * useful information.
862 if ((!file
->f_op
->fallocate
) ||
863 lo
->lo_encrypt_key_size
) {
864 q
->limits
.discard_granularity
= 0;
865 q
->limits
.discard_alignment
= 0;
866 q
->limits
.max_discard_sectors
= 0;
867 q
->limits
.discard_zeroes_data
= 0;
868 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, q
);
872 q
->limits
.discard_granularity
= inode
->i_sb
->s_blocksize
;
873 q
->limits
.discard_alignment
= inode
->i_sb
->s_blocksize
;
874 q
->limits
.max_discard_sectors
= UINT_MAX
>> 9;
875 q
->limits
.discard_zeroes_data
= 1;
876 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, q
);
879 static int loop_set_fd(struct loop_device
*lo
, fmode_t mode
,
880 struct block_device
*bdev
, unsigned int arg
)
882 struct file
*file
, *f
;
884 struct address_space
*mapping
;
885 unsigned lo_blocksize
;
890 /* This is safe, since we have a reference from open(). */
891 __module_get(THIS_MODULE
);
899 if (lo
->lo_state
!= Lo_unbound
)
902 /* Avoid recursion */
904 while (is_loop_device(f
)) {
905 struct loop_device
*l
;
907 if (f
->f_mapping
->host
->i_bdev
== bdev
)
910 l
= f
->f_mapping
->host
->i_bdev
->bd_disk
->private_data
;
911 if (l
->lo_state
== Lo_unbound
) {
915 f
= l
->lo_backing_file
;
918 mapping
= file
->f_mapping
;
919 inode
= mapping
->host
;
921 if (!(file
->f_mode
& FMODE_WRITE
))
922 lo_flags
|= LO_FLAGS_READ_ONLY
;
925 if (S_ISREG(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
926 const struct address_space_operations
*aops
= mapping
->a_ops
;
928 if (aops
->write_begin
)
929 lo_flags
|= LO_FLAGS_USE_AOPS
;
930 if (!(lo_flags
& LO_FLAGS_USE_AOPS
) && !file
->f_op
->write
)
931 lo_flags
|= LO_FLAGS_READ_ONLY
;
933 lo_blocksize
= S_ISBLK(inode
->i_mode
) ?
934 inode
->i_bdev
->bd_block_size
: PAGE_SIZE
;
941 size
= get_loop_size(lo
, file
);
943 if ((loff_t
)(sector_t
)size
!= size
) {
948 if (!(mode
& FMODE_WRITE
))
949 lo_flags
|= LO_FLAGS_READ_ONLY
;
951 set_device_ro(bdev
, (lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
953 lo
->lo_blocksize
= lo_blocksize
;
954 lo
->lo_device
= bdev
;
955 lo
->lo_flags
= lo_flags
;
956 lo
->lo_backing_file
= file
;
957 lo
->transfer
= transfer_none
;
959 lo
->lo_sizelimit
= 0;
960 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
961 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
963 bio_list_init(&lo
->lo_bio_list
);
966 * set queue make_request_fn, and add limits based on lower level
969 blk_queue_make_request(lo
->lo_queue
, loop_make_request
);
970 lo
->lo_queue
->queuedata
= lo
;
972 if (!(lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
973 blk_queue_flush(lo
->lo_queue
, REQ_FLUSH
);
975 set_capacity(lo
->lo_disk
, size
);
976 bd_set_size(bdev
, size
<< 9);
978 /* let user-space know about the new size */
979 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
981 set_blocksize(bdev
, lo_blocksize
);
983 lo
->lo_thread
= kthread_create(loop_thread
, lo
, "loop%d",
985 if (IS_ERR(lo
->lo_thread
)) {
986 error
= PTR_ERR(lo
->lo_thread
);
989 lo
->lo_state
= Lo_bound
;
990 wake_up_process(lo
->lo_thread
);
992 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
993 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
)
994 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
999 lo
->lo_thread
= NULL
;
1000 lo
->lo_device
= NULL
;
1001 lo
->lo_backing_file
= NULL
;
1003 set_capacity(lo
->lo_disk
, 0);
1004 invalidate_bdev(bdev
);
1005 bd_set_size(bdev
, 0);
1006 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1007 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
);
1008 lo
->lo_state
= Lo_unbound
;
1012 /* This is safe: open() is still holding a reference. */
1013 module_put(THIS_MODULE
);
1018 loop_release_xfer(struct loop_device
*lo
)
1021 struct loop_func_table
*xfer
= lo
->lo_encryption
;
1025 err
= xfer
->release(lo
);
1026 lo
->transfer
= NULL
;
1027 lo
->lo_encryption
= NULL
;
1028 module_put(xfer
->owner
);
1034 loop_init_xfer(struct loop_device
*lo
, struct loop_func_table
*xfer
,
1035 const struct loop_info64
*i
)
1040 struct module
*owner
= xfer
->owner
;
1042 if (!try_module_get(owner
))
1045 err
= xfer
->init(lo
, i
);
1049 lo
->lo_encryption
= xfer
;
1054 static int loop_clr_fd(struct loop_device
*lo
, struct block_device
*bdev
)
1056 struct file
*filp
= lo
->lo_backing_file
;
1057 gfp_t gfp
= lo
->old_gfp_mask
;
1059 if (lo
->lo_state
!= Lo_bound
)
1062 if (lo
->lo_refcnt
> 1) /* we needed one fd for the ioctl */
1068 spin_lock_irq(&lo
->lo_lock
);
1069 lo
->lo_state
= Lo_rundown
;
1070 spin_unlock_irq(&lo
->lo_lock
);
1072 kthread_stop(lo
->lo_thread
);
1074 lo
->lo_backing_file
= NULL
;
1076 loop_release_xfer(lo
);
1077 lo
->transfer
= NULL
;
1079 lo
->lo_device
= NULL
;
1080 lo
->lo_encryption
= NULL
;
1082 lo
->lo_sizelimit
= 0;
1083 lo
->lo_encrypt_key_size
= 0;
1084 lo
->lo_thread
= NULL
;
1085 memset(lo
->lo_encrypt_key
, 0, LO_KEY_SIZE
);
1086 memset(lo
->lo_crypt_name
, 0, LO_NAME_SIZE
);
1087 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1089 invalidate_bdev(bdev
);
1090 set_capacity(lo
->lo_disk
, 0);
1091 loop_sysfs_exit(lo
);
1093 bd_set_size(bdev
, 0);
1094 /* let user-space know about this change */
1095 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1097 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1098 lo
->lo_state
= Lo_unbound
;
1099 /* This is safe: open() is still holding a reference. */
1100 module_put(THIS_MODULE
);
1101 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
&& bdev
)
1102 ioctl_by_bdev(bdev
, BLKRRPART
, 0);
1105 lo
->lo_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1106 mutex_unlock(&lo
->lo_ctl_mutex
);
1108 * Need not hold lo_ctl_mutex to fput backing file.
1109 * Calling fput holding lo_ctl_mutex triggers a circular
1110 * lock dependency possibility warning as fput can take
1111 * bd_mutex which is usually taken before lo_ctl_mutex.
1118 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1121 struct loop_func_table
*xfer
;
1122 uid_t uid
= current_uid();
1124 if (lo
->lo_encrypt_key_size
&&
1125 lo
->lo_key_owner
!= uid
&&
1126 !capable(CAP_SYS_ADMIN
))
1128 if (lo
->lo_state
!= Lo_bound
)
1130 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
1133 err
= loop_release_xfer(lo
);
1137 if (info
->lo_encrypt_type
) {
1138 unsigned int type
= info
->lo_encrypt_type
;
1140 if (type
>= MAX_LO_CRYPT
)
1142 xfer
= xfer_funcs
[type
];
1148 err
= loop_init_xfer(lo
, xfer
, info
);
1152 if (lo
->lo_offset
!= info
->lo_offset
||
1153 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1154 lo
->lo_offset
= info
->lo_offset
;
1155 lo
->lo_sizelimit
= info
->lo_sizelimit
;
1156 if (figure_loop_size(lo
))
1159 loop_config_discard(lo
);
1161 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
1162 memcpy(lo
->lo_crypt_name
, info
->lo_crypt_name
, LO_NAME_SIZE
);
1163 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
1164 lo
->lo_crypt_name
[LO_NAME_SIZE
-1] = 0;
1168 lo
->transfer
= xfer
->transfer
;
1169 lo
->ioctl
= xfer
->ioctl
;
1171 if ((lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) !=
1172 (info
->lo_flags
& LO_FLAGS_AUTOCLEAR
))
1173 lo
->lo_flags
^= LO_FLAGS_AUTOCLEAR
;
1175 if ((info
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1176 !(lo
->lo_flags
& LO_FLAGS_PARTSCAN
)) {
1177 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1178 lo
->lo_disk
->flags
&= ~GENHD_FL_NO_PART_SCAN
;
1179 ioctl_by_bdev(lo
->lo_device
, BLKRRPART
, 0);
1182 lo
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1183 lo
->lo_init
[0] = info
->lo_init
[0];
1184 lo
->lo_init
[1] = info
->lo_init
[1];
1185 if (info
->lo_encrypt_key_size
) {
1186 memcpy(lo
->lo_encrypt_key
, info
->lo_encrypt_key
,
1187 info
->lo_encrypt_key_size
);
1188 lo
->lo_key_owner
= uid
;
1195 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1197 struct file
*file
= lo
->lo_backing_file
;
1201 if (lo
->lo_state
!= Lo_bound
)
1203 error
= vfs_getattr(file
->f_path
.mnt
, file
->f_path
.dentry
, &stat
);
1206 memset(info
, 0, sizeof(*info
));
1207 info
->lo_number
= lo
->lo_number
;
1208 info
->lo_device
= huge_encode_dev(stat
.dev
);
1209 info
->lo_inode
= stat
.ino
;
1210 info
->lo_rdevice
= huge_encode_dev(lo
->lo_device
? stat
.rdev
: stat
.dev
);
1211 info
->lo_offset
= lo
->lo_offset
;
1212 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1213 info
->lo_flags
= lo
->lo_flags
;
1214 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1215 memcpy(info
->lo_crypt_name
, lo
->lo_crypt_name
, LO_NAME_SIZE
);
1216 info
->lo_encrypt_type
=
1217 lo
->lo_encryption
? lo
->lo_encryption
->number
: 0;
1218 if (lo
->lo_encrypt_key_size
&& capable(CAP_SYS_ADMIN
)) {
1219 info
->lo_encrypt_key_size
= lo
->lo_encrypt_key_size
;
1220 memcpy(info
->lo_encrypt_key
, lo
->lo_encrypt_key
,
1221 lo
->lo_encrypt_key_size
);
1227 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1229 memset(info64
, 0, sizeof(*info64
));
1230 info64
->lo_number
= info
->lo_number
;
1231 info64
->lo_device
= info
->lo_device
;
1232 info64
->lo_inode
= info
->lo_inode
;
1233 info64
->lo_rdevice
= info
->lo_rdevice
;
1234 info64
->lo_offset
= info
->lo_offset
;
1235 info64
->lo_sizelimit
= 0;
1236 info64
->lo_encrypt_type
= info
->lo_encrypt_type
;
1237 info64
->lo_encrypt_key_size
= info
->lo_encrypt_key_size
;
1238 info64
->lo_flags
= info
->lo_flags
;
1239 info64
->lo_init
[0] = info
->lo_init
[0];
1240 info64
->lo_init
[1] = info
->lo_init
[1];
1241 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1242 memcpy(info64
->lo_crypt_name
, info
->lo_name
, LO_NAME_SIZE
);
1244 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1245 memcpy(info64
->lo_encrypt_key
, info
->lo_encrypt_key
, LO_KEY_SIZE
);
1249 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1251 memset(info
, 0, sizeof(*info
));
1252 info
->lo_number
= info64
->lo_number
;
1253 info
->lo_device
= info64
->lo_device
;
1254 info
->lo_inode
= info64
->lo_inode
;
1255 info
->lo_rdevice
= info64
->lo_rdevice
;
1256 info
->lo_offset
= info64
->lo_offset
;
1257 info
->lo_encrypt_type
= info64
->lo_encrypt_type
;
1258 info
->lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1259 info
->lo_flags
= info64
->lo_flags
;
1260 info
->lo_init
[0] = info64
->lo_init
[0];
1261 info
->lo_init
[1] = info64
->lo_init
[1];
1262 if (info
->lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1263 memcpy(info
->lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1265 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1266 memcpy(info
->lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1268 /* error in case values were truncated */
1269 if (info
->lo_device
!= info64
->lo_device
||
1270 info
->lo_rdevice
!= info64
->lo_rdevice
||
1271 info
->lo_inode
!= info64
->lo_inode
||
1272 info
->lo_offset
!= info64
->lo_offset
)
1279 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1281 struct loop_info info
;
1282 struct loop_info64 info64
;
1284 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1286 loop_info64_from_old(&info
, &info64
);
1287 return loop_set_status(lo
, &info64
);
1291 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1293 struct loop_info64 info64
;
1295 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1297 return loop_set_status(lo
, &info64
);
1301 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1302 struct loop_info info
;
1303 struct loop_info64 info64
;
1309 err
= loop_get_status(lo
, &info64
);
1311 err
= loop_info64_to_old(&info64
, &info
);
1312 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1319 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1320 struct loop_info64 info64
;
1326 err
= loop_get_status(lo
, &info64
);
1327 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1333 static int loop_set_capacity(struct loop_device
*lo
, struct block_device
*bdev
)
1340 if (unlikely(lo
->lo_state
!= Lo_bound
))
1342 err
= figure_loop_size(lo
);
1345 sec
= get_capacity(lo
->lo_disk
);
1346 /* the width of sector_t may be narrow for bit-shift */
1349 mutex_lock(&bdev
->bd_mutex
);
1350 bd_set_size(bdev
, sz
);
1351 /* let user-space know about the new size */
1352 kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, KOBJ_CHANGE
);
1353 mutex_unlock(&bdev
->bd_mutex
);
1359 static int lo_ioctl(struct block_device
*bdev
, fmode_t mode
,
1360 unsigned int cmd
, unsigned long arg
)
1362 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1365 mutex_lock_nested(&lo
->lo_ctl_mutex
, 1);
1368 err
= loop_set_fd(lo
, mode
, bdev
, arg
);
1370 case LOOP_CHANGE_FD
:
1371 err
= loop_change_fd(lo
, bdev
, arg
);
1374 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1375 err
= loop_clr_fd(lo
, bdev
);
1379 case LOOP_SET_STATUS
:
1380 err
= loop_set_status_old(lo
, (struct loop_info __user
*) arg
);
1382 case LOOP_GET_STATUS
:
1383 err
= loop_get_status_old(lo
, (struct loop_info __user
*) arg
);
1385 case LOOP_SET_STATUS64
:
1386 err
= loop_set_status64(lo
, (struct loop_info64 __user
*) arg
);
1388 case LOOP_GET_STATUS64
:
1389 err
= loop_get_status64(lo
, (struct loop_info64 __user
*) arg
);
1391 case LOOP_SET_CAPACITY
:
1393 if ((mode
& FMODE_WRITE
) || capable(CAP_SYS_ADMIN
))
1394 err
= loop_set_capacity(lo
, bdev
);
1397 err
= lo
->ioctl
? lo
->ioctl(lo
, cmd
, arg
) : -EINVAL
;
1399 mutex_unlock(&lo
->lo_ctl_mutex
);
1405 #ifdef CONFIG_COMPAT
1406 struct compat_loop_info
{
1407 compat_int_t lo_number
; /* ioctl r/o */
1408 compat_dev_t lo_device
; /* ioctl r/o */
1409 compat_ulong_t lo_inode
; /* ioctl r/o */
1410 compat_dev_t lo_rdevice
; /* ioctl r/o */
1411 compat_int_t lo_offset
;
1412 compat_int_t lo_encrypt_type
;
1413 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1414 compat_int_t lo_flags
; /* ioctl r/o */
1415 char lo_name
[LO_NAME_SIZE
];
1416 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1417 compat_ulong_t lo_init
[2];
1422 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1423 * - noinlined to reduce stack space usage in main part of driver
1426 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1427 struct loop_info64
*info64
)
1429 struct compat_loop_info info
;
1431 if (copy_from_user(&info
, arg
, sizeof(info
)))
1434 memset(info64
, 0, sizeof(*info64
));
1435 info64
->lo_number
= info
.lo_number
;
1436 info64
->lo_device
= info
.lo_device
;
1437 info64
->lo_inode
= info
.lo_inode
;
1438 info64
->lo_rdevice
= info
.lo_rdevice
;
1439 info64
->lo_offset
= info
.lo_offset
;
1440 info64
->lo_sizelimit
= 0;
1441 info64
->lo_encrypt_type
= info
.lo_encrypt_type
;
1442 info64
->lo_encrypt_key_size
= info
.lo_encrypt_key_size
;
1443 info64
->lo_flags
= info
.lo_flags
;
1444 info64
->lo_init
[0] = info
.lo_init
[0];
1445 info64
->lo_init
[1] = info
.lo_init
[1];
1446 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1447 memcpy(info64
->lo_crypt_name
, info
.lo_name
, LO_NAME_SIZE
);
1449 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1450 memcpy(info64
->lo_encrypt_key
, info
.lo_encrypt_key
, LO_KEY_SIZE
);
1455 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1456 * - noinlined to reduce stack space usage in main part of driver
1459 loop_info64_to_compat(const struct loop_info64
*info64
,
1460 struct compat_loop_info __user
*arg
)
1462 struct compat_loop_info info
;
1464 memset(&info
, 0, sizeof(info
));
1465 info
.lo_number
= info64
->lo_number
;
1466 info
.lo_device
= info64
->lo_device
;
1467 info
.lo_inode
= info64
->lo_inode
;
1468 info
.lo_rdevice
= info64
->lo_rdevice
;
1469 info
.lo_offset
= info64
->lo_offset
;
1470 info
.lo_encrypt_type
= info64
->lo_encrypt_type
;
1471 info
.lo_encrypt_key_size
= info64
->lo_encrypt_key_size
;
1472 info
.lo_flags
= info64
->lo_flags
;
1473 info
.lo_init
[0] = info64
->lo_init
[0];
1474 info
.lo_init
[1] = info64
->lo_init
[1];
1475 if (info
.lo_encrypt_type
== LO_CRYPT_CRYPTOAPI
)
1476 memcpy(info
.lo_name
, info64
->lo_crypt_name
, LO_NAME_SIZE
);
1478 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1479 memcpy(info
.lo_encrypt_key
, info64
->lo_encrypt_key
, LO_KEY_SIZE
);
1481 /* error in case values were truncated */
1482 if (info
.lo_device
!= info64
->lo_device
||
1483 info
.lo_rdevice
!= info64
->lo_rdevice
||
1484 info
.lo_inode
!= info64
->lo_inode
||
1485 info
.lo_offset
!= info64
->lo_offset
||
1486 info
.lo_init
[0] != info64
->lo_init
[0] ||
1487 info
.lo_init
[1] != info64
->lo_init
[1])
1490 if (copy_to_user(arg
, &info
, sizeof(info
)))
1496 loop_set_status_compat(struct loop_device
*lo
,
1497 const struct compat_loop_info __user
*arg
)
1499 struct loop_info64 info64
;
1502 ret
= loop_info64_from_compat(arg
, &info64
);
1505 return loop_set_status(lo
, &info64
);
1509 loop_get_status_compat(struct loop_device
*lo
,
1510 struct compat_loop_info __user
*arg
)
1512 struct loop_info64 info64
;
1518 err
= loop_get_status(lo
, &info64
);
1520 err
= loop_info64_to_compat(&info64
, arg
);
1524 static int lo_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1525 unsigned int cmd
, unsigned long arg
)
1527 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1531 case LOOP_SET_STATUS
:
1532 mutex_lock(&lo
->lo_ctl_mutex
);
1533 err
= loop_set_status_compat(
1534 lo
, (const struct compat_loop_info __user
*) arg
);
1535 mutex_unlock(&lo
->lo_ctl_mutex
);
1537 case LOOP_GET_STATUS
:
1538 mutex_lock(&lo
->lo_ctl_mutex
);
1539 err
= loop_get_status_compat(
1540 lo
, (struct compat_loop_info __user
*) arg
);
1541 mutex_unlock(&lo
->lo_ctl_mutex
);
1543 case LOOP_SET_CAPACITY
:
1545 case LOOP_GET_STATUS64
:
1546 case LOOP_SET_STATUS64
:
1547 arg
= (unsigned long) compat_ptr(arg
);
1549 case LOOP_CHANGE_FD
:
1550 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1560 static int lo_open(struct block_device
*bdev
, fmode_t mode
)
1562 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1564 mutex_lock(&lo
->lo_ctl_mutex
);
1566 mutex_unlock(&lo
->lo_ctl_mutex
);
1571 static int lo_release(struct gendisk
*disk
, fmode_t mode
)
1573 struct loop_device
*lo
= disk
->private_data
;
1576 mutex_lock(&lo
->lo_ctl_mutex
);
1578 if (--lo
->lo_refcnt
)
1581 if (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
) {
1583 * In autoclear mode, stop the loop thread
1584 * and remove configuration after last close.
1586 err
= loop_clr_fd(lo
, NULL
);
1591 * Otherwise keep thread (if running) and config,
1592 * but flush possible ongoing bios in thread.
1598 mutex_unlock(&lo
->lo_ctl_mutex
);
1603 static const struct block_device_operations lo_fops
= {
1604 .owner
= THIS_MODULE
,
1606 .release
= lo_release
,
1608 #ifdef CONFIG_COMPAT
1609 .compat_ioctl
= lo_compat_ioctl
,
1614 * And now the modules code and kernel interface.
1616 static int max_loop
;
1617 module_param(max_loop
, int, S_IRUGO
);
1618 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1619 module_param(max_part
, int, S_IRUGO
);
1620 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1621 MODULE_LICENSE("GPL");
1622 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1624 int loop_register_transfer(struct loop_func_table
*funcs
)
1626 unsigned int n
= funcs
->number
;
1628 if (n
>= MAX_LO_CRYPT
|| xfer_funcs
[n
])
1630 xfer_funcs
[n
] = funcs
;
1634 int loop_unregister_transfer(int number
)
1636 unsigned int n
= number
;
1637 struct loop_device
*lo
;
1638 struct loop_func_table
*xfer
;
1640 if (n
== 0 || n
>= MAX_LO_CRYPT
|| (xfer
= xfer_funcs
[n
]) == NULL
)
1643 xfer_funcs
[n
] = NULL
;
1645 list_for_each_entry(lo
, &loop_devices
, lo_list
) {
1646 mutex_lock(&lo
->lo_ctl_mutex
);
1648 if (lo
->lo_encryption
== xfer
)
1649 loop_release_xfer(lo
);
1651 mutex_unlock(&lo
->lo_ctl_mutex
);
1657 EXPORT_SYMBOL(loop_register_transfer
);
1658 EXPORT_SYMBOL(loop_unregister_transfer
);
1660 static struct loop_device
*loop_alloc(int i
)
1662 struct loop_device
*lo
;
1663 struct gendisk
*disk
;
1665 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1669 lo
->lo_queue
= blk_alloc_queue(GFP_KERNEL
);
1673 disk
= lo
->lo_disk
= alloc_disk(1 << part_shift
);
1675 goto out_free_queue
;
1678 * Disable partition scanning by default. The in-kernel partition
1679 * scanning can be requested individually per-device during its
1680 * setup. Userspace can always add and remove partitions from all
1681 * devices. The needed partition minors are allocated from the
1682 * extended minor space, the main loop device numbers will continue
1683 * to match the loop minors, regardless of the number of partitions
1686 * If max_part is given, partition scanning is globally enabled for
1687 * all loop devices. The minors for the main loop devices will be
1688 * multiples of max_part.
1690 * Note: Global-for-all-devices, set-only-at-init, read-only module
1691 * parameteters like 'max_loop' and 'max_part' make things needlessly
1692 * complicated, are too static, inflexible and may surprise
1693 * userspace tools. Parameters like this in general should be avoided.
1696 disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1697 disk
->flags
|= GENHD_FL_EXT_DEVT
;
1698 mutex_init(&lo
->lo_ctl_mutex
);
1700 lo
->lo_thread
= NULL
;
1701 init_waitqueue_head(&lo
->lo_event
);
1702 spin_lock_init(&lo
->lo_lock
);
1703 disk
->major
= LOOP_MAJOR
;
1704 disk
->first_minor
= i
<< part_shift
;
1705 disk
->fops
= &lo_fops
;
1706 disk
->private_data
= lo
;
1707 disk
->queue
= lo
->lo_queue
;
1708 sprintf(disk
->disk_name
, "loop%d", i
);
1712 blk_cleanup_queue(lo
->lo_queue
);
1719 static void loop_free(struct loop_device
*lo
)
1721 blk_cleanup_queue(lo
->lo_queue
);
1722 put_disk(lo
->lo_disk
);
1723 list_del(&lo
->lo_list
);
1727 static struct loop_device
*loop_init_one(int i
)
1729 struct loop_device
*lo
;
1731 list_for_each_entry(lo
, &loop_devices
, lo_list
) {
1732 if (lo
->lo_number
== i
)
1738 add_disk(lo
->lo_disk
);
1739 list_add_tail(&lo
->lo_list
, &loop_devices
);
1744 static void loop_del_one(struct loop_device
*lo
)
1746 del_gendisk(lo
->lo_disk
);
1750 static struct kobject
*loop_probe(dev_t dev
, int *part
, void *data
)
1752 struct loop_device
*lo
;
1753 struct kobject
*kobj
;
1755 mutex_lock(&loop_devices_mutex
);
1756 lo
= loop_init_one(MINOR(dev
) >> part_shift
);
1757 kobj
= lo
? get_disk(lo
->lo_disk
) : ERR_PTR(-ENOMEM
);
1758 mutex_unlock(&loop_devices_mutex
);
1764 static int __init
loop_init(void)
1767 unsigned long range
;
1768 struct loop_device
*lo
, *next
;
1771 * loop module now has a feature to instantiate underlying device
1772 * structure on-demand, provided that there is an access dev node.
1773 * However, this will not work well with user space tool that doesn't
1774 * know about such "feature". In order to not break any existing
1775 * tool, we do the following:
1777 * (1) if max_loop is specified, create that many upfront, and this
1778 * also becomes a hard limit.
1779 * (2) if max_loop is not specified, create 8 loop device on module
1780 * load, user can further extend loop device by create dev node
1781 * themselves and have kernel automatically instantiate actual
1787 part_shift
= fls(max_part
);
1790 * Adjust max_part according to part_shift as it is exported
1791 * to user space so that user can decide correct minor number
1792 * if [s]he want to create more devices.
1794 * Note that -1 is required because partition 0 is reserved
1795 * for the whole disk.
1797 max_part
= (1UL << part_shift
) - 1;
1800 if ((1UL << part_shift
) > DISK_MAX_PARTS
)
1803 if (max_loop
> 1UL << (MINORBITS
- part_shift
))
1808 range
= max_loop
<< part_shift
;
1811 range
= 1UL << MINORBITS
;
1814 if (register_blkdev(LOOP_MAJOR
, "loop"))
1817 for (i
= 0; i
< nr
; i
++) {
1821 list_add_tail(&lo
->lo_list
, &loop_devices
);
1824 /* point of no return */
1826 list_for_each_entry(lo
, &loop_devices
, lo_list
)
1827 add_disk(lo
->lo_disk
);
1829 blk_register_region(MKDEV(LOOP_MAJOR
, 0), range
,
1830 THIS_MODULE
, loop_probe
, NULL
, NULL
);
1832 printk(KERN_INFO
"loop: module loaded\n");
1836 printk(KERN_INFO
"loop: out of memory\n");
1838 list_for_each_entry_safe(lo
, next
, &loop_devices
, lo_list
)
1841 unregister_blkdev(LOOP_MAJOR
, "loop");
1845 static void __exit
loop_exit(void)
1847 unsigned long range
;
1848 struct loop_device
*lo
, *next
;
1850 range
= max_loop
? max_loop
<< part_shift
: 1UL << MINORBITS
;
1852 list_for_each_entry_safe(lo
, next
, &loop_devices
, lo_list
)
1855 blk_unregister_region(MKDEV(LOOP_MAJOR
, 0), range
);
1856 unregister_blkdev(LOOP_MAJOR
, "loop");
1859 module_init(loop_init
);
1860 module_exit(loop_exit
);
1863 static int __init
max_loop_setup(char *str
)
1865 max_loop
= simple_strtol(str
, NULL
, 0);
1869 __setup("max_loop=", max_loop_setup
);